International Journal of Advanced Science and Technology
Vol. 30, No. 1, (2021), pp. 50-55
Computer Vision Based Moving Object Detection and Tracking
Suresh Kumar1 , Prof. Yatin Kumar Agarwal2
1
[Link]. Scholar, 2Professor
12
Department of CSE, Greater Noida Institute Of Technology (GNIOT) Greater Noida, Uttar
Pradesh, India
Abstract
To detect and track objects, some approaches of detecting and tracking moving objects in stationary
scene are presented, including the detection methods of the time domain differential method, the
background differential method, optical flow method and tracking methods of the Kalman filter and
feature optical flow. Keywords-moving object detection and tracking; background model; Kalman
filter; optical flow. Object tracking is an important task within the field of computer vision. It is a
challenging problem. Many difficulties arises in tracking the objects due to abrupt object motion,
changing appearance patterns of both the object and the scene, non-rigid object structures, object-to-
object and object-to-scene occlusions, and camera motion. This paper selectively reviews the
research papers with regard to tracking methods on the basis of the object, their motion
representations and all detailed descriptions of representative methods in each category examining
their advantages/disadvantages.
Keywords:- Object Representation, Object Tracking, Object Detection, Computer Vision
I. INTRODUCTION
To develop the real world computer vision system, tracking of moving objects is very important task.
The proliferation of high-powered computers, the availability of high quality and inexpensive video
cameras, and the increasing need for automated video analysis has generated a great deal of interest in
object tracking algorithms [1]. Applications are like automatic video surveillance, motion-based
recognition, video indexing, humancomputer interaction, traffic monitoring, and vehicle navigation.
Real-time moving object detection and tracking [1][2] organically combines the image processing,
automation and information science technology to form a kind of new technology that moving targets
can be quickly detected in the images and target location can be extracted for tracking purpose.
Moving picture sequence can provide more useful information for low signal to noise ratio (SNR)
target detection, using image sequences can detect target which is difficult to detect in a single-frame
image. Image sequences formed by the moving target can be categorized as two cases: one is the
static background, the other is the varying background. The former case usually occurs in the camera
which is in a relatively static state, produces moving image sequences with static background. Then
we can use a temporal differencing algorithm or adaptive cancellation of the background method to
remove the background interference, the calculation is simple; The latter case usually occurs in the
target movement, while the camera is also in the relative movement state. Then it produces moving
image sequences of changing background and leads that the changes of the moving target in the image
are confused with the changes of the background itself. To deal with this case is complicated, if the
cancellation of the background method [3] selected, it needs to do inter-frame image stabilization [2]
and image registration [2]; else the prominent to an end method selected, it needs to do multi-frame
energy accumulation and noise suppression.
[Link] WORK
There is much research work in the field of object tracking in videos over the past decades. Some of
the work done has been discussed one by one below. Alexander Toshev, Ameesh Makadia et al. [5],
Presented shape based object recognition in Videos Using 3D Synthetic Object Models. This paper
sorted the problem of recognition of moving objects from the videos by synthetic 3D models. At first,
from the video, the silhouette images of the moving object is extracted by feature tracking, motion
grouping of tracks and co-segmentation of successive frames and then matched to 3D model
50
ISSN: 2005-4238 IJAST
Copyright ⓒ2021 SERSC
International Journal of Advanced Science and Technology
Vol. 30, No. 1, (2021), pp. 50-55
silhouettes. As a result, the matching of every 3D model to the video. This approach can recognize
objects in videos and estimate their rough pose by using only similar but not 3D models.
Mohammed Sayed and Wael Badawy [6], Presents a novel motion estimation method for mesh-based
video motion tracking. The method called mesh-based square-matching (MB-SM) motion estimation
method. This method is used in terms of computational cost reduction, efficiency and image quality. It
is a modified version of the hexagonal matching motion estimation method. The MB-SM motion
estimation is performed in two ways. First is the rough motion estimation using block-matching
algorithm. Second is the fine motion estimation to refine the motion vectors generated from the first
step. As result, MB-SM method has lower computation cost than the hexagonal matching motion
estimation method while it produces almost the same PSNR values.
Minglun Gong [7], Proposed estimating 3D geometry and motion of dynamic scenes based on
captured stereo sequences. A dynamic programming based technique is used for searching global
optimal disparity maps and disparity flow maps under an energy minimization framework. As a result,
both computations can benefit from each other and are capable of producing both 3D geometry and
motion information for dynamic scenes in nearly real time.
Ming-Yu Shih et al. [8], In this paper, a method of moving object detection on moving platforms is
proposed. This method composed of moving blob detection and shape refinement phases to provide
robust moving object detection result. By fusing motion field’s information from three consecutive
frames, positions of moving blobs were precisely detected. Next, using motion compensated
background models in intensity, r, and g color spaces, shapes of objects are well refined in the fused
background subtraction process. By combining moving blob and contextual background information,
alignment errors could be eliminated effectively to prevent foreground pixels from being adapted into
background models.
