On the Feasibility of Securing Vehicle-Pavement Interaction

WEI SUN∗ , The Ohio State University, Columbus, USA

KANNAN SRINIVASAN, The Ohio State University, Columbus, USA
Road surface information (e.g., smooth road or bumpy road with potholes or bumps) is important for safe driving (i.e., it’s
necessary to be aware of the road surface conditions during driving). However, the high-cost sensor (e.g., LiDAR and camera)
based road surface sensing approaches cannot work properly in inclement weather conditions (e.g., fogging and snowing) due
to the line-of-sight requirement. The low-cost and ubiquitous smartphone-based road surface sensing approach is not reliable
and safe to use, since it relies on the vibration of the vehicle body to sense the road surface (i.e., the vehicle’s tires need to
touch the bumps on the road surface). Can we automate the contact-free road surface sensing with low-cost sensors for safe
driving without requiring the vehicle’s tires to touch the bumps on the road surface?
In this paper, we propose Tago, a system that can achieve contact-free road surface sensing with commodity passive RFID
tags. Instead of deploying RFID tags or readers along the road or lamp post (i.e., infrastructure-based deployment), we deploy
the reader inside of the vehicle and attach the tag and the reader’s antenna at the front end of the vehicle like the vehicle’s
headlights (i.e., infrastructure-free deployment). However, there is a great challenge to obtain the clean reflection from the
road surface, since the reflection may be drown in the backscattered signals due to multipath effect. Moreover, it is not reliable
to use the composite signals received at the reader to sense the road surface conditions. Therefore, we first comprehensively
analyse the variation of composite signals received at the reader. Then, we propose a signal cancellation approach to extract
the clean reflections from the road surface, such that we can accurately sense the road surface conditions for safe driving. Our
experiments with different vehicles (e.g., Honda Civic Frankenfish, Folsom, Flutter and CR-V Warner) driven on different
roadways (e.g., urban and residential area) show that Tago can effectively sense the road surface information.
CCS Concepts: • Human-centered computing → Ubiquitous and mobile computing; • Security and privacy → Human
and societal aspects of security and privacy.
Additional Key Words and Phrases: Commodity Passive RFID Tag, Road Surface Sensing, Contact-Free Sensing, Safe Driving
ACM Reference Format:
Wei Sun and Kannan Srinivasan. 2022. On the Feasibility of Securing Vehicle-Pavement Interaction. Proc. ACM Interact. Mob.
Wearable Ubiquitous Technol. 6, 1, Article 29 (March 2022), 24 pages. https://doi.org/10.1145/3517230

1 INTRODUCTION
1.1 Why Do We Need Road Surface Sensing?
The bad road surface conditions (e.g., bumps or potholes on the road surface) are caused by severe and dynamic
weather conditions, unexpected and heavy traffic loads and the normal wear/tear. Even though the governments
around the world spend a lot of money on maintaining the quality of the roadways, the bad road surface conditions
are still hazardous to drivers and pedestrians [19]. It is difficult for the governments to maintain the good quality
∗ Corresponding authors

Authors’ addresses: Wei Sun, The Ohio State University, Columbus, USA, [email protected]; Kannan Srinivasan, The Ohio State University,
Columbus, USA, [email protected].

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that
copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first
page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy
otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from
[email protected].
© 2022 Association for Computing Machinery.
2474-9567/2022/3-ART29 $15.00
https://doi.org/10.1145/3517230

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022. 29
29:2 • Sun and Srinivasan

Directional
Tag
antenna RF beam

orm-f le tag perf

Squigg
Tx Rx

FEAT
URE

orm
ance
hnology
, gen

ALN
eral-pur 2
ALN
-966
•
•
Tag

pos

-9
e Squ

662
iggle

Sho
r

Fig. 1. Tago’s architecture. The RFID reader’s antenna and RFID tag are attached at the front end of the vehicle to achieve
contact-free road surface sensing. Specifically, the reader’s antennas (i.e., Tx antenna and Rx antenna) are attached at the
front end of the vehicle and the tag is attached on the reader’s antenna. The shadow area (i.e., shown in the left figure) in
front of the vehicle indicates the RF beam transmitted from the reader’s antenna. So, the sensing area of Tago is the road
surface in front of the vehicle. Specifically, the backscattered signals reflected off the road surface will be analysed to sense
bumps or potholes on the road surface.

of roadways in the long term due to the limited municipal budget and transportation agencies. Also, the process
of road maintenance will last a long period of time, which will further cause the traffic congestion in some main
roadways. So, we cannot expect the roadways to be smooth all the time. These bad road surface conditions could
approximately cause $3 billions in property damage every year [2], and degrade the efficiency of the vehicles
(e.g., waste the fuel) as the vehicle is driven on bumpy road instead of the smooth road. The bad road surface
conditions become even worse for the autonomous driving systems, which require us to design the robust and
reliable machine learning models to capture the complex road surface conditions. Therefore, it is necessary for
the vehicular systems or drivers to be aware of the road surface conditions for economic and safe driving. This is
also why we have eye-on-road driving guideline [17].

1.2 Problems of Prior Art on Road Surface Sensing

Camera, radar or LiDAR-based road surface sensing. The straightforward solution is to instrument the
high-cost and advanced sensors (e.g., camera, radar or LiDAR sensor) on the vehicle [1] to accurately sense the
road surface conditions, which have been employed by modern autonomous vehicles (e.g., Waymo [39]). However,
these advanced sensors are expensive. More importantly, camera and LiDAR sensors require the line-of-sight to
work properly, which is not reliable in inclement weather conditions (e.g., fogging and snowing).
Smartphone or crowd-sourced smartphone-based road surface sensing. Recently, low-cost IMU sensors
(e.g., GPS, magnetometer, gyroscope and accelerometer) embedded in the smartphone are leveraged to sense
the road surface conditions [9, 21]. However, these smartphone-based or crowd-sourced smartphone-based
approaches cannot provide reliable road surface sensing, since the smartphone’s MEMS sensors are prone to
noise, drifts and biases. More importantly, they cannot achieve contact-free road surface sensing for safe driving
and alert the drivers of the hazardous road conditions ahead, since they sense the road surface based on the
vibration of the vehicle’s body which requires the vehicle’s tires to touch the bumps.
Therefore, it is critical to achieve the low-cost, ubiquitous and contact-free road surface sensing for safe driving.
More details of prior art on road surface sensing can be found in §2. Next, we illustrate why we choose commodity
passive RFID for road surface sensing.

1.3 Why Do We Choose Commodity Passive RFID for Road Surface Sensing?
Recently, the commodity passive RFID system has been widely exploited for ubiquitous sensing (e.g., touch
sensing [16, 33], soil moisture sensing [31], food/liquid sensing [12, 36], gesture recognition [14, 34, 35], etc.) due

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:3

to its low cost and small form factor. Note that the ubiquitous smartphones can be used as RFID reader, which is
exploited in [5, 8]. More importantly, since the commodity passive RFID relies on the radio frequency signals to
sense the objects (e.g., road surface), it can inform the driver of the hazardous road conditions ahead before the
vehicle’s tires touch the bumps on the road surface. Therefore, we believe commodity passive RFIDs are the good
choice to automate contact-free road surface sensing for safe driving. The rationale of using RFIDs is they are
low-cost and they can be the complement of camera/lidar/radar systems in NLOS scenarios. We think that the
active tag needs the battery, which requires us to replace it regularly. So, it is not convenient. Moreover, Tago’s
design is not inherited from the tag. So, Tago’s design can be general to any backscatter sensors including the
active tags. The advanced senors such as Lidar sensors, which cannot provide the large enough signal wavelength
to penetrate fog, rain and snow [10, 18, 22, 24]. Typically, the Lidar signal has the wavelength smaller than
0.01mm. Fog particles have the size of 10 to 15mm and rain drops have the size of 0.5 to 5mm, which are larger
than 0.01mm.

1.4 Tago’s Design

In this paper, we propose Tago, a system that can sense the road surface conditions using the commodity passive
RFID system for safe driving. To do so, we deploy the RFID reader in the vehicle and attach the RFID tags at
the front end of the vehicle to achieve contact-free sensing as shown in Fig. 1. However, there are three great
challenges to build a functional system. First, to achieve the contact-free sensing, we cannot attach RFID tags on
the road surface like the other RFID-tagged objects sensing [30, 32, 35]. The other infrastructure-based settings
(e.g., deploying reader on the lamp post [3]) are not appropriate in our scenario. Second, the backscattered signals
reflected off the road surface will be drown in the direct-path signals, which will degrade the sensing sensitivity
if we use composite signals received at reader to sense the road surface. Moreover, the hardware distortion of
tags/reader will introduce extra phase offset to the backscattered signals. At last, the bumps can be randomly
distributed on the road surface. So, we cannot use only one tag to reliably sense these bumps due to the orientation
sensitivity of commodity passive RFID tags.
To this end, we will first comprehensively analyse above system settings to exploit the variation of reflections
from the road surface. It turns out that we cannot just use the composite signals received at the reader to sense the
road surface conditions. So, we propose a window-sized signal cancellation approach to obtain the weak reflection
from the road surface (i.e., an emerging technique put forward in recent years [43? ]), whereby cancelling out
the reflections from other objects (e.g., vehicle’s body and curb). Note that the pedestrians on the sidewalk and
the other vehicles passing by will not be captured by the tags attached at the front end of the vehicle (as shown
in §6), since we harness the orientation sensitivity of RFID system [40] by facing the tag and reader’s antenna
to the road surface. We also need to mitigate the hardware imperfections of tag and reader through one-time
calibration. Since the commodity passive RFID system is orientation-sensitive, we can deploy two tags on the
vehicle (like the vehicle’s headlights) to detect the randomly distributed bumps on the road surface.
Result. We built a prototype of Tago with USRP N210 as the RFID reader and the general-purpose commodity
passive RFID tags. The extensive experiments show that Tago can accurately sense the road surface conditions.
We deploy the prototype of Tago on different vehicles (i.e., Honda Civic Flutter, Frankenfish, Folsom and CR-V
Warner). We do the experiments in our campus road, residential area and urban area. The experimental results
show the capability of Tago on contact-free road surface sensing. More details can be found in Sec. 6.
Limitation. Our RFID-based contact-free road surface sensing system marks an important step to shrink the gap
between the contact and contact-free road surface sensing for safe driving. Nonetheless, we will discuss some
limitations (e.g., sensing range and road surface profiling) of Tago’s design in Sec. 7.
Contributions. We summarize our contributions as follows:

