2020 IEEE International Conference for Innovation in Technology (INOCON)

Bengaluru, India. Nov 6-8, 2020

Vehicle Detection Using Image Processing

Seelam Shanmukha Kalyan1,Voruganti Pratyusha2, Nandikonda Nishitha3 and T.K.Ramesh4
1,2,3,4
Department of Electronics and Communication Engineering,
Amrita School of Engineering, Bengaluru,
Amrita Vishwa Vidyapeetham, India
4
tk_ramesh@blr.amrita.edu, 1shanmukhakalyan03@gmail.com, 2pratyu2106@gmail.com, 3nishithanandikonda@gmail.com

Abstract- Nowadays, detection of the vehicles and their improvement in this domain. To overcome all these
classification is very essential and also it has a lot of importance drawbacks, an efficient algorithm is designed to detect
because of its use in many applications. One of the main vehicles in an image using image processing. As Image
applications is controlling and managing traffic. Vehicle processing is a vast domain which helps us to process the
detection and tracking plays a major role in the project of image as per our convenience and requirement by using
preventing road accidents using image processing. Tracking of various applications of this domain.
moving objects is important in monitoring surveillance videos
and capturing human motion. Taking its importance into In this paper, an algorithm is developed for vehicle
consideration, an efficient algorithm is proposed to detect detection by mainly using two applications. The two
vehicles in an image using image processing. The image is applications which are used are edge detection and
captured from the front view of the vehicles. So, this algorithm morphological processing which are two main applications of
detects vehicles using the front view. Each vehicle is detected image processing. After enhancing the objects in the image,
based on its size. The two major techniques used in this blob analysis is used to detect vehicles. Finally, rectangular
algorithm are edge detection and morphological processing.
box is assigned to each detected vehicle.
Edge detection as well as morphological processing are
important applications in image processing because of their This paper is focussed as: In section 2, We will be briefly
wide range of uses. Edge detection is used to enhance the discussing about all the researches, existing methods and also
objects in image. Morphological operations are used to remove the limitations of those methods. In section 3, the block
noise and as well as to adjust image in such a way to detect diagram of algorithm developed is shown and the proposed
objects in an image. The simulation work of this algorithm is methodology is explained in detail. Section 4 is concluded as
done in MATLAB which is a very strong scientific tool. summary of the method explained in section 3.
Keywords— Vehicle detection; vehicles; Image Processing; II. RELATED WORK
image; edge detection; morphological processing.
A unique algorithm for vehicle detection and vehicle
I. INTRODUCTION tracking in video frames has been designed in [6]. This
Today road fatalities and traffic management are the two system is developed using gaussian mixture models(GMM)
major issues faced by many citizens all over the country and method of blob detection. By learning the background,
especially in urban areas. According to the survey, in past first they differentiated foreground from background so that
few years around 4 lakh accidents happened each year. And the detection of vehicles becomes simple. This differentiation
40-50 percent of the unnatural deaths are contributed by is done by GMM. And then to find the moving objects
these road accidents. Among these accidents, metropolitan correctly, the noise is removed with the help of
cities has major contribution. The important fact is that 90 morphological operations. In each frame, the movement of
percent of the accidents are done due to lack of traffic objects is known by applying blob detection methods. To
monitoring [1]. Over speeding vehicles by youth, rash define rectangular boxes around each object, binary
driving, drunk and drive, red light jumping, using phone computation is done. Finally tracking and counting of
while driving are some of the important reasons which causes vehicles is done based on the rectangular boxes drawn. But in
accidents due to driver’s carelessness. These accidents had this model GMM is used which does not consider spatial
ruined many lives and sometimes government property is information and this is the major disadvantage for this
also damaged. So, this is the major challenge for the people system.
in India especially the people who are leading urban life. To The main motto of the model designed in [7] is detection
overcome these challenges, many methodologies have been of moving vehicles and segmentation using background
proposed using emerging technologies [2]. For most of the subtraction concept. The first step in this system is, we need
methods, the first and foremost important step is to detect to assume a stationary background and we need to take the
vehicles and their tracking. So, vehicle detection has a lot of difference between background image and the image in
importance in today’s life. With the help of vehicle detection, which vehicles should be detected i.e the input image. To
vehicles can be tracked time to time which avoids all the know the features of the moving object, segmentation is done
problem mentioned above [3]. Vehicle detection helps in which is one of the most important application in video
video surveillance application [4]. In past few years, many image processing. Segmentation is two-step process. First
algorithms have been developed for vehicle detection and thresholding is applied and then edge detection methods are
tracking. Most of the algorithms proposed are not effective applied over the image. Then various morphological
and useful in all the situations. And also, most of the methods operations are performed to remove the noise in the image.
failed to work during night time because these methods The major drawback in this project is that this model cannot
couldn’t recognize the difference between ambient light and be considered when the background is dynamic.
head light of the vehicles [5]. So, there is still a scope of

