International Journal of Advances in Engineering and Management (IJAEM)

Volume 3, Issue 8 Aug 2021, pp: 1794-1800 www.ijaem.net ISSN: 2395-5252

Real Time Vehicle Detection, Tracking and

Counting Using Raspberry – Pi
Gopika R, Mythri R, Rashmi Baluragi, Sushma G,
Dr. Ambedkar Institute of Technology, Bangalore
Mythri R, Dr. Ambedkar Institute of Technology, Bangalore
Dr. Ambedkar Institute of Technology, Bangalore
Dr. Ambedkar Institute of Technology, Bangalore.
---------------------------------------------------------------------------------------------------------------------------------------
Submitted: 10-08-2021 Revised: 25-08-2021 Accepted: 28-08-2021
---------------------------------------------------------------------------------------------------------------------------------------

ABSTRACT:Population explosion leads to increase monitored area. The need of traffic surveillance
in the number of physical objects means number of system is to provide construction engineers and
vehicles on road. This result, the number of road other associates to plan in an economical way and
accidents increases due to a very heavy traffic flow. proper decisions were taken based on density of the
Here traffic flow is monitored by using computer vehicles and the statistics obtained by the low cost
vision paradigm, where images provide a betterment electronic devices. It also provides solution to major
on the road view. In order to detect vehicles, problems such as vehicle accidents, vehicles theft
monitor and estimate traffic flow using low cost detection, managing parking area and other security
electronic devices, this research work utilizes threats.
camera module of raspberry-pi along with The main cause for the interest in traffic
Raspberry Pi 3. It aims to develop a remote access management activity is to utilize the computer
using raspberry-pi to detect, track and count vision techniques in real real-time conditions.
vehicles only when some variations occur in the The major challenge that caused hindrance
monitored area. The proposed system captures video to our work is vehicle segmentation in various
stream for providing video like vehicles in the atmospheric conditions such as night, snowy or
monitored area to compute the information and dusty wealth conditions.
transfer the compressed video stream for providing As a solution to it we have used a different
video based solution that is mainly implemented in pre-processing unit based on Histogram
Open CV by Python Programming. The proposed Equalization to improve the resolution of video and
method is considered as an economical solution for morphological operation to add or remove pixels in
industries in hich cost-effective solution are the boundaries of objects, where video depends on
developed for traffic management. shape and size of the structuring elements before
KEYWORDS: Computer Vision; Traffic flow; processing towards the next stage.
OpenCV; Vehicle Detection; Vehicle Tracking; It is also observed that the vehicles moving
Vehicle Count. towards same point either in lighter or darker region
or vice-versa might have the same colour as
I. INTRODUCTION background for detecting vehicle remains move
It is observed that, the major hindrance on challenging and this leads to fault in count of
road is due to heavy traffic flow during peak hours vehicles. So, as a solution to it we have employed
especially when people commute to work. The total background subtraction technique to register the
number of vehicles or object exceeds its capacity by vehicle ID if it crosses the given threshold.
causing a blockage for emergency vehicles such ass
fire fighter and rescue vehicles. In order to develop II. PROPOSED SYSTEM
an efficient, reliable, cleaner and safer mode of This System uses a well-known platform
transportation, it is necessary to make the road called Linux server for surveillance and recording
transportation system automated as much as video using Raspberry-Pi.
possible. The approach uses of Raspberry Pi to record a video
The primary focus is dedicated towards the when something moves inside the monitored area.
detection and tracking of vehicles, where it finally The proposed system efficiently manages to
keeps the count of vehicles in the particular distinguish the vehicles from the surrounding

DOI: 10.35629/5252-030817941800 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 1794
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 3, Issue 8 Aug 2021, pp: 1794-1800 www.ijaem.net ISSN: 2395-5252

environmental variability and improves the low Then the implementation of background subtraction
resolution videos through Histogram equalization algorithm helps to detects objects and track them
technique in order to maintain uniformity of videos based on the particle filter algorithm. Tracked
in terms of resolution and also in removal of noise vehicles are counted based on the threshold given to
from videos. different sized vehicles based on the area.