Sajjad Torkan, Alireza Behrad [9], proposed a new contour based tracking method using active
contour. Original greedy snake as a parametric active contour has weak performance in tracking target
with high velocity and large displacement between two successive frames. This is due to concave
parts of target boundary, shrinkage of contour if it is far from target boundary and the lack of target
motion information. To resolve this problem a greedy snake with adaptive curvature energy and
additional field energy term as an external energy. Kalman filtering is used for handling large target
displacement and contour concavities. Kalman filtering is also used for the estimation of target shape
and centroid in current frame. Applied on wide variety of video sequences results showed that the
algorithm capable of tracking target with high speed, large aspect change and contour concavities.
Baiyang Liu, Lin Yang et al. [10], proposed an adaptive tracking algorithm for lung tumors in
fluoroscopy using online learned collaborative trackers. No shape or motion priors are required for
this tracking algorithm. This saves many expensive expert annotations. Accurate tracking of tumor
movement in fluoroscopic video sequences is a clinically significant and challenging problem. This is
due to blurred appearance, unclear deforming shape, complicate intra and inter- fractional motion, and
other facts. Current offline tracking approaches are not adequate because they lack adaptivity and
often require a large amount of manual labeling. Therefore this adaptive online learning algorithm is
general to be extended to other medical tracking applications.
Pengwei LIU, Huiyuan WANG et al. [11], presented an approach for handling target detection and
tracking in dynamic scenes, in which, motion compensation is generated by pyramidal optical flow.
Because In most cases of target tracking in video sequences, the scene is dynamic due to the mobile
cameras and causes problems. Main contribution in this system is that it separates the values of
pyramidal optical flow into two groups: one represents background and the other foreground.
Experiments show that the algorithm achieves motion compensation very effectively. It is only
applied in two-dimensional motion cases such as that the camera moves only up and down.
51
ISSN: 2005-4238 IJAST
Copyright ⓒ2021 SERSC
International Journal of Advanced Science and Technology
Vol. 30, No. 1, (2021), pp. 50-55
Mark Ritch, Nishan Canagarajah [12], proposed a method to identify and track an object of interest
within compressed MPEG-2 video using only motion information. The system is designed to detect
interesting events taking place, such as moving object appearing and to track it without the need to
decode the video. Iterative rejection is used as a basis for initial segmentation and those macro blocks
whose motion vector is large than average are rejected. This is done by adopting a novel model based
approach. Experimental results on a number of sequences demonstrate its effectiveness in identifying
and tracking on object of interest from a compressed video stream without the need to fully decode
each frame and that the system performed better than using iterative rejection alone as a segmentation
method.
Huiqiong Chen, Derek Rivait and Qigang Gao [13], presents perceptual organization based method
for real-time license plate identification and tracking. In this method, video content is described by
Generic Edge Tokens (GETs), and an image is represented as a GET map. The structure provided by
GETs allows edge detection to be performed faster. A MGET graph representation is proposed for
coding motion content, in that a license plate is a sub-MGET-graph (SMG) which satisfies license
plate model. The SMG representing the license plate is identified by perceptually grouping plate
shape in the MGET graph. Experiments show that the method is able to effectively segment license
plates from the video sequences. Whether the video clips were taken during the day or at night did not
have a large impact on the program results. The application correctly identified 66 of these license
plates, giving an accuracy of approximately 98.5%.
[Link]
Detection of objects in motion is the first step towards non-stationary object tracking. Object
detection is the method of finding the non-stationary object in a video sequence. Some of the major
and important methods of detecting the moving objects are Frame differencing, Optical flow,
Background subtraction and Double difference etc. Object representation is the process of
demonstrating the objects. Object representation can be categorized as shape representation, color
representation, texture based representation and Motion oriented representation. Object tracking is the
process determining the position of the moving entity in a sequence of video.
Certain types of tracking algorithms are point tracking, Motion tracking, Shape tracking, Feature
tracking and Kernel based tracking. Step by step detail of the nonstationary object tracking is as
shown in figure 1.
2.1) Object detection Object detection is the method for recognizing the nonstationary or moving
object in a video sequence. This is the primary and main step towards moving object tracking
Different detection techniques are as despite in figure 2
52
ISSN: 2005-4238 IJAST
Copyright ⓒ2021 SERSC
International Journal of Advanced Science and Technology
Vol. 30, No. 1, (2021), pp. 50-55
3.1 Optical Flow Optical flow is substitute standard from of object detection in which the optical
flow arena of the image is calculated and grouping of those arenas is done rendering to appearances of
the image. The motion among dual video frames occupied at time t and t+ δt at every single location
is estimated in optical flow process. This technique gives the broad information regarding the
movement of the object. And also detects the object accurately compared to that of background
technique. This method is not widely used because of its huge calculation and it is very sensitive to
noise. It is not good for real-time occlusion condition. 2.1.2) Background subtraction Background
subtraction is the most widely used method for moving object detection. It can be of two types firstly
by considering first frame as the reference frame or background image. Secondly by considering
average of „n‟ frames as the background image. In this background subtraction method every pixel of
on-going frame is subtracted with the pixels of the background image. The equation (1) and (2) shows
the background subtraction method for first frame as the background image.
