Next-Generation Traffic Control: Adaptive Timer

and Emergency Vehicle Priority in Intelligent

Traffic Management
2024 International Conference on E-mobility, Power Control and Smart Systems (ICEMPS) | 979-8-3503-9439-9/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICEMPS60684.2024.10559373

Devika S G Govind A Lekshmi D

Dept. of Computer Science Engineering Dept. of Computer Science Engineering Dept. of Computer Science Engineering
SCT College of Engineering SCT College of Engineering SCT College of Engineering
Trivandrum, India Trivandrum, India Trivandrum, India
ORCID: 0009-0000-1056-6364 ORCID: 0009-0002-2573-4003 ORCID: 0009-0006-7911-2743

Abstract—Traffic has always been a troubling issue in our Intelligent traffic control systems are crucial for address-
society. The increased number of automobiles on roads has sub- ing these issues and require real-time detection and efficient
stantially increased traffic congestion polluting the environment response mechanisms. Applying automation and intelligent
and causing stress and anxiety to passengers in the traffic clog.
The existing system of fixed traffic control system even though control to roadside infrastructure and vehicles can enhance
widely used, lacks adaptability and operates on predefined signal traffic flow and safety in the existing transportation system.
timers. The lack of adaptability can lead to inefficient traffic flow, According to the survey conducted approximately 95% of
especially during high traffic congestion. To address this issue, respondents reported being impacted by traffic congestion in
we propose a more adaptive and intelligent system that leverages the study area. About 89% opined that it increases their anger,
adaptive timer control and prioritizes emergency vehicles like,
ambulance. The adaptive timer control system considers the 98% said that it is time-consuming, and 88% perceived that
count of vehicles at the previous junction as well as the count of it has an influence on the decline in their income. About
vehicles currently arriving at the intersection to update the signal 54% of the respondents claimed that traffic was a cause
timers. Additionally, to prevent delays in emergency response of frustration. In comparison, 95% of respondents felt that
times and to provide a better impact on public safety and regular traffic congestion during peak and off-peak hours
health outcomes we have provided an added feature to prioritize
emergency vehicles and release the traffic upon its arrival at causes mental stress [1].
the respective signal. The detection of the emergency vehicle is Drawbacks of existing System are:
to be carried out using the YOLO v8(You Look Only Once) 1) Excessive and avoidable waiting periods occur on roads
algorithm along with siren recognition using a classifier model. when traffic densities fluctuate: The inability of fixed count-
The proposed system offers a significant improvement over the down timers to adapt to real-time traffic conditions results in
existing traffic control infrastructures and has the potential for
aiding resilient and smooth transportation. significant delays and congestion, especially during peak hours
Index Terms—AI Traffic control, Traffic management, Emer- or sudden traffic surges. With India’s increasing automobile
gency, Adaptive traffic system, Intelligent transport systems, Smart population and limited road capacity, the rigid nature of fixed
surveillance, Computer Vision, Machine Learning, YOLO v8 timers exacerbates the problem by failing to optimize traffic
flow efficiently. The lack of dynamic adjustment in countdown
I. I NTRODUCTION timers hampers the smooth movement of vehicles and can
The existing traffic systems in urban areas face numerous contribute to increased fuel consumption and environmental
challenges due to the increasing number of vehicles and the pollution. This can be tackled by placing a system that sets
limited capacity of road networks. Fixed signal timers used timer based on car count.
in traffic control systems at intersections lead to inefficiencies 2) Lack of adequate measures to ensure clearance for
and congestion. The demand for road capacity necessitates emergency vehicles delays their prompt response.: Delayed
new solutions for traffic control that can adapt to changing response times caused by traffic congestion can have severe
conditions. The efficiency of transportation systems directly consequences, as ambulances, fire engines, and police vehicles
has a socio-economic impact. Real-time traffic information struggle to reach their destinations quickly. These delays not
systems and intelligent control methods can affect transporta- only impact the effectiveness of emergency services but also
tion costs and time. Traffic congestion is considered one of pose a risk to public safety and can result in potential loss of
the major barriers to sustainable urban development because it life and property. Implementing effective traffic management
increases travelling time, energy consumption, road collisions, strategies that prioritize the clearance and smooth movement of
and environmental pollution. Its principal manifestation is emergency vehicles is crucial for ensuring timely and efficient
the gradual slowing of traffic, increasing travel times, fuel emergency response services. The number of accidents per
consumption, pollution, and other costs [1]. day is increasing in today’s scenario. As reported by Times

