See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341459808

Embedded Vision Based Driving Assistance System for Safe Driving With Road
Accident Detection and Reporting

Article · January 2018

CITATIONS READS

0 391

2 authors, including:

Harish Gollaprolu
Vignan's Lara Institute of Technology and Sciences
22 PUBLICATIONS 143 CITATIONS

SEE PROFILE

All content following this page was uploaded by Harish Gollaprolu on 18 May 2020.

The user has requested enhancement of the downloaded file.

IJSART - Volume 4 Issue 4 – APRIL 2018 ISSN [ONLINE]: 2395-1052

Embedded Vision Based Driving Assistance System for

Safe Driving With Road Accident Detection and
Reporting
Syed Rasheed 1, G.Harish2
1, 2
Dept of Electronics and Communication Engineering
1, 2
Vignan’s Lara Institute of Technology & Science, Vadlamudi, Guntur,
Andhra Pradesh-522213, India.

Abstract- Automated driving will have a great impact on a big gap between what is tested in a standard PC and what
society, creating new possibilities for mobility and reducing finally runs in the embedded platform. Furthermore, there is
road accidents. The presence of vision technologies inside the not a standard hardware and software platform, so various
vehicles is expected to grow as an increase in the automation solutions have been proposed by the industry and the scientific
levels. However, embedding a vision-based driver assistance community, as it is usual on still non-mature markets.
system supposes a major challenge because of the special
features of vision algorithms. Additionally, some outstanding To help reduce the huge casualty, tremendous
challenges are also identified. A learning algorithm (Machine innovations in driver assistance systems are on the horizon,
learning classification algorithms) is used to optimizing including Lane Keeping Systems (LKS) [2] and Lane
solution. This paper mainly concentrated on vision-based Departure Warning Systems (LDWS) [3]. Among these,
driving assistant system using computer vision and post-crash camera-based lane detection [4] has been a core technology
analysis using a black box. which many of the newer driver safety systems adapt. Lane
marker detection recognizes white or yellow markings from
Keywords- Advance driving assistance system (ADAS), Lane the image captured by a camera mounted in the car to provide
departure warning system (LDWS), Forward collision warning lane position for automating driving systems [4].
system (FCWS) and Post Crash Analysis.
II. PROPOSED SYSTEM
I. INTRODUCTION
This project mainly concentrated on a vision-based driving
Using transportation is an everyday practice, a assistant system using computer vision. One of the key
characteristic of the modern world, but the human role is technologies in this project is advanced driver assistance
basically taken for granted. People are using different systems (ADAS). Safety features are designed for avoiding
transportation modes such as cars, buses, trains, ships or collisions and accidents by using driving assistant technology
aircraft. From all these modes of transportation, road traffic that alerts the driver of potential dangers or by implementing
injuries are consistently one of the three causes of death for safeguards
people aged between 5 to 44 years old. More than 1.2 million
people died on the world's roads every year, and as many as The proposed system mainly consists of three parts as follows.
50 million others are injured [1].
A. The vision-based driving assistance system
The major role of computer vision in understanding B. Road accident detection and reporting system using
and analyzing the driving scene is of great importance to IoT
building more intelligent driver assistance systems. However, C. Post-crash analysis using the Black box
the implementation of these computer vision-based
applications in a real automotive environment is not A. The vision-based driving assistance system
straightforward. The vast majority of works of the scientific
literature test their driver assistance algorithms on standard An embedded vision system for driving assistance
PCs. When these algorithms are ported to an embedded need to be fulfilling trade-off between several requirements
device, they see their performance degraded and sometimes such as dependability, real-time performance, low cost, small
they cannot even be implemented. Since there are several size, low power consumption by considered all these factors
requirements and constraints to be taken into account, there is this system is designed. This driving assistance system mainly
Page | 2271 www.ijsart.com
IJSART - Volume 4 Issue 4 – APRIL 2018 ISSN [ONLINE]: 2395-1052