Figure 1: Proposed system design

III. SYSTEMDESIGN

Figure 2: Raspberry-Pi

This paper makes use of as shown in figure 2 It establishes a connection between camera and
network.
1. Raspberry-Pi 3
Raspberry-Pi 3 is 52 RAM with 1.2GHz. Remote access of Raspberry-Pi
2. Raspberry-Pi camera module To operate under Raspberry-Pi, it is necessary to
Raspberry-Pi has a connector to plug in a install all the recent features and drivers for updated
camera module to capture the video. OS with a correct access to an internet connection.
3. Power supply
Raspberry-Pi uses 400mA of current that plugs the Commands to be followed in Raspberry-Pi
micro USB. $ sudo apt-get update
4. Micro SD Card This command updates the OS with recent features
Raspberry-Pi uses SD Card to store and install and drivers.
libraries and run OS for this device. Minimum $ sudo raspi-config
64GB or higher is required. Raspi-config opens the configuration tool that is
5. Wi-Fi USB adaptor written and maintained by Raspbian OS.

DOI: 10.35629/5252-030817941800 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 1795
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 3, Issue 8 Aug 2021, pp: 1794-1800 www.ijaem.net ISSN: 2395-5252

Figure 3: Raspberry-Pi connection

Raspberry -pi connection is shown in figure 3.

Open RaspiCam Remote app provide login details such as IP address, username and password as shown in
figure 4.

Figure 4: Remote login

DOI: 10.35629/5252-030817941800 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 1796
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 3, Issue 8 Aug 2021, pp: 1794-1800 www.ijaem.net ISSN: 2395-5252

DOI: 10.35629/5252-030817941800 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 1797
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 3, Issue 8 Aug 2021, pp: 1794-1800 www.ijaem.net ISSN: 2395-5252

IV. SYSTEM IMPLEMENTION Path followed by an object with the

Proposed system make use of three important purpose to determine the observed direction of
modules namely target on a near real-time surveillance and security
A. Vehicle Detection for traffic control without affecting human computer
B. Vehicle Tracking intervention.
C. Vehicle Counting The major goal of tracking is to determine the target
object in sequential frames of video. Object change
Vehicle Detection in shape and size over time in such scenarios so
Computervision paradigm provides vision motion model for recovering trajectories and models
for identifying objects that belong to classes that with high accuracy for a small number of vehicles.
might be a vehicle or person in a video. The use of Bounded boxes around a detected object are seen.
object detection in computer vision paradigm is to Centroid of bounding boxes determines the
solve real world challenges in areas like image object detected fortracking current object centroids
search and video surveillance to detect a person or and compute the distance between each pair of
vehicle. objectsuch as Euclidean distance. The object needs
to be registered if the number of consecutive frames
To detect vehicles, we subtract image of of the objects is disappeared. In order to register a
the road without vehicles from another image new centroid as a trackable object it must satisfy
having vehicles on road. The background pixels condition that input centroid as a trackable object it
would cancel each other out and the vehicles or must satisfy condition that input centroids should be
objects present in the foreground will appear. greater than the number of existing centroids.

Background subtraction is easy to Vehicle Counting

implement and suitable threshold is given for Vehicle tracked are counted when they
detecting vehicles is an area that remains less than leave the frame or cross a line at an exit point of the
20000. Figure 6 shows objects within the given frame. To count vehicles moving in two different
threshold are pedestrians, animals, people with carts directions we make use of counting lines that is
etc. Also, objects change in size and shape as they down count as red line and up count as blue line.
move across the view. Counted vehicles are classifies based on the
perimeter, if the perimeter of bounding box is lesser
Figure 7 shows the masked image of the than 300 it is counted as bike, if the perimeter of
detected objects. Masking is done to highlight the bounding box is less than 500 it is counted as car
desired objects in a video frame. The objects with an and if the perimeter of bounding box is greater than
area smaller than the threshold value are ignored. 500 it is counted truck/bus.Figure 8 shows the
Selecting Region of Interest (ROI) helps in detecting vehicles counted and classified based the threshold
objects, for vehicle detection and further given to them.
implementation is done by tracking the detected
Vehicles. V. RESULT ANALYSIS
From experimental analysis it is evident
Background Subtraction that vehicle detected at an accuracy rate of about
Background subtraction is process of extracting the 97.39% and vehicle tracked at a rate of about
target image from original image. ID 98.26% as shown below.
Origin: It is the original image is coloured or gray The objects within the given threshold are
scaled image of 8-bit or 32-bit floating point. detected. It is observed that vehicle are not the only
Target: The target image is either 32-bit or 64-bit objects that move on/across the road. The masked
floating point. image of objects detected. Masking is done to
Alpha: Weight of input image. Speed of updation is highlight the desired objects in a video frame.
decided by alpha, set a lower value for this variable Figure 6 shows the vehicles counted and
in existing frames. classified based the threshold given to them.
Finally, information about the vehicles crossed with
Target(x,y)=(1-alpha).target(x,y) + alpha.origin(x,y) height, width and vehicle ID is maintained only to
track the vehicles for counting either moving
Vehicle Tracking upwards or downwards as shown in figure 7.