Authorized licensed use limited to: University of Plymouth. Downloaded on September 20,2024 at 10:54:21 UTC from IEEE Xplore. Restrictions apply.
of India in the year 2016, approximately 146,133 people lost within a VANET system. Roadside Units (RSUs) ex-
their lives in road accidents in India. Tragically, around 30% change information and regulate traffic signals. Hacking
of these fatalities are attributed to delayed ambulance services. incidents to be handled by shifting traffic to uncompro-
This issue can be side-lined to a huge extent by prioritizing mised RSUs, identifying and blocking malicious sources.
emergency vehicles over others at the signals. • Reference [7] proposes a system that acquires live move-
3) The lack of consideration for traffic conditions at pre- ment of vehicles. The camera shifts to the next lane
vious traffic signals leads to inefficient flow.: Ignoring traffic based on time intervals. Vehicle detection and recognition
conditions at previous traffic signals can result in inefficient is done using the Haar Cascade algorithm. The Video
flow, leading to an irregular movement of vehicles and po- Processing module present in the system calculates the
tential congestion on the roads. Sudden traffic congestion is traffic density (the count) and along with time assigned
often a consequence of the domino effect caused by inadequate for each vehicle, the signal lights are switched.
coordination between traffic signals, where poor flow manage- III. PROPOSED SYSTEM
ment at one signal can impact the entire traffic network. By
utilizing the traffic count at previous signals to predict traffic Our System aims to develop and deploy a system that
at a given signal, we can improve traffic flow management enables easy flow of vehicles at traffic signals using modern
and reduce congestion on the roads. This approach allows for deep learning algorithms along with conventional cameras and
better coordination between signals and helps create a more sensors. Rules and regulations were followed in the proposed
efficient and synchronized traffic flow throughout the network. system as per the rules [13] by the government of India.

II. LITERATURE REVIEW

• Reference [2] proposes a structure that integrates Central-
ized control system, a mobile application for ambulance
drivers and a web portal for administrators. The Cen-
tralized control system facilitates real time monitoring of Fig. 1. Proposed system flowchart
the ambulance coordinates. The GPS helps navigate to
the spot, update their status and availability through the The proposed system consists of the following:
mobile application. The web portal provides administra- 1) Vehicle detection and Counting Module: The proposed
tors with data to view the performance and analyse the system leverages the YOLO (You Only Look Once) algorithm
data on a continuous basis. for accurate and efficient vehicle detection. YOLO v8’s fast
• Reference [3] proposes a strategy that adjusts the timing processing time and high accuracy makes it a suitable choice
of traffic lights based on real-time traffic conditions. Uti- for this purpose. To enhance vehicle detection capabilities, a
lizes intelligent algorithms and sensors to collect data on custom YOLO v8 model was fine-tuned specifically for identi-
patterns, density, and traffic volume which is analysed to fying various vehicle classes such as cars, bikes, trucks, buses
determine the optimal timing for traffic lights at different and emergency vehicles. The advantages of using YOLO v8
intersections. for vehicle detection include real-time performance, enabling
• Reference [4] proposes a framework that incorporates the system to process a continuous stream of video received
modules such as object detection module and signal from the camera and instantly identify vehicles. Additionally,
switching module. YOLO algorithm detects and classifies YOLO’s ability to detect multiple objects in a single pass
vehicles into different categories and the Signal Switching improves efficiency and reduces processing time compared to
Algorithm sets the green signal timer as per the traffic traditional detection methods. This allows the proposed system
density and in a cyclical manner the signals are switched. to provide timely and reliable vehicle detection for effective
• Reference [5] puts forward a framework for integrating traffic control and management. The dataset to train the model
intelligent transportation systems and ensuring effective was obtained from two different sources. The labelled dataset
communication between different components of a trans- was obtained from VehicleID (PKU VehicleID) . The dataset
portation network. The Traffic-Roadway Section provides contained images of 26,267 vehicles . Then the model was
design for networking of traffic signals and the Intelligent trained using the pre- trained weights downloaded from the
Transportation Systems Unit (ITS Unit) is responsible for YOLO website. The trained model of the output was stored
determining the type of communication system that can according to Darknet format. The ”.cfg” file holds the network
be used, based on the project type. architecture configuration and hyperparameters of the YOLO
• Reference [6] suggests a strategy to prioritize emergency v8 model, defining its structure. The “. weights” file contains
vehicles using Vehicle Ad-Hoc Networks (VANET) and the learned weights and parameters of the trained YOLOv8
the Internet of Things (IoT). It measures the distance model, representing the knowledge gained during the training
between an intersection and the emergency vehicle, con- process. The confidence level was set to 80% in order for more
siders factors such as incident type and emergency vehicle successful detection. OpenCV module was used to capture
type. It includes a Traffic Management Server (TMS), the video from camera and passed to identify the presence
vehicles, Roadside Units (RSUs), and sensors deployed of vehicles and also count the number of vehicles.