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:4 • Sun and Srinivasan

• To the best of our knowledge, Tago is the first contact-free road surface sensing system for safe driving,
using commodity passive RFID tags attached at the front end of the vehicle.
• We comprehensively analyse Tago’s system settings. Then, we propose a window-sized signal cancellation
approach to obtain the weak reflection from the road surface. To achieve reliable sensing, we attach two
RFID tags at the front end of the vehicle (like the vehicle’s headlights) to sense the randomly distributed
bumps on the road surface.
• The extensive experimental results show the efficiency of our Tago’s design on road surface sensing,
comparing with the state-of-the-art smartphone-based road surface sensing approaches.
Below, we will present the prior work on road surface sensing, which will further comprehend our Tago’s
low-cost (i.e., in comparison to the cost of camera and LiDAR sensors) and contact-free (i.e., in comparison to the
smartphone or crowd-sourced smartphone-based approaches) road surface sensing.

2 RELATED WORK ON ROAD SURFACE SENSING

In the past literature, the road surface sensing has been exploited using different kinds of sensors such as stereo
cameras, LiDAR sensors and embedded sensors in the smartphones. So, we will discuss the related work in each
of the above sensors-based road surface sensing techniques and illustrate why they are not the good choice for
road surface sensing in comparison to Tago.

2.1 Advanced Sensor-Based Road Surface Sensing

The road surface condition can be obtained through the remote sensing, using the satellites or aerial imagery [38].
The remote sensing approaches can achieve large-scale sensing. But, it will cause the high-cost deployment
and data collection, especially when we need to frequently update the sensing data. More importantly, this
expensive remote sensing approach provides the poor accuracy and cannot provide the fine-grained road surface
information (e.g., bumps on the road surface). To improve the sensing accuracy, we can deploy some advanced
sensors (e.g., cameras and LiDAR sensors) on the vehicle to sense the road surface [1, 13, 45]. For example, the
Automatic Road Analyzer (ARAN [1]) is a vehicle instrumented with many high-quality sensors such as laser,
camera and LiDAR to collect the road surface information, by fusing all these sensing data. These sensors can
detect and localize the bumps on the road surface, estimate the road grade for efficient fuel usage and measure
the dimension of the pothole with high accuracy. The high accuracy is achieved with these expensive sensors,
which can just work properly in line-of-sight scenario. The high deployment cost limits its scalability. Moreover,
these sensors perform poorly in bad weather conditions such as fogging and snowing. The current autonomous
vehicles (e.g., Tesla [28] and Waymo [39]) highly rely on these sensors, since these sensors can provide high
accuracy with deep learning algorithms. However, they have high computational complexity and deployment
cost to train the deep neural networks in different weather conditions to have robust system.
Obviously, these advanced and expensive sensors could provide high accuracy in line-of-sight scenario. But,
they cannot work properly in non-line-of-sight scenario such as fogging and snowing. To alleviate the high
cost and achieve ubiquitous sensing, the smartphone-based road surface sensing has been exploited, as the
ubiquitous smartphone is embedded with different kinds of IMU sensors such as magnetometer, gyroscope, GPS
and accelerometer [9, 11, 21, 25, 46], which can be leveraged to sense the road surface.

2.2 Smartphone or Crowd-Sourced Smartphone-Based Road Surface Sensing

Nericell [21] can monitor road and traffic information with different sensors embedded in the smartphone (e.g.,
GPS, accelerometer and microphone). 𝑃 2 [9] uses the vibration and GPS sensors to assess the road surface
conditions. Since the smartphone-based road surface sensing approaches rely on IMU sensors to sense the road
surface, there are two main drawbacks using the smartphone to sense road surface conditions as follows:

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:5

Road surface
Data collection module Signal processing module Road surface sensing module information

Multipath mitigation Phase readings profile

orm-f
Squiggle
Bumpy road

tag perf

high-pe
FEATURE

rforman
hnology
ce, general

ALN
ALN-96
-purpos
•
•

62

-96
e Squiggl

62
e

Shor
Hardware imperfection Smooth road
Backscattered signals mitigation Road surface detection

Fig. 2. Tago’s operation workflow. Tago mainly consists of signal processing module and road surface sensing module.

• First, the gyroscope and accelerometer integrated into the smartphone are usually used for road surface
sensing, which are prone to noise, biases and drifts. For example, gyroscope is usually implemented with
Coriolis force [26], which will have the linear output and damping is ignored at the low frequency. When
the frequency increases, the damping will gradually become dominant which will cause the oscillation
with forced vibration. The frequency offset caused by the drifts will degrade the gyroscope data readings.
• Second, the physical principle of IMU sensors relies on vehicle’s body vibration to sense the road surface.
So, smartphone-based road surface sensing approach is contact sensing. This violates our design goal of
safe driving, which requires the system to alert the driver before the vehicle hits the bump on the road
surface. Therefore, it is essential to have contact-free road surface sensing approach.
Later on, the researchers try to improve the smartphone-based road surface sensing, through minimizing the
IMU sensor’s erroneous and leveraging the concept of crowdsourcing. However, they do not actually achieve
contact-free road surface sensing for safe driving. For example, RoadCare [29], [46] and [11] sense the road
surface conditions with multiple users’ smartphones, which will be uploaded to the central cloud to share these
road surface conditions with the other drivers. Actually, they still rely on the vehicle’s body vibration to sense
the road surface, which cannot achieve the goal of safe driving. Moreover, the crowdsourcing method incurs
significant overhead and may raise privacy and incentive issues.

3 TAGO OPERATION
Tago is a road surface sensing system that leverages the commodity passive RFID tags to detect the bumps on the
road surface. To do so, we attach the reader’s antenna and the tag at the front end of the vehicle.
Fig. 2 illustrates Tago’s operation workflow, which consists of two main modules: signal processing module
and road surface sensing module. Tago’s flow of operations includes the following steps: (1) The reader will
receive the backscattered signals from the tag; (2) The hardware imperfections of tag and reader’s antennas
should be mitigated with one-time calibration, since the tag and reader’s hardware imperfections are constant;
(3) We mitigate the multipath effect through signal cancellation. Since tag and reader’s antennas are deployed at
the front end of the vehicle as shown in Fig. 1, we mainly cancel the reflections from the vehicle’s body, curb
and the direct-path signals between reader’s antenna and the tag. After the signal cancellation, the remaining
part will be the backscattered signal reflected off the road surface; (4) We mainly extract the phase readings from
the backscattered signals reflected off the road surface for road surface sensing. To differentiate the bumpy and
smooth road surface, we compare the variance of phase readings to the predefined threshold. The bumps or
potholes on the road surface will introduce significant phase variation to the phase readings.
In the following, we will elaborate the design details of Tago. We start from modeling the RFID sensing in §4.1.
Then, we analyse the difficulties and challenges of contact-free RFID sensing in §4.2. To address the challenges,
we propose signal cancellation algorithm to obtain the backscattered signals reflected off the road surface in §4.3.
At last, we introduce reliable road surface sensing with two tags in §4.4

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:6 • Sun and Srinivasan

road reflection
other reflection
Virtual point
Q composite signal
1800

LOS
Tag
RFID reader
NLOS

Road surface 0 I

Fig. 3. Left figure shows the back-and-forth signal propagation between the reader and tag (i.e., LOS path), and the NLOS
signal propagation reflected off the virtual point (e.g., road curb and vehicle’s body) and road surface. Right figure shows the
vector representation of signal paths in IQ domain. The variation of road surface will cause the variation of backscattered
signals received at the reader.