978-1-7281-9744-9/20/$31.00 ©2020 IEEE 1

Authorized licensed use limited to: American University of Beirut. Downloaded on May 25,2021 at 23:39:29 UTC from IEEE Xplore. Restrictions apply.
 An algorithm for vehicle detection is proposed in [8] by In this paper [13] they explained about how the speed of
using thresholding, image differencing, binary morphological the vehicle is tracked by its motion with sequence of images
process and edge detection. First, the frames are extracted as input. For this model background subtraction is used. This
from video. In this algorithm there are two phases- pre- model consists mainly 5 stages for tracking the vehicle
processing phase and vehicle detection phase. First the speed.1) Pre-processing: In this stage the video is first
background image is captured for reference image. Then converted into frames and then frame rate is extracted from
thresholding is applied to both reference image and input that. In the next step noise is removed by using median
image to convert them to binary images. Then subtraction filtering. 2)Moving vehicle detection: In this step, first we do
between both the images is performed and edge detection background subtraction i.e., foreground detection. In this step
methods are applied to identify the boundaries of the each and every pixel's intensity values are modified using
vehicles. The image is again processed with morphological Gaussian mixture model (GMM). After differentiating the
operations. Finally, based on the number of pixels, all the foreground, those pixels are grouped using 8 connectivity
vehicles are classified. This model cannot be used when the connected components. 3)Feature Extraction: In this step
background is dynamic and when there are shadows in the they are finding the edges of the image by using the j.shi and
image which is the drawback of this system. c. Tomasi method.4) vehicle tracking: It traces the moving
objects with respect to successive video frames. It is done by
In this paper [9], a model has been designed for vehicle the extracting features of objects in a frame and finds the
detection in traffic monitoring using image processing. In objects in sequence. 5) speed estimation: Here, distance is
this model two methodologies are defined for day time calculated by comparing the centroids of the objects in
processing and night time processing. Background previous and the next frame. Time is noted when vehicle
subtraction is performed to identify the objects. During the enters the ROI (Region of interest). Then based on the
night time, to identify the difference between headlight and formula we will calculate speed.
ambient light, the image intensity is analysed. And object
counting methods are applied to count vehicles. As told (Speed =distance travelled /time taken).
earlier this model is also not useful in all the situations
because the background cannot be static always All the discussed models except morphological
processing cannot be used for the case of static background.
In this paper [10], different methodologies which are In the next section, we proposed an algorithm to detect
useful for object classification and identification are vehicles with morphological processing combined with edge
explained. The methodologies explained in this paper are detection which gives better output and better accuracy
optical flow, background subtraction, temporal differencing, compared to the existing.
frame differencing. These are the classical approaches used
for vehicle detection. However, all these methods have some III. EFFECTIVE VEHICLE DETECTION
limitations. Frame differencing method cannot obtain ALGORITHM
complete contour of the object. Temporal differencing is not An efficient algorithm for vehicle detection has been
helpful when the target moves slowly because it results to designed in this section. Fig 1 shows us the basic design of
inappropriate output. And for optical flow method, large the method developed. This method is divided into four
number of calculations has to be performed. phases as – Basic processing phase, Edge detection phase,
In paper [11], an algorithm for vehicle detection is Morphological operation phase and vehicle detection phase.
proposed which is done using two phases i.e the pre- As each phase of the algorithm has its own importance,
processing phase and operational phase which is also termed all the four phases of this method play an important role in
as morphological processing phase. In phase 1, the image is detecting the vehicle in an image given as input. The four
firstly converted into grey scale and then noise in the image phases are explained as follows:
is removed with the help of median filtering. In phase 2, the
image is binarized and morphological operations are applied
on the image. A structuring element is first defined and then
with the help of that structuring element, all the operations
are performed. And then Sobel operator is used for edge
detection. Finally, the vehicles are detected and counted.
In this paper [12] , an algorithm is designed to detect
vehicles in the traffic scenes which are dynamic. The picture
is captured by the camera which is fixed at a position to
monitor and control traffic on roads. First, to detect any
object in the image, we need to enhance the edges of the
objects. To find the edge features of the objects, this
algorithm has used Sobel operator, Scharr operator and also
blob analysis. First the input is video which is converted into
image frames. For each image frame, the region of interest is
calculated. And then to enhance the edges, edge detection is
performed. The operator used for edge detection is Sobel
operator. Next step is to find all the connected components
and their labelling based on connected components achieved
from the image. Final step applied is blob analysis which is
used to detect objects in every image. After all the objects Fig. 1. Basic design of the algorithm
getting detected, rectangular box is assigned to each vehicle.