Authorized licensed use limited to: University of Plymouth. Downloaded on September 20,2024 at 10:54:21 UTC from IEEE Xplore. Restrictions apply.
 tion, turns, other vehicle carelessness, ambulance operator
carelessness, road conditions, and traffic regulation violations.
The study showed that the average traffic jam caused an
ambulance delay of 10 minutes and a dense traffic condition
caused an ambulance delay of 12.98 minutes longer than rare
traffic conditions [9]. The proposed system features emergency
vehicle detection and siren noise detection. recognition of
emergency vehicles results in an emergency system giving true
value and false in all other conditions.

Fig. 2. YOLOv8 Model was fine-tuned with 160 epochs on image size 512.

The survey highlighted the prevalence of RCNN (Recur-

rent Convolutional Neural Network) and YOLO architectures
in existing methods, suggesting that future research should
focus on developing a vehicle detection method based on
the YOLOv8 architecture. However, the survey also mentions
the need for improvements in tracking methods as existing
approaches heavily rely on manual intervention and camera
views. Moreover, the similarities between different classes of
vehicles and the small size of vehicles in satellite images
pose challenges that require fine-tuning of network parameters
for desired results. Overall, the advantages of using YOLO
for vehicle detection include real-time performance, efficient
object detection, and the ability to identify multiple vehicle
classes. However, addressing limitations in tracking methods
and fine-tuning network parameters are areas that need further
attention and improvement [8].

Fig. 4. Shows the proposed Emergency Vehicle Identification Module

• Vehicle Recognition: YOLO v8 model used for vehicle

identification is also used here. It can detect and localize
emergency vehicles with high precision, minimizing false
positives and negatives. There are 3 classes of Emergency
vehicles: ambulances, fire engine and police. The sepa-
rately labeled dataset was obtained from images.cv and
Fig. 3. Showcases test images that were subjected to our vehicle detection Roboflow. YOLO performs object detection by dividing
model. On the top, the original image is displayed, while on the bottom, the the input image into a grid and predicting bounding
output is presented, demonstrating the application of the vehicle detection
model. The output includes bounding boxes around the detected vehicles and
boxes and class probabilities for each grid cell. This
their confidence score and label. enables the model to detect multiple objects in a single
pass, including multiple classes of emergency vehicles
2) Emergency Vehicle Identification Module: There were within the same image, without the need for additional
6 factors that mostly influence ambulance delays: intersec- computational overhead.

Authorized licensed use limited to: University of Plymouth. Downloaded on September 20,2024 at 10:54:21 UTC from IEEE Xplore. Restrictions apply.
 resizing the images to a consistent size (e.g., 224x224
pixels). The VGG (Visual Geometry Group) architecture
of CNN was used as it consists of multiple convolutional
layers with small 3x3 filters, followed by max-pooling
layers for downsampling. VGG16, which has 16 layers,
was used here. After monitoring performance of the
model , epoch value was set to 50. Librosa library in
python was used to input the live noise.

Fig. 6. A sound wave graph of a sample siren sound is shown, which was
used in VGG16.

3) Signal Communication Network Module: To tackle the

Fig. 5. The left figure shows the original image, while the image on the right
side shows correctly identified emergency vehicles in boxes with class labels. flow of irregular distribution of traffic in a city, we consider
Three classes of vehicle are identified: police, ambulance and fire engine. the face of past traffic signal count also while setting up
the countdown timer. So, it’s important to maintain a proper
communication network among signals. The reference [5] puts
• Siren Noise Detection: As emergency vehicles may not forward the method of using optic fibres for easy communi-
always be on an emergency run, it is important to identify cation between signals as it is found to have the maximum
whether there is an emergency. So, the conventional efficient communication. The paper also puts forward the use
method is by the identification of a siren alarm that of 12 or 24 strand fibre to enhance the traffic communication.
goes loud when activated hence it’s important to have
a sound recognition system. The reference [10] presents
a study on detecting emergency vehicles, such as ambu-
lances, fire engines, and police cars, based on their siren
sounds. To address the issue of drivers potentially missing
siren warnings, particularly when using in-vehicle audio
systems, an automatic detection system called SirenNet
is proposed. SirenNet is an ensemble model utilizing
a CNN (Convolutional Neural Network) with two net-
work streams. The first stream, WaveNet, processes raw
waveform data, while the second stream, MLNet, works
with a combined feature consisting of MFCC (Mel-
frequency cepstral coefficients)and log-mel spectrograms.
Experimental results on a diverse dataset demonstrate that
combining raw data with MFCC and log-mel features
achieves a promising accuracy of 98.24% for siren sound
detection. Notably, the system performs well even with
short samples of 0.25 seconds, achieving an accuracy
of 96.89%. The proposed system has potential benefits Fig. 7. The proposed signal communication optic fiber is placed in a locality
in Trivandrum to demonstrate communication between 4 road junctions and
for both drivers and autopilot systems. Based on this 3 road junction (T intersection) signals. Here dotted lines denote the optic
we created a CNN for image recognition of sound wave cable running between the signals.
images. The sound dataset was obtained from audioset
tool inside research.google.com. Another dataset was also 4) Mathematical Weightage: According to Reference [11]
obtained to train for background noises which mostly in statistics the assumed mean is a method for calculating the
contain horn sounds. The sound wave graphs of the arithmetic mean and standard deviation of a data set which is
dataset was generated using sonicvisualiser and was used used to get the batch size. Batch size refers to the number of
as training dataset for the model labeled as “Siren” vehicles we expect to wait at a signal. It is assumed as the
and “BackgroundNoise”. The preprocessing step includes average number of vehicles we expect to wait.