4 TAGO DESIGN
4.1 Employing RFID Signal for Sensing
Since the commodity passive RFID tags are battery-free, they need to be activated by the external reader
through the high-power signals (i.e., continuous wave). Let’s assume the transmitted signal from the reader is
𝑆𝑟𝑡 = |𝑆𝑟𝑡 | 𝑒 𝑗𝜃𝑟 (𝜃 𝑟 denotes the phase offset due to the reader’s hardware imperfection), which will be propagated
to the nearby tags. The tag will receive signals 𝑆𝑡 = 𝑆𝑟𝑡 ℎ𝑟𝑡 ℎ𝑡 , ℎ𝑟𝑡 = |ℎ𝑟𝑡 | 𝑒 𝑗𝜃𝑟𝑡 , ℎ𝑡 = |ℎ𝑡 | 𝑒 𝑗𝜃𝑡 , where ℎ𝑟𝑡 denotes the
wireless channel from the reader to the tag and ℎ𝑡 denotes the tag’s antenna gain. Suppose that the reflection
coefficient is 𝛼 = |𝛼 | 𝑒 𝑗𝜃𝛼 , which is a constant value given the specific RFID tag. The received signals at the
tag will be backscattered to the reader through ON-OFF keying modulation. Then, the reader will receive tag’s
reflections 𝑆𝑟 ′ = 𝑆𝑡 𝛼ℎ𝑡 ℎ𝑡𝑟 = 𝑆𝑟𝑡 𝛼ℎ𝑡2ℎ𝑎2 = 𝑆𝑟𝑡 𝛼ℎ𝑡2ℎ𝑎2 𝑒 𝑗 (𝜃𝑟 +𝜃𝛼 +2𝜃𝑎 +2𝜃𝑡 ) , where 𝜃 𝑎 = 𝜃 𝑟𝑡 , 𝜃 𝑎 = 𝜃 𝑡𝑟 and ℎ𝑎 = ℎ𝑟𝑡 = ℎ𝑡𝑟
due to the reciprocity property of the wireless channel.
After the reader receives the backscattered signals from the tag, we can extract the signal strength and phase.
The received signal strength of the backscattered signal is denoted as follows:
𝑆𝑖𝑔𝑛𝑎𝑙 𝑠𝑡𝑟𝑒𝑛𝑔𝑡ℎ = 20 𝑙𝑜𝑔 |𝑆𝑟 ′ | = 20 𝑙𝑜𝑔 |𝑆𝑟𝑡 | + 20 𝑙𝑜𝑔 |𝛼 | + 40 𝑙𝑜𝑔 |ℎ𝑎 | + 40 𝑙𝑜𝑔 |ℎ𝑡 |
The received signal phase of the backscattered signal is denoted as follows:
𝑆𝑖𝑔𝑛𝑎𝑙 𝑝ℎ𝑎𝑠𝑒 = 𝜃 𝑟 + 𝜃 𝛼 + 2𝜃 𝑎 + 2𝜃 𝑡
Since the phase readings can provide more finer resolution than the amplitude readings [20, 30, 32, 35, 40, 42],
most of the work employs the phase readings to do RFID-based sensing. So, we will also leverage the signal phase
extracted from tag’s backscattered signals to achieve road surface sensing. Note that the extracted signal phase
from the backscattered signals is related to the hardware imperfection of the tag (i.e., 𝜃 𝑡 , 𝜃 𝛼 ) and reader (i.e., 𝜃 𝑟 ),
which will be addressed in §4.3. Below, we will address the challenge of enabling contact-free RFID sensing.

4.2 Analysing the Impacted Factors on Contact-Free Road Surface Sensing

Instead of deploying the tags or reader on existing infrastructure (e.g., lamp posts along the road), we attach
the reader’s antenna and tag at the front end of the vehicle to achieve contact-free road surface sensing. From
the above discussion, we can see that there are two types of backscattered signals between the reader and the

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:7

dynamic component
static component
Q composite signal Q Q
Q

0 I 0 0
I 0 I I

Fig. 4. The impacted factors on using composite signals to sense the road surface conditions. The leftmost figure shows that
the amplitude variation of the dynamic component affects the phase variation of the composite signals. The second figure
shows that the phase variation of dynamic component will affect the phase variation of the composite signals. The third
figure shows that the amplitude of the static component affects the phase variation of the composite signals. The size of
circle indicates the amplitude of the dynamic component. The rightmost figure shows the location of the bump or pothole on
the road surface will affect the phase variation of the composite signals.

tag: line-of-sight (LOS) signals and non-line-of-sight (NLOS) signals as shown in Fig. 3. The signals directly
propagated between the reader’s antenna and the tag indicate the LOS path. There are some other objects (e.g.,
vehicle’s body) which can reflect the signals. We regard these objects as a virtual point. The virtual point and
road surface cause the NLOS signal transmission.
The vector representation of LOS and NLOS signal paths are shown in Fig. 3. The LOS path and the signal
path traversing over the virtual point can be seen as the static component. The reflection from the road surface
can be regarded as the dynamic component due to the change of road surface (e.g., from smooth to bumpy).
Since the road surface can change, the backscattered signals from the road surface will change (i.e., amplitude or
phase variation) due to the different propagation distance. We use the size of the circle in Fig. 3 to indicate the
signal amplitude reflected off the road surface. The reader will just receive the composite signals. When the phase
difference between the static component and the average dynamic component is 180◦ as shown in Fig. 3, we can
see the obvious phase variation (i.e., Δ𝜃 ) of the composite signals. However, it is difficult to know if this phase
variation is enough to sense the road surface conditions. Next, we present factors that will affect the sensing
ability of using composite signals for road surface sensing.
Fig. 4 shows the impacted factors that can affect the phase variation of the composite signals received at
the reader. From the leftmost figure, we can see that the amplitude of the signals reflected off the road surface
will affect the phase variation of the composite signals received at the reader. As the amplitude variation of the
dynamic component becomes larger, the phase variation of the composite signals becomes larger. However, the
amplitude of the signals reflected off the road surface depends on the material contents of the road surface and
the distance between the reader’s antenna and the road surface. The second figure shows that the phase variation
of the dynamic component will affect the phase variation of the composite signals. As the phase variation of
dynamic component becomes larger, the phase variation of the composite signals becomes larger. The third
figure shows that the amplitude of the static component will affect the phase variation of the composite signals.
As the amplitude of the static component becomes smaller, the phase variation of composite signals becomes
larger. However, it is difficult to control above three factors to maximize the phase variation of the composite
signals received at the reader [41]. Note, the location of bumps or potholes on the road surface will also affect the

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:8 • Sun and Srinivasan

o o
hrot hrot

hRt hRt
hRt hT
hT hT
hrt hrt
hrt

htr htr
htr
hRr hRr
hRr hrvt
v
(a) (b) (c)

Fig. 5. Modeling the propagation of the backscattered signal. (a) When there are no other objects around the commodity
passive RFID system, we just consider line-of-sight propagation between the reader and tag. (b) When there are other objects
around the commodity passive RFID system, we regard 𝑜 as the other reflectors. We should consider the propagation path
traversing over 𝑜 from reader to the tag. (c) Extracting the reflection from the road surface (i.e., denoted as 𝑣) through the
cancellation. Note that our objective is to obtain channel ℎ𝑟 𝑣𝑡 for road surface sensing.

variation of composite signals as shown in the rightmost figure. So, we need to directly extract the reflections
from the road surface to obtain the road surface information.
Instead of maximizing the phase variation of the composite signals by introducing another tag to harness the
tag-tag coupling effect (e.g., TagSMM [41] is designed for vibration sensing) for road surface sensing, we propose
to extract the clean backscattered signals reflected off the road surface to sense the road surface conditions. This
is because TagSMM need carefully deploy non-target tag to harness the coupling effect between the target tag
and non-target tag, which is not reliable when both of the tags are attached at the front end of the vehicle due
to the vibration of the vehicle’s body. RF-EAR [44] leverages the Orthogonal Matching Pursuit algorithm to
characterize different vibration sources by scaling the different factors for different vibration sources, which
is designed for vibration sensing. In Tago, the tag and reader’s antenna are co-located on the front end of the
vehicle (i.e., infrastructure-free), which is different from the existing infrastructure-based RFID sensing systems.

4.3 Enabling Contact-Free Road Surface Sensing through Signal Cancellation

To obtain the clean reflections from the road surface, we propose the backscattered signal cancellation approach
to sense the road surface.
Since we deploy the reader’s antenna and tags at the front end of the vehicle, the tag will modulate the
composite signal from the reader (i.e., signals transmitted from reader’s transmit antenna) and the background
reflectors (i.e., signals reflected off the different objects around the tag). Then, the composite signal will be
reflected off the tag through ON-OFF keying modulation. The reflections from the bump on the road surface will
be drown in the background reflections, thereby it’s essential to disentangle the road surface reflections from the
background reflections. This is because the direct-link signals will be the dominant due to Tago’s settings.
Modeling backscatter communication in LOS. We first present the case when there are no reflectors around
commodity passive RFID system (i.e., LOS setting) as shown in Fig. 5(a). As mentioned in Sec. 4.1, in commodity
passive RFID system, the reader should initialize the communication and activate the tags by sending the constant
continuous wave (i.e., cw). After the tag is activated, it will response by backscattering the cw signals with
ON-OFF keying modulation. Let’s assume the reader sends an request 𝑥 (i.e., complex scalar value) to the tag
within its communication range. The tag will backscatter the received signal through ON-OFF keying modulation.