Authorized licensed use limited to: American University of Beirut. Downloaded on May 25,2021 at 23:39:29 UTC from IEEE Xplore. Restrictions apply.
A. Basic processing phase: In this phase, connected components analysis is done to
In this phase, the image frame which is captured from the remove the connected components which are fewer in
camera fixed at a position is taken as input to the algorithm. number comparing with the size mentioned in the command.
As the image taken is a colour image, the image is firstly The size is initialized based on the size of the object to be
converted from RGB to Gray as it is the first and foremost detected. And finally, using the regionprops, rectangular box
step followed in image processing. Then the frame which is is allocated to each vehicle detected in an image.
obtained after converting to grey scale should be binarized. The flowchart of the proposed method is as shown in fig
Binarizing the image helps us in processing the image 2. First, the image is taken as input to the algorithm. Then
efficiently in later stages. this image is converted from RGB to grey scale for better and
B. Edge detection phase: easy processing. In the next step, the image is binarized with
the suitable threshold taken. After binarizing the image, the
This is the important phase in this methodology because edge detection operator is applied. The operator used here is
for detecting any type of objects present in an image, first we “Sobel” operator. This operator is used to emphasize and
need to highlight them. So, we should now highlight the brighten the edges of the objects in the image. With the help
boundaries of the object for which edge detection is applied. of edge detection, the tracking of vehicle would become
Edge detection is one of the most important applications in easy. Then dilation and erosion operations are performed
image processing. In this algorithm we have used “Sobel” based on the shape and size of structuring element taken.
operator for edge detection. Dilation connects all the necessary objects in the image.
C. Morphological operation Phase: Erosion removes unwanted pixels from the image. Then
based on the connected components set of pixels vehicles are
In this phase a series of morphological operations are
detected and bounding boxes are drawn around each vehicle.
applied on the image to make it suitable for the vehicle
detection in the next phase. Erosion and dilation comes under IV. RESULTS
morphological operations. The operations are performed with
the help of structuring element. Structuring element of The entire vehicle detection algorithm is implemented in
suitable size and suitable shape are chosen based on the size MATLAB simulation software which is developed by
and shape of the object to be detected. MathWorks. The output at each step of the algorithm is
shown in this section.
D. Vehicle detection phase:

Fig. 3. Image Captured by the camera

Fig. 4. Phase1 Output

Fig. 2. Flow chart for the vehicle detection

Authorized licensed use limited to: American University of Beirut. Downloaded on May 25,2021 at 23:39:29 UTC from IEEE Xplore. Restrictions apply.
Fig. 5. Phase 2 Output

Fig. 9. Output of existing method

This vehicle detection algorithm gives better output than

many of the existing algorithms even during the night time.
The result of the output is shown below in fig 10.

Fig. 6. Phase 3 Output

Fig. 10. Output during night time

Fig. 7. Phase 4 Output

As seen in fig 7, the bike is also getting detected for this

proposed algorithm. Whereas, in the existing algorithm only
cars are getting detected. And also, as shown in fig 8, the
person is also getting detected for proposed method whereas
person is not getting detected for existing algorithm as shown
in fig 9.

Fig. 11. Accuracy Plot

The perfect output can be obtained in the night time by

placing the ambient light sensor in the system which has
ability to differentiate between ambient light and headlight of
the vehicles during the night time. This helps us in detecting
all the required vehicles in the image. Now to show that
proposed algorithm is more accurate than the existing
algorithm, the results and accuracies of the previous existing
algorithm and the proposed algorithm are compared using the
accuracy plot shown in fig 11. This accuracy plot is drawn by
considering random five real time images and by comparing
the results of both algorithms.
Fig. 8. Output of proposed method