Authorized licensed use limited to: University of Plymouth. Downloaded on September 20,2024 at 10:54:21 UTC from IEEE Xplore. Restrictions apply.
 a = assumed mean, i.e., current batch size
X
f i = traffic frequency
di = xi − a
= deviation of ith class of traffic count
Fig. 8. T-Intersection simulation output
from current batch size
X
f i = n = Total number of observations. Reference [13] puts forward that, 4 Road intersections are
a crucial part of the system and frequently act as a bottleneck
The time calculation formula can be broken down step-by- for traffic. The most recent data indicates both site limits
step: and increase in traffic. The engineer must create criteria and
procedures based on the fundamental relationships between
capacities, traffic flow, geometry design, and safety. This
allows users to move directly from suggested geometry to
In the scaled weighted average calculation, the current value actual estimates of the operating condition. The findings of
is multiplied by 0.8 and the previous count value is multiplied this study provide an insight into recommended solutions for
by 0.2. This assigns different weights to control their influence. these issues as well as safety measures for reducing traffic
The time calculation involves scaling the weighted average by jams at intersections. We used these guidelines to design 4
dividing it by batch size. Then, the scaled weighted average is Road intersections for our simulation.
multiplied by the maximum allotted time to determine the time
percentage to be allocated. Subtracting the time percentage
from maximum time gives the counter time.

This creates a countdown effect as the weighted average

increases. Lastly, rounding the calculated time to the nearest
integer ensures that the time value is represented in whole
minutes, which is typically used for time intervals. These
mathematical operations are used to compute the time for Fig. 9. 4-road intersection simulation output
each red timer based on the weighted average of current and
past values, ensuring that the allocated time corresponds to the The 4-road intersection has been demonstrated in the sim-
significance of the weighted average and creates a countdown ulation as shown in figure 9. The previous junction count and
effect. the current vehicle count is displayed along with the timers.
5) System Simulation Module: The proposed traffic system To simplify the simulation, the previous count is generated
was analysed and observed using a Python simulation created randomly while the current vehicle count is calculated from the
for both 4-road intersections and 3-road intersections. Refer- screen. We can see the signal timers and the traffic congestion
ence [12] demonstrates the detailed study of T-intersections in as time passes. An emergency flag is also set to detect whether
Andhra Pradesh and clearly analyses how the lack of proper an ambulance arrives or not.Hence based on the current count
traffic signals contributes to huge number of accidents. The of vehicles and the previous junction count the signal timers
study of road traffic in the city shows that the primary cause should be set accordingly which act as parameters.
of accident is collision of vehicles at the intersections. The
collision may be rear shunt on approach to the junction, right- IV. PERFORMANCE EVALUATION
angled collision, principal right turn collision or pedestrian The performance of the proposed system is analyzed based
collision. These collisions can be eliminated if the signal is on the simulation created. It was observed that every time
properly designed. So, the main objective of the dissertation an emergency vehicle enters the signal, the light immediately
is to provide better and safe mobility of traffic through efficient turns green while the rest of the signals at the intersection
signal design at the intersection. The signal is designed as per remained red. In all other cases the minimum red signal timer
IRC guidelines so that the signal could ensure proper traffic is set as 15 seconds as it’s considered as the time required for
flow. The T-intersection simulation has been built following pedestrian crossing and maximum red signal timer is set to 60
the IRC guidelines. seconds is scheduled preventing over-lodging of vehicles. The