concentrated on three parts: 1. Lane Departure Warning Identifying the object is itself is a big task a learning-
Systems (LDWS); 2. Forward collision warning system based algorithm is used to identify objects. Support Vector
(FCWS); 3.Traffic signs recognition. Entire Vision based Machine (SVM) is used to classify objects based on obtained
driving assistance system access the data from the camera and features from the camera. Lighting condition and weather
performing basic image operation by using Open CV (an open condition have a greater impact in obtaining futures. This will
source computer vision library). Lane detection is basically to be corrected by applying filter maintain different threshold
identify the margin of the road by using edge detection. levels. Distance is estimated based on size and shape of the
Coming to FCWS objected are identified and opposite vehicle vehicle and there relative pixel values, FCWS flow chart given
distance is calculated. Object identification and traffic signs in Fig 2 in this assistance to driver performed if the distance is
recognition are performed by using machine learning. less than 10 meters and generate an alarm signal by the buzzer
at the same time a audio “you are nearby a vehicle”.
1. LDWS
3. Traffic signs recognition
At the time of ignition of the vehicle, camera
accessed and edge detection and filters are performed on each Indian Traffic signs are trained using neural networks
frame of video. We considered region of interest as trapezoid. algorithms. Each and every traffic sign images are taken in
During processing of frames huge noise is produced and different light conditions and images are trained based on their
unnecessary lines are produced several considerations are labels. There are several proposing steps are considered like
taken to eliminate them and detect straight lines. image colour segmentation. Attain reason of interest and
classifying using multiple neural networks after these
(1)
proposing steps intimate driver about traffic sing as explained
in Fig. 2 and respect audio generated for respected sing to
Where c is the curvature of the lane, m is slop, b is
intimate.
offset of the lane, x and y are coordinate values
B. Road accident detection and reporting system using IOT
Based on slop value a condition is performed if slop
is more then 80 or less than -80 intimate vehicle driver by
Road accident is detected by impact switches [7], this
generating audio “you are crossing lane” from the speaker.
are placed in the front, back, both sides of the car. During
accident huge pressure occurred on impact switches then it
said to be accident occurred according to proposed system in
such a case immediately the location of the accident and
vehicle number are sent to concern authority likes ambulance,
police etc., by using IOT simultaneously a continuous RF
signal is transmitted after the accident occurred this will help
full to identify the vehicle.

Fig. 1 Flow chart for lane detection

2. FCWS

Page | 2272 www.ijsart.com

IJSART - Volume 4 Issue 4 – APRIL 2018 ISSN [ONLINE]: 2395-1052

Fig. 2 Flow chart for FCWS and traffic sign recognition

C. Post-crash analysis using Black Box

The Post-crash analysis main aim is to identify, how

the crash occurs? It collect data from the sensor will be tack
care of Arduino to reduce the burden to raspberry pi because Fig. 3 Flow chart for road accident detection and reporting
Vision based driving assistance system itself required a huge
amount of computation power. It collects data from humidity, III. HARDWARE
temperature, gyro and accelerometer sensors and transformed
to raspberry pi through the serial port. Change in Gyro values Hardware architecture consists of Raspberry Pi which
means change rotation motion similarly change in is the main part of entire hardware architecture used for
accelerometer means a change in axial motion. When the performing driving assistance. We consider sensors such as
accident occurred then sensor data sent to concern mail-id temperature, humidity, accelerometer and gyro sensors. These
using IOT [6]. Fig.12 is line chart for Black Box information sensors are connected to the arduino. Here, accelerometer and
this will help to identify crash analysis [8]. gyro sensor data used to identify a change in orientation and
rotation based placemat of the vehicle. Temperature and
humidity sensor is used to identify internal vehicle
temperature conditions.SD card contains OS and file system.
It is inserted in the SD card slot. Impact switches nothing but
vibrating sensor. If impact switch is activated then road
accident is detected. The camera is directly connected to the
raspberry pi. Speaker and buzzer are used to intimate vehicle
driver. Speaker is directly connected to 3.5mm audio Jake.
Raspberry Pi HDMI port connected to display. GPS and
arduino both are connected to the serial port.

Page | 2273 www.ijsart.com

IJSART - Volume 4 Issue 4 – APRIL 2018 ISSN [ONLINE]: 2395-1052

average in FCWS and Traffic Sign Detection. Design,

development and validation we follow a V-model [4]. Testing
was done on two aspects testing using test dataset and testing
on public roads. Fig. 6, 7 and 8 are related to vision-based
driving assistance system testing on highway roads.Fig.9 SMS
intimation to concern authority in case accident occurred.
Table I is Black box information sent to concern mail id after
the accident occurred it contain time, date, humidity,
temperature, accelerometer (Ax, Ay and Az) and gyro meter
(Gx, Gy and Gz) sensor values. Fig 10 is related to line chart
for this sensor values.
Fig. 4 Hardware architecture

IV.SOFTWARE ARCHITECTURE

The software architecture itself explanative to

understand. How entire software organized? Object detection
and traffic sign are trained on hardware by using Support
Vector Machine (SVM). Training, validation and testing is
done on image data set which contains Indian traffic sign and
vehicle images. Coming to operating system standalone not
necessary due to memory constraints and computational
ishues. If background applications of os are ruing then the
performance of vision algorithms are reduced due to them by Fig. 6 Lane departure warning system
considering all this a basic kernel with required support to
application of OS is enough due to this constraint.