Authorized licensed use limited to: American University of Beirut. Downloaded on May 25,2021 at 23:39:29 UTC from IEEE Xplore. Restrictions apply.
 V. CONCLUSION AND FUTURE SCOPE [5] G. Muslu and B. Bolat, "Nighttime Vehicle Tail Light Detection with
Rule Based Image Processing," 2019 Scientific Meeting on Electrical-
Nowadays, Vehicle detection has a lot of use and Electronics & Biomedical Engineering and Computer Science
importance in many of the real time day to day applications. (EBBT), Istanbul, Turkey, 2019, pp. 1-4, doi:
Many researches have been done in this domain and many 10.1109/EBBT.2019.8741541.
methodologies have been developed. But there are some [6] P. K. Bhaskar and S. Yong, "Image processing based vehicle
detection and tracking method," 2014 International Conference on
limitations for the existing models which are discussed in Computer and Information Sciences (ICCOINS), Kuala Lumpur,
section 2. So, an efficient algorithm has been proposed in 2014, pp. 1-5, doi: 10.1109/ICCOINS.2014.6868357.
section 3 which gives better output in all the situations based [7] A. S. Mohan and R. Resmi, "Video image processing for moving
on the results shown in section 4. This proposed algorithm is object detection and segmentation using background subtraction,"
simulated with real time data set of around 100 images taken 2014 First International Conference on Computational Systems and
at morning time, afternoon time, evening time and also night Communications (ICCSC), Trivandrum, 2014, pp. 288-292, doi:
10.1109/COMPSC.2014.7032664.
time. This algorithm works better in all the situations which
can be proved by looking at the results and the accuracy plot [8] H. T. Pandu ranga, M. Ravi kiran, S. Raja shekar and S. K. Naveen
kumar, "Vehicle detection and classification based on morphological
which is shown and discussed in the previous section. technique," 2010 International Conference on Signal and Image
Though this algorithm does not work perfectly, these results Processing, Chennai, 2010, pp. 45-48, doi:
are better than existing algorithm. In this model, as we can 10.1109/ICSIP.2010.5697439.
see in the output image, the car in the shadow is not detected. [9] P. N. Chowdhury, T. Chandra Ray and J. Uddin, "A Vehicle
To overcome this, we can use deep learning methods in Detection Technique for Traffic Management using Image
which first the training is done by using vehicle dataset. And Processing," 2018 International Conference on Computer,
Communication, Chemical, Material and Electronic Engineering
also to detect the vehicles at different times of the day, we (IC4ME2), Rajshahi, 2018, pp. 1-4, doi:
can change threshold values which are suitable for the image 10.1109/IC4ME2.2018.8465599.
intensities by using fuzzy logics. [10] J. S. Kulchandani and K. J. Dangarwala, "Moving object detection:
Review of recent research trends," 2015 International Conference on
REFERENCES Pervasive Computing (ICPC), Pune, 2015, pp. 1-5, doi:
[1] Y. Qu, L. Jiang and X. Guo, "Moving vehicle detection with 10.1109/PERVASIVE.2015.7087138.
convolutional networks in UAV videos," 2016 2nd International [11] B. Pawar, V. T. Humbe and L. Kundnani, "Morphology based moving
Conference on Control, Automation and Robotics (ICCAR), Hong vehicle detection," 2017 International Conference on Big Data
Kong, 2016, pp. 225-229, doi: 10.1109/ICCAR.2016.7486730. Analytics and Computational Intelligence (ICBDAC), Chirala, 2017,
[2] S. Santra, S. Roy, P. Sardar and A. Deyasi, "Real-Time Vehicle pp. 217-223, doi: 10.1109/ICBDACI.2017.8070837.
Detection from Captured Images," 2019 International Conference on [12] S. A. Nur, M. M. Ibrahim, N. M. Ali and F. I. Y. Nur, "Vehicle
Opto-Electronics and Applied Optics (Optronix), Kolkata, India, detection based on underneath vehicle shadow using edge features,"
2019, pp. 1-4, doi: 10.1109/OPTRONIX.2019.8862323. 2016 6th IEEE International Conference on Control System,
[3] Pingting Luo, Fuqiang Liu, Xiaofeng Liu and Yingqian Yang, Computing and Engineering (ICCSCE), Batu Ferringhi, 2016, pp.
"Stationary vehicle detection in aerial surveillance with a UAV," 2012 407-412, doi: 10.1109/ICCSCE.2016.7893608.
8th International Conference on Information Science and Digital [13] P. M. Dhulavvagol, A. Desai and R. Ganiger, "Vehical Tracking and
Content Technology (ICIDT2012), Jeju, 2012, pp. 567-570. Speed Estimation of Moving Vehicles for Traffic Surveillance
[4] B. F. Momin and T. M. Mujawar, "Vehicle detection and attribute Applications," 2017 International Conference on Current Trends in
based search of vehicles in video surveillance system," 2015 Computer, Electrical, Electronics and Communication (CTCEEC),
International Conference on Circuits, Power and Computing Mysore, 2017, pp. 373-377, doi: 10.1109/CTCEEC.2017.8455043.
Technologies [ICCPCT-2015], Nagercoil, 2015, pp. 1-4, doi:
10.1109/ICCPCT.2015.7159405.

Authorized licensed use limited to: American University of Beirut. Downloaded on May 25,2021 at 23:39:29 UTC from IEEE Xplore. Restrictions apply.