DOI: 10.35629/5252-030817941800 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 1798
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 3, Issue 8 Aug 2021, pp: 1794-1800 www.ijaem.net ISSN: 2395-5252

Figure 6: Count of vehicles on up and downstream

Figure 7: information of vehicle crossed

TABLE 1
Proposed system count with accuracy rate

A B C D E F G H
60 1 3 4 4 4 100% 100
%
300 2 7 9 9 9 100% 100
%
450 1 6 17 18 18 94.1 100
1 % %
140 1 2 38 40 42 95% 95.2
1 7 1 %
181 3 4 82 85 81 96.3 95.3
2 4 8 % %
251 6 4 11 11 11 97.2 98.4
2 7 5 2 5 3 % %

A – Number of input frames vehicles in each frame based on the information

B – Up count obtained from previous frame. Experimental results,
C – Down count implemented with Open CV, indicates that the
D – Total number of vehicles in truth accuracy rate reaches to 97.1% for object detection
E – Vehicle detected and 98.4% for object tracking. From the table, we
F – Vehicle tracked observe that the proposed method is effective to
G – Accuracy of vehicle detection detect, track, and count moving vehicles in a more
H – Accuracy of vehicle tracking accurate and successful manner irrespective of
weather conditions.
VI. CONCLUSION
In this, we have presented the unitize VII. FUTURE SCOPE
techniques to achieve improvement and out As a future work, major improvements can
performance in the vehicle detection and counting be included by upgrading to higher version of
process. Major technique used to improve detection raspberry pi. This greatly reduces the processing
of vehicles is the use of background subtraction time. Traffic on highways cause occlusions, here
algorithm. The proposed method eliminates the two vehicles are combined to treat two vehicles as a
unnecessary portion and differentiates the vehicles single entity. Due to heavy wind, camera might get
in a more accurate manner. Further, we track affected due to vibrations. This leads to partial

DOI: 10.35629/5252-030817941800 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 1799
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 3, Issue 8 Aug 2021, pp: 1794-1800 www.ijaem.net ISSN: 2395-5252

detection of vehicles.
Alarm system might be incorporated as an
additional future enhancement.

REFERENCES
[1]. Li, D., Liang, B., & Zhang, W. (2014). Real-
time moving vehicle detection, tracking and
counting system implemented with OpenCV.
2014 4th IEEE International Conference on
Information Science and Technology.
[2]. Sorwar, T., Azad, S. B., Hussain, S. R., &
Mahmood, A. I. (2017). Real -time Vehicle
monitoring for traffic surveillance and
adaptive change detection using
Humanitarian Technology Conference (R10-
HTC).
[3]. Yang, H., & Qu, S. (2017). Real time vehicle
detection and counting in complex traffic
scenes using background subtraction model
with low-rank decomposition Intelligent
Transport Systems.
[4]. Soleh, M.,Jati, G., Sasongko, A. T., Jatmiko,
W., &Hilman, M. H.(2017). A real time
vehicle counting based on adaptive tracking
approach for highway videos.
[5]. Aqel, S., Hmimid, A., Sabri, M. A., &Aarab,
a. (2017) road traffic: Vehicle detection and
classification. 2017 Intelligent System and
Computer Vision (ISCV).
[6]. https://www.researchgate.net/publication/281
039935_Real-time moving vehicle
detection_tracking
_and_counting_system_implemented_with_O
penCV.
[7]. Foresti, G. L., &snidaro, L, (2005). Vehicle
Detection and Tracking for Traffic
Monitoring. Lecture Notes in Computer
Science.

DOI: 10.35629/5252-030817941800 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 1800