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:9

So, the reader will receive the backscattered signal as follows:

𝑦𝑑 = ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟𝑡 ℎ𝑅𝑡 𝑧𝑥 + ℎ𝑅𝑟 ℎ𝑟𝑟 ℎ𝑅𝑡 𝑥, (1)
where ℎ𝑡𝑟 denotes the wireless channel from tag to reader’s receive antenna, and ℎ𝑟𝑟 denotes the channel from
the reader’s transmit antenna to receive antenna. ℎ𝑟𝑡 denotes the channel from reader’s transmit antenna to
tag. 𝑧 denotes the data transmitted by tag. We use ℎ𝑅𝑡 , ℎ𝑅𝑟 and ℎ𝑇 to denote the hardware distortion at reader’s
transmit antenna, receive antenna and tag respectively. ℎ𝑅𝑟 ℎ𝑟𝑟 ℎ𝑅𝑡 𝑥 represents the self-interfered signals received
at the reader’s receive antenna. The impinged signals on the tag body are ℎ𝑟𝑡 ℎ𝑅𝑡 𝑥. Since the passive RFID tag
simply backscatters the impinged signals from the reader, we only consider ℎ𝑇 once. Then, the backscattered
signals received by the reader’s receive antenna are ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 𝑧ℎ𝑟𝑡 ℎ𝑅𝑡 𝑥. Note that the RFID reader is capable to
do self-interference cancellation to eliminate the signal leakage at reader. Therefore, the reader will receive the
following signal:
𝑦𝑑 = ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟𝑡 ℎ𝑅𝑡 𝑧𝑥 + ℎ𝑟𝑟 ℎ𝑅𝑡 ℎ𝑅𝑟 𝑥 − ℎ𝑟𝑟 ℎ𝑅𝑡 ℎ𝑅𝑟 𝑥
= ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟𝑡 ℎ𝑅𝑡 𝑧𝑥 (2)

Backscatter communication in NLOS. When there are other objects (e.g., vehicle’s body or curb) around the
commodity passive RFID system. The reader will receive the composite reflections from the tag and other objects
as shown in Fig. 5(b). Instead of modulating the signal from the reader, the tag will modulate the composite
signal from the reader and other objects. So, the impinged signal on the tag is ℎ𝑟𝑡 ℎ𝑅𝑡 𝑥 + ℎ𝑟𝑜𝑡 ℎ𝑅𝑡 𝑥, where ℎ𝑟𝑡 ℎ𝑅𝑡 𝑥
denotes the impinged signal from the reader and ℎ𝑟𝑜𝑡 ℎ𝑅𝑡 𝑥 denotes the impinged signal traversing over the other
objects. We use ℎ𝑟𝑜𝑡 to denote the channel traversing over the other objects from the reader to tag. After the
reader cancels out the self-interference signal, the reader will receive the following signal:
𝑦𝑜 = ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟𝑡 ℎ𝑅𝑡 𝑧𝑥 + ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟𝑜𝑡 ℎ𝑅𝑡 𝑧𝑥 . (3)
As we can see from the above equation, our objective is to estimate the channel affected by the other objects (e.g.,
road surface). However, ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟𝑡 ℎ𝑅𝑡 will dominate the received signal 𝑦𝑜 , which will affect the road surface
sensing. Next, we will illustrate how to extract the channel traversing over the road surface through cancellation.
Extracting the reflection from road surface through cancellation. Since our objective is to obtain the
reflections from the road surface, we need to cancel out the reflections from the other objects such as the
vehicle’s body. Fig. 5(c) shows the signal paths between the reader and tag, when we drive the vehicle on the
road. We use 𝑣 to denote the reflection from the road surface. So, the tag will modulate the impinged signal
ℎ𝑟𝑡 ℎ𝑅𝑡 𝑥 + ℎ𝑟𝑜𝑡 ℎ𝑅𝑡 𝑥 + ℎ𝑟 𝑣𝑡 ℎ𝑅𝑡 𝑥 with ON-OFF keying modulation, where ℎ𝑟 𝑣𝑡 ℎ𝑅𝑡 𝑥 denotes the signal traversing
over the road surface. We use ℎ𝑟 𝑣𝑡 to denote the channel traversing over the road surface from the reader to the
tag. After self-interference cancellation, the reader will receive the signal as follows:
𝑦 𝑣 = ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟𝑡 ℎ𝑅𝑡 𝑧𝑥 + ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟𝑜𝑡 ℎ𝑅𝑡 𝑧𝑥 + ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟 𝑣𝑡 ℎ𝑅𝑡 𝑧𝑥 . (4)
The backscattered signals received at the reader mainly consist of two components: the direct reflection and
multi-path reflection. For the multi-path reflections from tag to reader, reflected signals from tag come from
two sources: directly impinged signals from reader and the other objects’ reflections, which will be used by the
tag to modulate its own data. For the tag’s reflected signals directly coming from the reader, when they are
backscattered by the tag over other objects, they are also included in the second component of right side of Eq.(4)
due to the reciprocity of dominant and over-the-air wireless channel. For the tag’s reflected signals coming from
the other objects’ reflections, when they are backscattered by the tag over other objects, they become weaker due
to multiple attenuation. So, in Eq.(4), we just count the reflections over the other objects once. To demonstrate
this, we show the IQ constellation of the backscattered channel before and after signal subtraction in Fig. 6. As

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:10 • Sun and Srinivasan

Fig. 6. Backscattered channel in IQ domain before and after dominant signal cancellation in Eq.(4). Note, Eq.(5) shows the
signal cancellation process.

we can see, after the signal subtraction, the channel samples are distributed around origin point. This is because
we cancel out the dominant contribution from the composite backscattered channel. This indicates that we can
ignore the reflections traversing over objects multiple times and the impact of RFIC’s impedance due to their
weak impact.

Algorithm 1 Road surface sensing with signal cancellation algorithm

Require:
• Commodity RFID passive system and the vehicle;
Ensure:
• The expected channel traversing over the road surface from the reader to tag ℎ𝑟 𝑣𝑡 ;
1: Distortion mitigation: Estimating 𝐻 ′ and
1 𝑑𝑟𝑡
ℎ𝑟𝑡 = ℎ𝑡𝑟 = 2 𝑒 −𝑗2𝜋 𝜆 𝑚𝑜𝑑 2𝜋 ;
𝑑𝑟𝑡
2: Backscattered signal extraction: Obtaining the backscattered signals received at the reader
𝑦𝑜 = ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟𝑡 ℎ𝑅𝑡 𝑧𝑥 + ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟𝑜𝑡 ℎ𝑅𝑡 𝑧𝑥;
3: Real-time backscattered signal cancellation: Obtaining the expected channel through cancellation
𝑦 𝑣′ = 𝑦 𝑣 − 𝑦𝑜 = ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟 𝑣𝑡 ℎ𝑅𝑡 𝑧𝑥,
where 𝑦 𝑣 is obtained through equation (4);
ℎ𝑣
4: Return the channel estimation ℎ𝑟 𝑣𝑡 = 𝑑 ;
−𝑗 2𝜋 𝑟𝑡 𝑚𝑜𝑑 2𝜋
𝐻 ′ 12 𝑒 𝜆
𝑑𝑟𝑡

To extract the reflections from the road surface, we need to do the subtraction as follows:
𝑦 𝑣′ = 𝑦 𝑣 − 𝑦𝑜 = ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟 𝑣𝑡 ℎ𝑅𝑡 𝑧𝑥, (5)
where ℎ 𝑣 = ℎ𝑅𝑟 ℎ𝑡𝑟 ℎ𝑇 ℎ𝑟 𝑣𝑡 ℎ𝑅𝑡 will be extracted through the backscattered channel estimation. Since we deploy
tag and reader’s antenna at the front end of the vehicle, the reflections from the vehicle’s body are quite stable.
So, we can do above subtraction, especially in the outdoor environment. Note that the above subtraction can
also cancel out the effect of engine’s vibration. Since our objective is to obtain the channel ℎ𝑟 𝑣𝑡 , we still need to
eliminate the impact of hardware distortion and the line-of-sight propagation between the reader and tag. The

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:11

Left tag

Right tag

Fig. 7. Two tags are attached at the left and right front Fig. 8. Left tag and right tag exhibit the different phase
of the vehicle respectively to sense the road surface con- readings over time, indicating the different road surface
ditions. conditions.

line-of-sight propagation channel between the reader and tag can be easily modeled as follows:
1 −𝑗2𝜋 𝑑𝑟𝑡 𝑚𝑜𝑑 2𝜋
ℎ𝑟𝑡 = ℎ𝑡𝑟 = 2
𝑒 𝜆 , (6)
𝑑𝑟𝑡
where 𝑑𝑟𝑡 denotes the distance between the reader and tag. We have ℎ𝑟𝑡 = ℎ𝑡𝑟 due to the geometric symmetry. As
we can see, ℎ𝑟𝑡 and ℎ𝑡𝑟 will be easily calculated with known distance 𝑑𝑟𝑡 between the reader and tag. Next, we
have to compensate the hardware distortion, which can be calibrated once and used for the rest of life [43]. Then,
we obtain the expected channel ℎ𝑟 𝑣𝑡 reflected off the road surface.
ℎ𝑣
ℎ𝑟 𝑣𝑡 = 𝑑𝑟𝑡
, (7)
𝐻 ′ 𝑑12 𝑒 −𝑗2𝜋 𝜆 𝑚𝑜𝑑 2𝜋
𝑟𝑡

where 𝐻 ′ = ℎ𝑅𝑟 ℎ𝑇 ℎ𝑅𝑡 denotes the calibrated hardware distortion of the reader and tag.
Road surface sensing with signal cancellation algorithm. We present Algorithm 1 to extract the reflection
from the road surface. As we can see, we need three steps to extract the reflection from the road surface as
follows:
• Distortion mitigation: We first need to estimate the impact of hardware distortion and line-of-sight
propagation channel between the reader and tag in line-of-sight setting. After pre-calibration, we will
obtain 𝐻 ′ and equation (6).
• Backscattered signal extraction: We deploy our commodity passive RFID system at the front end of the
vehicle to sense the road surface conditions. Specifically, we will obtain the received signal at the reader
shown in equation (3).
• Real-time backscattered signal cancellation: As we drive the vehicle, we collect the received signal at
the reader shown in equation (4). We do subtraction in equation (5), which is obtained through subtraction
of the consequent and window-sized backscattered signals. Then, we estimate the channel traversing over
the road surface from the reader to the tag as shown in equation (7). By doing this, the impact of vehicle’s
body vibration and the other external reflectors (e.g., curb) could be mitigated, such that we can observe a
peak/trough indicating the bump/pothole in the phase profile.