Authorized licensed use limited to: University of Plymouth. Downloaded on September 20,2024 at 10:54:21 UTC from IEEE Xplore. Restrictions apply.
maximum time 60 seconds is set whenever the current count [5] Oregon Department of Transportation Traffic Standards and Asset Man-
of vehicle is 0 as it can be observed from the Table I given agement Unit “Chapter 7 – Interconnect ITS Communication Plan”,
June 2017.
below. [6] Sumi, Lucy, and Virender Ranga. ”Intelligent traffic management system
for prioritizing emergency vehicles in a smart city.” International Journal
of Engineering 31.2 (2018): 278-283.
[7] Sable, Tanvi, et al. ”Density and time based traffic control system using
video processing.” ITM Web of Conferences. Vol. 32. EDP Sciences,
2020.
[8] Maity, Madhusri, Sriparna Banerjee, and Sheli Sinha Chaudhuri. ”Faster
r-cnn and yolo based vehicle detection: A survey.” 2021 5th international
conference on computing methodologies and communication (ICCMC).
IEEE, 2021.
[9] Wiwekananda, Ketut Shri Satya, et al. ”Understanding factors of ambu-
lance delay and crash to enhance ambulance efficiency: an integrative
literature review.” J Comm Emp Health 3 (2020): 213.
[10] Tran, Van-Thuan, and Wei-Ho Tsai. ”Acoustic-based emergency vehicle
detection using convolutional neural networks.” IEEE Access 8 (2020):
75702-75713.
[11] Drusch, Robert L. ”Estimating annual average daily traffic from short-
term traffic counts.” Highway Research Record 118 (1966): 85-95.
[12] Surisetty, Ramesh, and Soma N. Sekhar. ”Designing of a Traffic Sig-
naling System at T-Intersection.” International Journal of Engineering
Fig. 10. Graph showing comparison between existing 30-second timer as per Research and Application 7.4 (2017): 82-86.
[15] and the proposed system [13] Dr. A. MAHESH BABU, J. Jatin Karthik, M. Surya Kiran and Syed
Shabbir “Design and Analysis of Traffic Signal at Signalized Junction”,
Journal of Engineering Sciences, Vol 13 Issue 7, July/2022, ISSN:0377-
V. CONCLUSION AND FUTURE WORKS 9254
[14] Ministry of road transport highways, Government of India “Traffic Signs
The sole purpose of this system is to integrate adaptive ensure safety: Book on road safety signage and signs”, Fifth Edition
systems for emergency vehicles along with the implementation :2015
of intelligent traffic systems. In this system, there is a com- [15] US Department of transportation Federal Highway Administration ”Traf-
fic Signal Timing Manual”, CHAPTER 5, Signal Timing Manual -
munication channel established between signals which helps to Second Edition
pass on traffic count at previous signals and also uses traffic
count at the present signal to establish an adjusted vehicle
count and use that to determine the waiting time set for the
red signal timer. The emergency vehicle that is in emergency
mode could bypass as the signal turns green for the vehicle.
The vehicle counts and emergency vehicle detection part takes
place using YOLO v8 while sound recognition happens as
VGG16, a CNN variant recognizes the sonic wave using
imaging. Thus, the proposed system offers a significant im-
provement over the existing traffic control infrastructures and
has the potential for more resilient and smooth transportation.
The domain of adaptive traffic light control systems could be
expanded into another dimension by incorporating the concept
of IoT (Internet of Things) to enable live communication with
emergency vehicles. Moreover, disaster response systems like
water level gauges or seismic sensors could be used such that
when calamities occur, it alerts the traffic signal to show red
automatically to prevent vehicles from unknowingly reaching
those destinations.
R EFERENCES
[1] Fattah, Md Abdul, Syed Riad Morshed, and Abdulla-Al Kafy. ”Insights
into the socio-economic impacts of traffic congestion in the port and
industrial areas of Chittagong city, Bangladesh.” Transportation Engi-
neering 9 (2022): 100122.
[2] George, Anita Acha, et al. ”Golden aid an emergency ambulance
system.” 2017 International Conference on Networks Advances in
Computational Technologies (NetACT). IEEE, 2017.
[3] Shinde, Swapnil Manohar. ”Adaptive traffic light control system.” 2017
1st international conference on intelligent systems and information
management (ICISIM). IEEE, 2017.
[4] Gandhi, Mihir M., et al. ”Smart control of traffic light using artificial
intelligence.” 2020 5th IEEE international conference on recent advances
and innovations in engineering (ICRAIE). IEEE, 2020.

Authorized licensed use limited to: University of Plymouth. Downloaded on September 20,2024 at 10:54:21 UTC from IEEE Xplore. Restrictions apply.