[Link] RESULTS
The algorithm can be implemented by Microsoft Visual Studio, test sequences use the 640×480
surveillance video. We do the experiment on a computer, and the experimental results are shown in
Figure3.
Figure 3. Results of moving target detection on static background:(a)original frame;(b)result frame
53
ISSN: 2005-4238 IJAST
Copyright ⓒ2021 SERSC
International Journal of Advanced Science and Technology
Vol. 30, No. 1, (2021), pp. 50-55
Experiment results show that the method proposed by this paper can effectively extract the moving
target, especially when the velocity of target is not fast and the background is not very complicated, it
can meet the requirements of real-time and accuracy.
5. CONCLUSIONS AND FUTURE WORK
We have presented the survey of object tracking methods and all categories of moving objects that is
object representation, object tracking and object detection from any type of video. This will help us to
significantly improve and facilitate the performance of certain computer vision tasks, such as tracking,
video surveillance, motion-based recognition, video indexing, human-computer interaction, traffic
monitoring, and vehicle navigation. Our future work will focus on: 1) Movement detection and
capable of finding the objects which are in motion in every frame with respect to the previous frame.
Till now in many proposed works, we have been plotting the movement of objects through videos but
none of them is capable of plotting graph of the moving object if the background is changed at any
instant. 2) The coordinates of tracked video in real time can import to any other software to work with
the results calculated by our algorithm. 3) Our algorithm is capable of tracking the objects and making
a 3d graph in mesh based scope in between any number of fps (frames per second) video.
REFERENCES
[1] Pengwei LIU, Huiyuan WANG et al., Motion Compensation Based Detecting and Tracking
Targets in Dynamic Scene, IEEE, 2010, 703-706.
[2] Sajjad Torkan, Alireza Behrad, A New Contour Based Tracking Algorithm Using Improved
Greedy Snake, IEEE, 2010.
[3] Baiyang Liu, Lin Yang et al., An Adaptive Tracking Algorithm Of Lung Tumors In Fluoroscopy
Using Online Learned Collaborative Trackers, IEEE, 2010, 209-212.
[4] Alexander Toshev, Ameesh Makadia, Kostas Daniilidis, Shape-based Object Recognition in
Videos Using 3D Synthetic Object Models, IEEE, 2009, 288-295.
[5] Makadia et al, Motion-based Background Modeling for Moving Object Detection on Moving
Platforms, IEEE, 2007, 1178-1182.
[6] Mohammed Sayed and Wael Badawy, Motion-Based Video Object Tracking In The Compressed
Domain, IEEE, 2007, 301-304.
[7] Minglun Gong, Object Tracking:A Survey, ACM Computing Surveys, 38(4), 2006.
[8] Ming-Yu Shih et al. , A Thesis, Thapar Institute of Engineering and Technology, Patiala, ME,
2006.
[9] Sajjad Torkan, Alireza Behrad Algorithm for Estimating 3D Geometry and Motion in Near Real-
time, IEEE, 2006.
[10] Huiqiong Chen, Derek Rivait and Qigang Gao, Real-Time License Plate Identification by
Perceptual Shape Grouping and Tracking, IEEE, 2006, 1352-1357.
[11] Pengwei LIU, Huiyuan WANG et al, A novel motion estimation method for mesh- based video
motion tracking, IEEE, 2004, 337-340.
[12] Li-Qun Xu, Simultaneous Tracking And Segmentation Of Two Free Moving Hands In A Video
Conferencing Scenario, IEEE, 2003, 49-52.
54
ISSN: 2005-4238 IJAST
Copyright ⓒ2021 SERSC
International Journal of Advanced Science and Technology
Vol. 30, No. 1, (2021), pp. 50-55
[13] Mark Ritch, Nishan Canagarajah Particle Filter With Multiple Motion Models For Object
Tracking In Diving Video Sequences, IEEE, 2008, 224-228.
[14] Harini Veeraraghavan, Paul Schrater and Nikolaos Papanikolopoulos, Switching Kalman Filter-
Based Approach for Tracking and Event Detection at Traffic Intersections, IEEE, 2005, 1167-1172.
[15] Yumi Iwashita, Ryo Kurazume et al., Fast Implementation of Level Set Method and Its Real-time
Applications, IEEE, 2004, 6302- 6307.
55
ISSN: 2005-4238 IJAST
Copyright ⓒ2021 SERSC