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:12 • Sun and Srinivasan

Smartphone

Reader
Antenna & tag Laptop

Fig. 9. Experimental setup. The left figure shows RFID reader’s two antennas and one tag are attached at the front end of
Honda Civic Flutter. The same setup for Honda CR-V Warner can be seen in Fig. 1. The right figure shows RFID reader, PC
and smartphone (i.e., used for system performance comparison) are deployed inside of Honda CR-V Warner. Note that RFID
reader is connected with two directional antennas (i.e., Tx transmit antenna and Rx receive antenna). We attach RFID tag on
the reader’s antennas as shown in the zoom-in figure of Fig. 1.

4.4 Enabling Reliable Sensing with Two RFID Tags

Since the commodity passive RFID system is orientation-sensitive and the bumps are randomly distributed on the
road surface, it is important to detect these bumps with multiple tags attached to the vehicle for reliable sensing.
Considering the beamwidth of the reader’s directional antenna (i.e., 105◦ [4]) and the width of the vehicle and
road, we can just attach two tags at the front of the vehicle as shown in Fig. 7. Each tag will be responsible to
sense half side of the road surface, which looks like two headlights of the vehicle.
Fig. 8 shows the phase readings over time from two tags. As we can see, two tags present different phase
readings, since the road surface conditions are different. We can deploy more tags at the front of the vehicle to
achieve reliable and fine-grained road surface sensing. However, we have to consider the low tag reading rate
due to the collisions, which will degrade RFID’s sensing ability. We will discuss this problem in Sec. 7.

5 IMPLEMENTATION AND EVALUATION

Hardware. We have implemented a prototype of Tago with commodity passive RFID tags and USRP N210 [23]
as reader, which is FCC-compliant working at the frequency band of 902-928MHz. We only extract phase and
amplitude of the backscattered signal from the reader. Fig. 9 illustrates Tago’s hardware components. The circular
polarized antennas [4] are connected to the reader, which has antenna gain of 11dBi and beamwidth of 105◦ . We
use different commodity passive RFID tags from atlasRFID store [6] (e.g., ALN-9662, Monza 4D and R6-P [7])
attaching to the front end of the vehicle during the experiments, which are widely used with reading range of
5-15 meters at best [37]. The USRP reader can interrogate the tags and receive the backscattered signals from
tags to extract phase and amplitude readings, via the Ethernet cable connecting with a host PC.
Software. The reader’s implementation is obtained from the previous work [15], which can enable the RFID
communication using USRP N210 as reader. We just use this reader to collect data. Our road surface sensing
algorithms are implemented offline with MATLAB for signal processing. The commodity passive RFID system
complies with C1G2 standard using slotted ALOHA protocol to interrogate tags. There are two main steps
during tag interrogation. At the first step, the activated tags will reply RN16 to the reader for channel access
permission (i.e., query step). During query step, the reader will initialize some parameters such as M (i.e., the
number of frames/inventory rounds), Q (2𝑄 indicates the number of time slots in one frame/inventory round),
four sessions (i.e., used for multi-reader reading), flags (e.g., SL flag and inventoried flag) and so on. In default, we
set Q=0, meaning there is one time slot in each frame, since we only have one tag. This can maximize the tag

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:13

2 2

Phase
Phase

0 vehicle is moving vehicle is static 0

vehicle is vehicle is static

-2 -2 moving

0 1000 2000 3000 0 1000 2000 3000

Inventory Round Inventory Round

Fig. 10. Phase readings over time, when driving the ve- Fig. 11. Phase readings over time, when driving the ve-
hicle on the smooth road with speed of 10mph. hicle on the smooth road with speed of 20mph.
-1.6
-0.8
-1.8
-1
Phase

-2

Phase
-1.2
-2.2
-1.4 vehicle is
vehicle is vehicle is static
-2.4 moving
-1.6 static
vehicle is moving
-1.8 -2.6
500 1000 1500 2000 2500 500 1000 1500 2000 2500
Inventory Round Inventory Round

Fig. 12. Phase readings over time, when driving the ve- Fig. 13. Phase readings over time, when driving the ve-
hicle on the bumpy road with speed of 10mph. hicle on the bumpy road with speed of 20mph.

reading rate (Usually, we use tag reading rate to measure how often we can obtain the channel information in
RFID communication). The value of 𝑀 depends on how long we drive the vehicle (we set it as 5000 by default).
The other parameters are set as default value (e.g., zero). At the second step, the reader will chose one tag to
communicate and the chosen tag will reply EPC to the reader. The channel information extraction happens in the
second step with preamble based channel estimation. The reader will report the phase angle through channel
sounding using EPC packets.
Compared approach. We mainly compare the performance of Tago with smartphone-based road surface sensing.
We use iphone 7 to measure the data readings from the gyroscope of the smartphone. The smartphone will be
deployed and fixed at the different positions inside of the vehicle (e.g., windshield, air-conditioning vents, etc.) as
shown in Fig. 9.
Experimental setup. During the experiments, we use different vehicles (e.g., Honda Civic and Honda CRV
Warner shown in Fig. 9). We will drive the vehicle on the different road conditions (e.g., bumpy and smooth road)
with different speed (e.g. 10mph, 15mph and 20mph). We drive the vehicle in urban (Fig. 26) and residential area
(Fig. 28) for road surface sensing.

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:14 • Sun and Srinivasan

Fig. 14. CDF of phase variance when driving on the smooth and bumpy road with different speed.

6 EXPERIMENTAL RESULTS
6.1 Micro Benchmarks
6.1.1 Effect of Sensing Smooth/Bumpy Road. The different road surface conditions will reflect the backscattered
signals differently.
Method. To see the effect of sensing smooth and bumpy road, we drive the vehicle on the smooth and bumpy
road respectively with different driving speed. Moreover, we compare the phase readings, when vehicle is moving
and static. To do so, we drive the vehicle for a period of time and stop the vehicle to see the variation of phase
readings.
Result. Fig. 10 and Fig. 11 show the phase readings over time, when we drive the vehicle on the smooth road
with speed of 10mph and 20mph and stop the vehicle during driving. We can see that the phase variation is stable
over the time after we stop the vehicle (i.e., the vehicle is static). Fig. 12 and Fig. 13 show the phase readings over
time, when we drive the vehicle on bumpy road with speed of 10mph and 20mph and stop the vehicle during
driving. As we can see, the phase variation is significant, when we drive the vehicle on the bumpy road. When
we stop the vehicle, the phase readings become stable. This is because the road surface conditions will affect
the backscattered signals, such that we can use the variation of backscattered signals to sense the bumpy and
smooth road surface. Fig. 14 shows the CDF of the phase variance, when we drive the vehicle on smooth and
bumpy road with different speed. We can see that the phase variance is consistent, when we drive the vehicle
with different speed on the same road surface. When we drive the vehicle on the bumpy road, the phase variance
becomes larger in comparison to the phase variance when we drive the vehicle on smooth road. So, Tago can
sense the road surface conditions with contact-free commodity passive RFID sensing system.
6.1.2 Impact of Vehicle’s Vibration, Passing-by Passengers and Other Vehicles. The backscattered signals are
reflected off the different objects around the vehicle such as the vehicle’s body, passing-by passengers and
vehicles.
Method. To measure the impact of the passing-by passengers and other vehicles, we drive the vehicle on the
smooth road at urban area, which can mitigate the impact of road surface condition on the backscattered signals.
During the experiments, there are passengers walking on the sidewalk and the other vehicles passing by our
vehicle. We will show the phase and amplitude readings over time (i.e., inventory rounds), when we drive the
vehicle and the vehicle is static respectively.
Results. Fig. 15 and Fig. 16 show the phase and amplitude readings over inventory rounds, when the vehicle
is static and its engine is on. We can see the stable phase and amplitude readings. The vibration of vehicle’s
engine will not affect backscattered signals. This is because Algorithm 1 has already accounted for the vibration

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:15

2
2

Amplitude
Phase

1
-2

0 100 200 300 400 0

0 100 200 300 400
Inventory Round Inventory Round
Fig. 15. Phase readings over time, when the vehicle is
Fig. 16. Amplitude readings over time, when the vehicle
static and the engine is on.
is static and the engine is on.
3

2
2

Amplitude
Phase

1
-2

0 200 400 600 0

0 200 400 600
Inventory Round Inventory Round
Fig. 17. Phase readings over time, when we drive the
Fig. 18. Amplitude readings over time, when we drive
vehicle on the smooth road at urban area.
the vehicle on the smooth road at urban area.

of vehicle’s body. Then, we drive the vehicle on the smooth road to measure the impact of passing-by passengers
and other vehicles. Fig. 17 and Fig. 18 show the phase and amplitude readings over inventory rounds, when we
drive the vehicle on the smooth road at urban area. As we can see, the phase and amplitude readings are quite
stable, indicating that the passing-by passengers and other vehicles will not affect the backscattered signals. This
is because the commodity passive RFID system uses the directional antenna with narrow beamwidth, which will
only concentrate the signals in the front of the vehicle. The passing-by passengers and other vehicles will not be
in the reader’s field-of-view, such that they cannot affect the backscattered signals.
6.1.3 Impact of Pothole Size. The pothole on the road surface may have different size. Intuitively, the bigger
pothole is more easier to be detected than the smaller pothole.
Method. To see the impact of pothole size on road surface sensing, we drive the vehicle with speed of 10mph
passing by the big pothole (i.e., with size of 85x85x4cm) and small pothole (i.e., with size of 28x28x3.5cm) on the
road surface.
Result. Fig. 19 shows the phase readings from the tag, when we drive the vehicle passing the big pothole and
small pothole. When we drive the vehicle passing the big pothole, we can see a fat trough in the phase reading
profile. But, when we drive the vehicle passing a small pothole, we can see a sharp trough in the phase reading
profile. These results indicate that the size of pothole will affect Tago’s road surface sensing. Potentially, we can

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:16 • Sun and Srinivasan

Fig. 19. Left figure shows the phase readings over time, when the vehicle passes the big pothole with size of 85x85x4cm.
Right figure shows the phase readings over time, when the vehicle passes the small pothole with size of 28x28x3.5cm.