Fig. 7 Forward collision warning system

Fig. 5 Software architecture

V. RESULTS AND DISCUSSION

There is no ideal platform for implementing vision-

based algorithms for embedded applications, it varies with
graphics processing. Some of the most commonly used
embedded platforms are not much faster as real-time for the Fig. 8 Traffic signs recognition
vision system. We required an efficient platform to built high
computation performance. Raspberry is also not much
sufficient to perform these operations but several
considerations are taken to obtain higher efficiency such as
dynamic threshold and splitting entire image training data set Fig. 9 SMS intimation to concern authority
into Batches and trained one by one. The proposed system can
achieve 89.34% detection rate and 3.4% of false alarm rate on
Page | 2274 www.ijsart.com
IJSART - Volume 4 Issue 4 – APRIL 2018 ISSN [ONLINE]: 2395-1052

TABLE I between the three levels of processing of the implemented

Black box file vision algorithm (low level, mid-level and high level), is of
big importance when designing both the hardware architecture
and the finally optimized software. It is also remarkable that it
is not possible to fulfill completely all the requirements, so
there must be a trade-off between the several design requisites.

In this paper, an ADAS consisting of FCWS and

LDWS functions is proposed and implemented on a Free scale
with a monocular camera, which provides safety information
to drivers. The proposed dynamic threshold method conquers
the problems resulted from different weather conditions, and
the proposed multiple frame approval reduces the effect of
different scenes and windshield wiper. Two algorithms,
LDWS and FCWS, are integrated together in order to produce
vision assistance for day and night highway driving, which
detects not only the forward vehicle but also the potential cut-
in vehicles. Automatic visual inspection systems can assist the
human inspector but cannot replace or duplicate many of the
unique abilities which the human being has. In this paper, the
principle benefits of visual inspection systems help to identify
lane, traffic signs, and object classification. This project is
mainly to improve the feature of middle range cars and safety
of passengers.

REFERENCES

[1] World Health Organization: ‘Global status report on road

safety 2015’, 2015.
[2] V. Cerone, M. Milanese and D. Regruto. Combined
Automatic Lane-Keeping and Driver's Steering Through a
2-DOF Control Strategy. IEEE Transactions on Control
Systems Technology, 17(1): 135-142, 2009.
[3] J.M. Clanton, D.M. Bevly and A.S. Hodel. A Low-Cost
Solution for an Integrated Multisensory Lane Departure
Warning System. IEEE Transactions on Intelligent
Transportation Systems, 10(1): 47- 59, 2009.
[4] Gorka Velez , Oihana Otaegui. Embedding vision-based
advanced driver assistance systems: a survey. IET Intell.
Transp. Syst., 2017, Vol. 11 Iss. 3, pp. 103-112.
[5] Arun Nandewal, Abhishek Tripathi, Satyam Chandrra.
Indian Traffic Sign Detection and Classification Using
Neural Networks. An International Journal of Engineering
Fig. 10 Line chart for black box information
Sciences, Special Issue ICTMS-16 , 25-26 JUNE 2016.
[6] C.Mohamedaslam,Ajmal Roshan.T, Mohamed
VI. CONCLUSION
Sahal.M.T, Nisi.K. A Smart Vehicle For Accident
Prevention Using Wireless Blackbox And Eyeblink
This paper provided an insight into vision-based
Sensing Technology Along With Seat Belt Controlled
embedded ADAS. The key elements of this field were
Ignition System. 2016 Online International Conference on
explained, starting from the features and requirements that
Green Engineering and Technologies (IC-GET).
these systems have. These features and requirements guide all
[7] Sayem Chaklader, Junaed Alam, Monirul Islam, Ali
decisions taken about hardware and software. The balance
Shihab Sabbir. Black Box: An Emergency Rescue

Page | 2275 www.ijsart.com

IJSART - Volume 4 Issue 4 – APRIL 2018 ISSN [ONLINE]: 2395-1052

Dispatch System for Road Vehicles for Instant

Notification of Road Accidents and Post Crash Analysis.
3rd International Conference On Informatics, Electronics
& Vision 2014.
[8] N. Watthanawisuth, T. Lomas and A.Tuantranont.
Wireless Black Box Using MEMS Accelerometer and
GPS Tracking for Accidental Monitoring of Vehicles. the
IEEE-EMBS International Conference on Biomedical and
Health Informatics (BHI 2012) Hong Kong and
Shenzhen, China, 2-7 Jan 2012.

Page | 2276 www.ijsart.com

View publication stats