1.5 1.5 Left tag

Right tag
1
Left tag
Phase

Phase
0.5 Right tag 1
0
-0.5
0.5
-1
200 400 600 800 1000 1200 200 400 600 800 1000 1200
Inventory Round Inventory Round

Fig. 20. Pothole located at the left side of the Fig. 21. Pothole located at the right side of the
road is detected by the left tag attached on the road is detected by the right tag attached on
vehicle. the vehicle.

leverage the phase readings to create the profile of the pothole and bump on the road surface. However, since
the tags are just attached at the front end of the vehicle, we can only obtain one dimensional information of the
pothole or bump (e.g., width). Profiling the pothole or bump on the road surface (e.g., the length, height and
width of the bump or pothole) will be left for our future work.
6.1.4 Effect of Using Two Tags. The commodity passive RFID system employs the directional antenna to concen-
trate RF signals for long-range RFID communication. And, the potholes or bumps are randomly distributed on
the road surface.
Method. To reliably sense the road surface conditions and detect the potholes or bumps on the road surface, we
attach two tags at the left and right front end of the vehicle like the vehicle’s headlights. We will first drive the
vehicle attached with two tags to pass the pothole on the left side of the road surface. Then, we drive the vehicle
attached with two tags to pass the pothole on the right side of the road surface.
Result. Fig. 20 and Fig. 21 show the phase readings of backscattered signals from two tags, when we drive the
vehicle passing the pothole located at the left or right side of the road surface. We can see that the phase variation
of left tag is significant, when the pothole is located at the left side of the road surface. The phase variation of
right tag is significant, when the pothole is located at the right side of the road surface. Since the commodity
passive RFID system is orientation-sensitive, only one tag can sense the pothole located at the left or right side of

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:17

Bump is sensed

tago
gyro. x
gyro. y
gyro. z

Random peak

Fig. 22. Phase readings of backscattered signal for the RFID-based contact-free road surface sensing and the rate of rotation
measurements from gyroscope for smartphone-based contact road surface sensing.

the road surface. This potentially indicates that the passing-by passengers and other vehicles will not affect the
backscattered signals due to the orientation sensitivity of RFID tags. Note, we can deploy more than two tags at
the front end of the vehicle to enhance the sensing ability of Tago for road surface sensing. However, we have to
consider the low tag reading rate, as we increase the number of tags.

6.2 Contact-Free Road Surface Sensing for Safe Driving

6.2.1 Comparison to the Smartphone-Based Road Surface Sensing. The principle of smartphone-based road
surface sensing relies on the MEMS gyroscope or accelerometer to detect the vehicle’s body vibration caused by
the bump or pothole on the road surface. Therefore, the smartphone-based road surface sensing system requires
the vehicle’s front tires to hit the bumps or potholes on the road surface to accurately detect them, which is
actually violating the goal of safe driving. More details can be found in Sec. 2.
Method. The commodity passive RFID system can provide the contact-free road surface sensing by deploying
the reader’s antenna and tag at the front end of the vehicle. As shown in Fig. 9, we deploy reader’s antenna
and RFID tag at the front of the vehicle for contact-free road surface sensing. There is a smartphone fixed at
the arbitrary location inside of the vehicle (e.g., windshield). During the experiments, we will drive the vehicle
approaching the bump as shown in Fig. 9. But, we will stop the vehicle before the vehicle’s front tires touch the
bump. By doing this, we can imagine that the sensing data read from the smartphone will not be affected by the
bumps, since the vehicle’s body does not vibrate. However, the backscattered signals from RFID tags attached at
the front end of the vehicle will be affected.
Result. Fig. 22 shows the phase readings from the RFID tag and the rate of rotation measurements from gyroscope
embedded in the smartphone. We can see that the phase readings are abruptly changed at 35 seconds. However,
the rate of rotation measurements from the gyroscope keeps stable over time during the experiments. Therefore,
we can claim that the smartphone-based road surface sensing indeed requires vehicle’s body vibration caused by
the bumps or potholes on the road surface, which cannot achieve safe driving. However, the commodity passive
RFID system can achieve contact-free sensing, which does not require the vehicle’s tires to touch the bumps or
potholes due to the fact of RF signals propagation. In comparison to the smartphone-based road surface sensing,
Tago indeed achieves safe driving. Therefore, we believe commodity passive RFID system is a good choice to
achieve road surface sensing for safe and economic driving due to its low-cost (considering the smartphone can
be the reader [5, 8]), small form-factor and contact-free sensing.
6.2.2 Effect of Contact-Free Road Surface Sensing with Tago. We have shown the contact-free sensing using
Tago in comparison to the smartphone-based road surface sensing, which will enable the safe driving. Then, the
problem becomes how far Tago can detect the pothole/bump on the road surface.

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:18 • Sun and Srinivasan

tago
gyro.x
gyro.y
gyro.z

Fig. 23. Phase readings from the tag and gyroscope data from the smartphone, when we drive the vehicle in the speed of
10mph passing the bump on the road surface.

tago
tagogyro.x
gyro.xgyro.y
gyro.ygyro.z
gyro.z

Fig. 24. Phase readings from the tag and gyroscope data from the smartphone, when we drive the vehicle in the speed of
15mph passing the bump on the road surface.

Method. Here, we measure the phase readings from the tag and the data from the gyroscope of the smartphone,
when we drive the vehicle in different speed. We will see how far away our Tago can sense the bump on the road
surface in comparison to using the smartphone.
Result. Fig. 23 shows the phase readings from tag and gyroscope data from smartphone, when we drive the
vehicle in the speed of 10mph passing the bump on the road surface. As we can see, Tago can sense the bump on
the road surface about 2.2 seconds earlier than using the smartphone to sense the bump. Therefore, Tago can
sense the bump on the road surface in contact-free way, which can allow about 2.2 seconds for the vehicular
system to react to the hazardous road conditions. Fig. 24 shows the phase readings from tag and gyroscope data
from smartphone, when we drive the vehicle in the speed of 15mph passing the bump on the road surface. We
can see that Tago can sense the road surface about 1.6 seconds earlier than using the smartphone to sense the
bump. So, the vehicular system has about 1.6 seconds to react to the hazardous road conditions. Fig. 25 shows the
phase readings from tag and gyroscope data from smartphone, when we drive the vehicle in the speed of 20mph
passing the bump on the road surface. As we can see, Tago can sense the road surface about 1.0 seconds earlier
than using the smartphone to sense the bump. So, the vehicular system has about 1.0 seconds to react to the
hazardous road conditions. The above results indicate that Tago can achieve the goal of safe driving due to its
contact-free sensing.

6.3 Case Study

In this subsection, we measure the accuracy of sensing the bumps on the road surface using Tago, when we drive
the vehicle on the roadways at the urban and residential area respectively.
Method. To see the performance of Tago on roadways, we drive the vehicle at residential community and urban
area. There are several bumps on the road surface. Our Tago is responsible to detect these bumps on the road
surface.
Result. Fig. 26 shows the detected bumps along the driving path at the urban area. Fig. 28 shows the detected
bumps along the driving path at residential community. We also indicate the ground-truth location of the bumps

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:19

0.2
1.4

Phase (radians)
phase
tago
x x
gyro.

randian/s
y
gyro. y 0
1.3 z
gyro. z

-0.2
1.2
0 5 10 15 20 25
Time (s)

Fig. 25. Phase readings from the tag and gyroscope data from the smartphone, when we drive the vehicle in the speed of
20mph passing the bump on the road surface.

(i.e., blue dots) along the driving path. The detected bumps are indicated in the figures with the red stars. As we
can see, Tago can accurately detect the bumps on the road surface. Fig. 27 and Fig. 29 show the phase readings
from the tag and gyroscope data from the smartphone, when we drive the vehicle at urban area shown in Fig. 26
and residential area shown in Fig. 28 respectively. We drive the vehicle in the speed of around 15mph. The road
surface between two bumps is quite smooth in these two areas. So, we can see that both of the phase readings
and gyroscope data are quite stable, when we drive on the smooth road. Furthermore, we have two observations
as follows. First, Tago can accurately detect the bumps on the road surface. Second, Tago can sense the bumps on
the road surface earlier than using the smartphone to sense the bumps, thereby the contact-free road surface
sensing is achieved.

7 DISCUSSION
7.1 Fine-Grained Road Surface Sensing
Our paper aims to detect the road surface conditions. Specifically, the bumps or potholes on the road surface
can be detected based on the backscattered signals reflected off the road surface. For fine-grained road surface
sensing, we expect to differentiate the bumps and potholes on the road surface and sense the geometry of these
bumps/potholes. These fine-grained road surface information can be used for road maintenance analysis. To do
so, the straightforward idea is using some advanced sensors (e.g., camera, Lidar and radar sensors). We need
deploy multiple sensors on the vehicle to obtain the 3D geometry of bump/pothole. Can we use RFID tags to sense
the fine-grained road surface information? To differentiate the pothole and bump on the road surface, we can use
one RFID tag deployed in the front of the vehicle and leverage the machine learning algorithm to do classification
based on the backscattered signals reflected off the road surface. To characterize 3D geometry of bump or pothole
on the road surface, we probably need to deploy multiple tags at the front end of vehicle. But, we cannot make
sure that the backscattered signals from the tag array can capture the different parts of the bump or pothole for
3D geometry sensing. Furthermore, using tag array will introduce extra processing latency, since commodity
passive RFID system interrogates tag with slotted ALOHA protocol. The collision-free tag interrogation protocols
(e.g., parallel decoding) can be used. But, these protocols have downsides (e.g., they can only work with tens of
tags and require the hardware modification of tag or reader), which constrain them to be widely deployed.

7.2 Long Range Sensing with Active Tags and Distributed Beamforming
The passive RFID tag is limited by its communication range. To increase the communication range, we can use
the active tags. But, these active tags require the battery replacement regularly. Recently, we notice that the
tunnelling RFID tag can achieve long range sensing, which is not commercialized. In comparison to these active
tags and customized tags, passive RFID tags are ubiquitous. We can enhance communication range of passive

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:20 • Sun and Srinivasan

2.5 0.5

Phase (radians)

randian/s
0
2 tago
gyro.x -0.5
gyro.y
1.5 gyro.z -1

100 150 200 250 300

time (s)

Fig. 27. Phase readings from the tag and gyroscope data
from the smartphone, when we drive the vehicle in the
urban area shown in Fig. 26.
Fig. 26. Driving vehicle at the urban area. The red line
shows the driving path. The red star indicates the de-
tected bumps using Tago. The blue dot indicates the
ground-truth bumps along the driving path.

1.5
tago
3 gyro.x

Phase (radians)
1
gyro.y

randian/s
gyro.z
2.5 0.5

0
2
-0.5

40 60 80 100 120 140

time (s)

Fig. 29. Phase readings from the tag and gyroscope data
Fig. 28. Driving vehicle at the residential area. The red
from the smartphone, when we drive the vehicle in the
line shows the driving path. The red star indicates the
residential area shown in Fig. 28.
detected bumps using Tago. The blue dot indicates the
ground-truth bumps along the driving path.

RFID tags with distributed beamforming using multiple readers like PushID [37]. To see how PushID can work
for Tago, we believe that there are two design spaces. The straightforward idea of applying PushID to Tago is
deploying multiple readers on the vehicle, and these readers can collaborate with each other to do beamforming
to activate the tag. However, this approach seems bulky due to the size of reader’s directional antenna and the
vehicle. The another way is to deploy one reader and one tag for each vehicle. Then, the tag is activated by the
readers deployed on the adjacent vehicles through opportunistic beamforming. As the proliferation of connected
and automated vehicle (CAV) and edge computing, the readers at the different vehicles can collaborate with each
other through Internet backbone. However, this deployment has one downside. It will introduce collisions across
the readers, which requires the readers deployed at different vehicles to collaborate with each other for collision
resolving. Note that the limited communication range is not inherited from Tago’s design. Our algorithm and
design can be extended to the general low-power backscatter tag.

7.3 Crowdsourcing-Based Road Surface Information Sharing

Due to the proliferation of connected and automated vehicle (CAV), the vehicles collaborate with each other
and connect with the remote cloud. This CAV architecture can further enhance Tago’s design and deployment.
Imagine the road surface information collected by one vehicle using Tago can be shared with other vehicles
through Internet backbone. The other vehicles will obtain the road surface information directly from google map.
Note the current satellite-based navigation system cannot show the road surface information to the drivers. The

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:21

0.6
0.25
0.4
0.5

Phase (radians)
0.2
0.2

Amplitude
Amplitude

0.15 0
0
0.1 -0.2

-0.4
0.05 -0.5
-0.6
0 200 400 600 800 0 200 400 600 800
0
6 9 11 Inventory Round Inventory Round
Time (s)
Distance (m)

Fig. 31. Amplitude readings of Fig. 32. Phase readings of backscat-

Fig. 30. Amplitude of backscattered backscattered signals, when people
tered signals, when people perform the
signals over different tag-reader dis- perform the gesture 10 meters away
gesture 10 meters away from Tago’s
tance. from Tago’s setup. setup.

data processing and analysis can be moved to the cloud to relief the computation at the vehicle. One downside of
this scheme is the communication latency at the Internet backbone, especially for highly mobile vehicles. We hope
this barrier can be tackled, as 5G becomes ubiquitous. Note that as more vehicles employ Tago for road surface
sensing and the traffic becomes crowded on the road, the collisions across the reader from different vehicles may
degrade the performance of road surface sensing. We hope the collision-free schemes for RFID communication
could solve this problem. Moreover, the directivity of Tago will mitigate the inter-reader interference.

7.4 Communication Range and Sensing Range of Passive RFIDs

In commodity passive RFID system, the tag is battery-free, which can be activated by the external reader. So,
the communication range is highly depending on the reader’s capability to illuminate the tag. So, the reader is
usually instrumented with the directional antenna to concentrate the signals for powering up the tag. To fully
power up the tag, the reader transmits the constant continuous wave. In our case, since we attach the tag on the
reader’s antenna, the tag will always be activated. This means that the tag can always communicate with the
reader. Tago relies on the backscattered signals reflected off the road surface to sense the bumps. The sensing
range is highly depending on how far the transmitted RF signals from the reader can arrive and be received at
the reader after being reflected off the road surface and modulated by the tag. Intuitively, the sensing range will
be larger than the communication range, since the tag needs to be activated when the power of impinged RF
signals is larger than a threshold consistently. However, in RFID-based sensing, the channel state information of
backscattered signals are extracted from the EPC packets. In another words, the RF signals reflected off the road
surface need to be modulated by the tag for preamble-based channel estimation to extract the desired channel for
road surface sensing. Therefore, the tag needs to be activated and communicate with the reader, such that we
can extract the phase or amplitude information to do sensing. In this regard, the sensing range is same as the
communication range in RFID-based sensing. We also do the following micro-benchmark experiments to show
the communication and sensing range of Tago.
First, we attach reader’s antennas at the front end of the vehicle. We do not attach the tag to the reader’s
antenna like Tago’s setup. We measure the amplitude of the backscattered signals, when we put tag 6, 9 and 11
meters away from the reader’s antennas in front of the vehicle. Fig. 30 shows the amplitude of backscattered
signals, when the tag is deployed 6, 9 and 11 meters away from the reader’s antennas. We can see that the
amplitude decreases as the tag-antenna distance increases. The reader can still read the tag, even though the tag
is 11 meters away from the reader. This experiment indicates that the RFID system’s communication range can
be more than 10 meters. This indicates that the sensing range of Tago could be larger than 10 meters. Next, we
will showcase the reader can indeed capture the variation 10 meters far away from the Tago’s setup.

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:22 • Sun and Srinivasan

We directly show 10m sensing range of Tago through a case study. We evaluate the sensing range in Tago’s
setup. We attach reader’s antennas at the front end of the vehicle, and the tag is attached on the reader’s antenna.
Then, we ask one person who is 10 meters away from Tago’s setup to perfrom the gesture in front of the vehicle.
Tago will process the backscattered signals. Fig. 31 and Fig. 32 show the amplitude and phase readings over time,
when people perform the gesture 10 meters away from Tago. As we can see, Tago has the sensing range of 10
meters.

8 CONCLUSION
In this paper, we propose a system, Tago that can sense the road surface conditions with commodity passive
RFID system. Instead of deploying the reader or tags on the existing infrastructure (e.g., lamp post and roadways),
we deploy the reader inside of the vehicle and the tag will be attached at the front end of the vehicle to achieve
contact-free sensing. We comprehensively analyse Tago’s settings. Then, we propose the road surface sensing
with the signal cancellation algorithm to achieve the contact-free road surface sensing. The extensive experiments
show the efficiency of Tago on road surface sensing in comparison to the state-of-the-art smartphone-based road
surface sensing.

ACKNOWLEDGMENTS
We are grateful to the anonymous reviewers for their insightful feedback. This work was supported by the
National Science Foundation under Grant 1618520.

REFERENCES
[1] Automatic road analyzer. https://www.wtae.com/article/aran-on-the-prowl-automatic-road-analyzer-gathers-information-to-map-
future-state-road-repairs-1/7462861.
[2] Understanding the Link Between Unsafe Road Conditions and Car Accidents. https://www.makeroadssafe.org/understanding-the-link-
between-unsafe-road-conditions-and-car-accidents/.
[3] Omid Abari, Deepak Vasisht, Dina Katabi, and Anantha Chandrakasan. 2015. Caraoke: An e-toll transponder network for smart cities.
In Proceedings of the 2015 ACM Conference on Special Interest Group on Data Communication. 297–310.
[4] Alien. 2020. ALR-8697-8. https://www.alientechnology.com/products/antennas/alr-8697-8/.
[5] Zhenlin An, Qiongzheng Lin, and Lei Yang. 2018. Cross-frequency communication: Near-field identification of uhf rfids with wifi!. In
Proceedings of the 24th Annual International Conference on Mobile Computing and Networking. 623–638.
[6] atlasRFIDstore. 2020. Commodity passive RFID tags. https://www.atlasrfidstore.com/rfid-tag-sample-pack-uhf-passive/.
[7] atlasRFIDstore. 2020. Commodity Passive RFID tags. https://www.atlasrfidstore.com/rfid-tags/.
[8] Joshua F Ensworth and Matthew S Reynolds. 2015. Every smart phone is a backscatter reader: Modulated backscatter compatibility with
bluetooth 4.0 low energy (ble) devices. In 2015 IEEE international conference on RFID (RFID). IEEE, 78–85.
[9] Jakob Eriksson, Lewis Girod, Bret Hull, Ryan Newton, Samuel Madden, and Hari Balakrishnan. 2008. The pothole patrol: using a mobile
sensor network for road surface monitoring. In Proceedings of the 6th international conference on Mobile systems, applications, and services.
29–39.
[10] Junfeng Guan, Sohrab Madani, Suraj Jog, Saurabh Gupta, and Haitham Hassanieh. 2020. Through Fog High-Resolution Imaging Using
Millimeter Wave Radar. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 11464–11473.
[11] Abhishek Gupta, Shaohan Hu, Weida Zhong, Adel Sadek, Lu Su, and Chunming Qiao. 2020. Road Grade Estimation Using Crowd-Sourced
Smartphone Data. In 2020 19th ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN). IEEE, 313–324.
[12] Unsoo Ha, Junshan Leng, Alaa Khaddaj, and Fadel Adib. 2020. Food and Liquid Sensing in Practical Environments using RFIDs. In 17th
{USENIX} Symposium on Networked Systems Design and Implementation ({NSDI} 20). 1083–1100.
[13] Jens Jauch, Johannes Masino, Tim Staiger, and Frank Gauterin. 2017. Road grade estimation with vehicle-based inertial measurement
unit and orientation filter. IEEE Sensors Journal 18, 2 (2017), 781–789.
[14] Haojian Jin, Zhijian Yang, Swarun Kumar, and Jason I Hong. 2018. Towards wearable everyday body-frame tracking using passive rfids.
Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 1, 4 (2018), 1–23.
[15] Nikos Kargas, Fanis Mavromatis, and Aggelos Bletsas. 2015. Fully-coherent reader with commodity SDR for Gen2 FM0 and computational
RFID. IEEE Wireless Communications Letters 4, 6 (2015), 617–620.

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
 On the Feasibility of Securing Vehicle-Pavement Interaction • 29:23

[16] Keiko Katsuragawa, Ju Wang, Ziyang Shan, Ningshan Ouyang, Omid Abari, and Daniel Vogel. 2019. Tip-Tap: Battery-free Discrete 2D
Fingertip Input. In Proceedings of the 32nd Annual ACM Symposium on User Interface Software and Technology. 1045–1057.
[17] Rushil Khurana and Mayank Goel. 2020. Eyes on the Road: Detecting Phone Usage by Drivers Using On-Device Cameras. In Proceedings
of the 2020 CHI Conference on Human Factors in Computing Systems. 1–11.
[18] Martin Laurenzis, Frank Christnacher, Emmanuel Bacher, Nicolas Metzger, Stéphane Schertzer, and Thomas Scholz. 2011. New approaches
of three-dimensional range-gated imaging in scattering environments. In Electro-Optical Remote Sensing, Photonic Technologies, and
Applications V, Vol. 8186. International Society for Optics and Photonics, 818603.
[19] Jaeyoung Lee, BooHyun Nam, and Mohamed Abdel-Aty. 2015. Effects of pavement surface conditions on traffic crash severity. Journal
of Transportation Engineering 141, 10 (2015), 04015020.
[20] Yunfei Ma, Nicholas Selby, and Fadel Adib. 2017. Minding the billions: Ultra-wideband localization for deployed rfid tags. In Proceedings
of the 23rd Annual International Conference on Mobile Computing and Networking. 248–260.
[21] Prashanth Mohan, Venkata N Padmanabhan, and Ramachandran Ramjee. 2008. Nericell: rich monitoring of road and traffic conditions
using mobile smartphones. In Proceedings of the 6th ACM conference on Embedded network sensor systems. 323–336.
[22] Kun Qian, Shilin Zhu, Xinyu Zhang, and Li Erran Li. 2021. Robust Multimodal Vehicle Detection in Foggy Weather Using Complementary
Lidar and Radar Signals. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 444–453.
[23] Ettus Research. 2020. USRP N210. https://www.ettus.com/product/details/UN210-KIT/.
[24] Guy Satat, Matthew Tancik, and Ramesh Raskar. 2018. Towards photography through realistic fog. In 2018 IEEE International Conference
on Computational Photography (ICCP). IEEE, 1–10.
[25] Fatjon Seraj, Berend Jan van der Zwaag, Arta Dilo, Tamara Luarasi, and Paul Havinga. 2015. RoADS: A road pavement monitoring
system for anomaly detection using smart phones. In Big data analytics in the social and ubiquitous context. Springer, 128–146.
[26] Henry M Stommel, Dennis W Moore, and Dennis W Moore. 1989. An introduction to the Coriolis force. Columbia University Press.
[27] ]sun2020allergie Wei Sun and Kannan Srinivasan. [n. d.]. Allergie: Relative Vehicular Localization with Commodity RFID System. In
2020 IEEE International Conference on RFID (RFID). IEEE, 1–8.
[28] Tesla. 2020. Autonomous vehicle. https://www.tesla.com/.
[29] Saurabh Tiwari, Ravi Bhandari, and Bhaskaran Raman. 2020. Roadcare: a deep-learning based approach to quantifying road surface
quality. In Proceedings of the 3rd ACM SIGCAS Conference on Computing and Sustainable Societies. 231–242.
[30] Jue Wang, Fadel Adib, Ross Knepper, Dina Katabi, and Daniela Rus. 2013. RF-compass: Robot object manipulation using RFIDs. In
Proceedings of the 19th annual international conference on Mobile computing & networking. 3–14.
[31] Ju Wang, Liqiong Chang, Shourya Aggarwal, Omid Abari, and Srinivasan Keshav. 2020. Soil moisture sensing with commodity RFID
systems. In Proceedings of the 18th International Conference on Mobile Systems, Applications, and Services. 273–285.
[32] Jue Wang and Dina Katabi. 2013. Dude, where’s my card? RFID positioning that works with multipath and non-line of sight. In
Proceedings of the ACM SIGCOMM 2013 conference on SIGCOMM. 51–62.
[33] Ju Wang, Jianyan Li, Mohammad Hossein Mazaheri, Keiko Katsuragawa, Daniel Vogel, and Omid Abari. 2020. Sensing finger input
using an RFID transmission line. In Proceedings of the 18th Conference on Embedded Networked Sensor Systems. 531–543.
[34] Jingxian Wang, Chengfeng Pan, Haojian Jin, Vaibhav Singh, Yash Jain, Jason I Hong, Carmel Majidi, and Swarun Kumar. 2019. RFID
Tattoo: A Wireless Platform for Speech Recognition. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
3, 4 (2019), 1–24.
[35] Jue Wang, Deepak Vasisht, and Dina Katabi. 2014. RF-IDraw: virtual touch screen in the air using RF signals. ACM SIGCOMM Computer
Communication Review 44, 4 (2014), 235–246.
[36] Ju Wang, Jie Xiong, Xiaojiang Chen, Hongbo Jiang, Rajesh Krishna Balan, and Dingyi Fang. 2017. TagScan: Simultaneous target imaging
and material identification with commodity RFID devices. In Proceedings of the 23rd Annual International Conference on Mobile Computing
and Networking. 288–300.
[37] Jingxian Wang, Junbo Zhang, Rajarshi Saha, Haojian Jin, and Swarun Kumar. 2019. Pushing the range limits of commercial passive
RFIDs. In 16th {USENIX} Symposium on Networked Systems Design and Implementation ({NSDI} 19). 301–316.
[38] Yinsong Wang, Yajie Zou, Kristian Henrickson, Yinhai Wang, Jinjun Tang, and Byung-Jung Park. 2017. Google Earth elevation data
extraction and accuracy assessment for transportation applications. PloS one 12, 4 (2017), e0175756.
[39] waymo. 2020. Autonomous vehicle. https://waymo.com/.
[40] Teng Wei and Xinyu Zhang. 2016. Gyro in the air: tracking 3D orientation of batteryless internet-of-things. In Proceedings of the 22nd
Annual International Conference on Mobile Computing and Networking. 55–68.
[41] Binbin Xie, Jie Xiong, Xiaojiang Chen, and Dingyi Fang. 2020. Exploring commodity RFID for contactless sub-millimeter vibration
sensing. In Proceedings of the 18th Conference on Embedded Networked Sensor Systems. 15–27.
[42] Lei Yang, Yekui Chen, Xiang-Yang Li, Chaowei Xiao, Mo Li, and Yunhao Liu. 2014. Tagoram: Real-time tracking of mobile RFID tags
to high precision using COTS devices. In Proceedings of the 20th annual international conference on Mobile computing and networking.
237–248.

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.
29:24 • Sun and Srinivasan

[43] Lei Yang, Qiongzheng Lin, Xiangyang Li, Tianci Liu, and Yunhao Liu. 2015. See through walls with COTS RFID system!. In Proceedings
of the 21st Annual International Conference on Mobile Computing and Networking. 487–499.
[44] Panlong Yang, Yuanhao Feng, Jie Xiong, Ziyang Chen, and Xiang-Yang Li. 2020. RF-Ear: Contactless Multi-device Vibration Sensing and
Identification Using COTS RFID. In IEEE INFOCOM 2020-IEEE Conference on Computer Communications. IEEE, 297–306.
[45] Xiaobin Zhang, Liangfei Xu, Jianqiu Li, and Minggao Ouyang. 2013. Real-time estimation of vehicle mass and road grade based on
multi-sensor data fusion. In 2013 IEEE Vehicle Power and Propulsion Conference (VPPC). IEEE, 1–7.
[46] Weida Zhong, Qiuling Suo, Fenglong Ma, Yunfei Hou, Abhishek Gupta, Chunming Qiao, and Lu Su. 2019. A Reliability-Aware Vehicular
Crowdsensing System for Pothole Profiling. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 3, 4
(2019), 1–26.

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 1, Article 29. Publication date: March 2022.