International Journal of Electronics, Communication

& Instrumentation Engineering Research and

Development (IJECIERD)
ISSN (P): 2249-684X; ISSN (E): 2249-7951
Vol. 8, Issue 2, Apr 2018, 1-6
© TJPRC Pvt. Ltd.

KALMAN FILTER BASED MULTIPLE OBJECT TRACKING SYSTEM

ANITA KULKARNI1 & ELIZABETH RANI2

1
Associate Professor, EIE department, VNR Vignana Jyothi IET, Hyderabad , India
2
Professor & Head, EIE department, GITAM, Vishakhapatnam, India
ABSTRACT

This paper presents a method for tracking multiple objects from a given video dataset. We can track many
objects at a time efficiently using Kalman Filter and Optical flow algorithm. Here in this paper, we propose improved
optical flow algorithm which works with high accuracy and overcomes occlusion in a video. So, improved optical flow
algorithm is found to be more promising as it gives better accuracy in less computation.

KEYWORDS: Multiple Objects, Kalman Filter & Less Computation

Received: Jan 10, 2018; Accepted: Jan 30, 2018; Published: Feb 26, 2018; Paper Id.: IJECIERDAPR20181

INTRODUCTION

Original Article
Object tracking is a crucial task within the field of computer vision. There are three important steps in
video analysis:interesting moving objects detection, tracking of such objects from frame to frame, and analysis of
object tracks to recognize their behaviors. The complexity of object tracking is due to different noises present in
the input from the video, illumination variations in the input, unusual movement of the object and different
occlusions. Most of the tracking algorithms assume that the moving object moves in smooth and no sudden change.
There are two main goals in our project: to correctly recognize moving objects of interest, and to track those
moving objects throughout their life span. Noise is the primary problem. Movement of the camera and background
may lead to pixel movement in the image. This is of no importance. Size constraint is utilized to remove
movements which are very small to contribute to the object movement.

From the input image, the subject to be tracked is to be identified. This is the next work. Conventional
technique to track the system will give poor results. When the subject is in motion, we find that Template Matching
give errors. Hence Template Matching andMotion detection can be used together on the objects that are moving.
Results have shown this to be a robust method in both recognizing the objects as well as tracking them. The rest of
the report presents a brief literature review; a description of the experiment design and the program; presentation of
the results, conclusion and discussion for future work.

LITERATURE REVIEW

Motion detection and object tracking is a very rich area of research in computer vision. The main issues
that make this research area difficult are:

Computational Expense

If an algorithm for detecting motions and tracking objects is to be applied to real-time applications, then it
needs to be computationally inexpensive so that a modern PC has enough power to run it. Yet many algorithms in

www.tjprc.org [email protected]
2 Anita Kulkarni & Elizabeth Rani

this research area are very computationally expensive since they require computing values for each pixel in each image
frame.

Moving Background Rejection

Unwanted movements contributed by the tree branches and so are to be eliminated by the algorithm. If at all if the
object is too small than the background then mistakes in the classifications can be done. Also the speed with which the
object is moving relative to the background can lead to errors.

Method of Occlusion

Number of algorithms is available which suggest solutions for tracking under tiny occlusions, still they fail to
track the object if it is occluded for a long period of time.

Modelling Targets of Interest

Many algorithms use a reasonably detailed model of the targets in objects detection and consequently require a
large number of pixels on target in order to detect and track them properly. This is a problem for real-world applications
where it is frequently impossible to obtain a large number of pixels on target.

Adapting to Illumination Variation

Many situations under real time have changes in the scene illumination and all the techniques should keep this in
mind and try to solve this issue. If the algorithm is written which is based on the intensity technique thee is a probability
that it may not work in sucha situation.

PROPOSED WORK

Kalman filter is a region based method for finding the regions of object in the next frame. The center of object is
found first, and then we use Kalman filter for predicting the position of the object in the next frame. The filter we consider
here will try to estimate the linear system’s state and this is Gaussian distributed. The method using Kalman
filteriscomposed of two steps, prediction and correction as shown in figure.1. Optimal estimate of its position, at each time
of step is provided by the Kalman filter for subjects in movement which also has some unwanted noise due to this
movement.

The optimality is guaranteed if all noise is Gaussian. Then the filter minimizes the mean square error of the
estimated parameters (e.g. position, velocity). The

As the fresh information come they are processed immediately by the Kalman filter and also it is very useful.
Here we need a linear system mixed with unknown disturbances that is dynamic to design a Kalman filter. The Kalman
filter tries to estimate the state a ∈R nofthat system which is governed by the vector difference equation:

Impact Factor (JCC): 6.7847 NAAS Rating: 2.92

 KALMAN Filter Based Multiple Object Tracking System 3

Figure: The Kalman Filter Predict/Correct Model

The random variables ck, dk represent the process and measurement noise respectively. They are assumed to be
zero mean, white noise with covariance matrices respectively. The matrix A is called the state transition matrix and relates
the previous state ak−1 to the current state ak, if no noise was present. The size of A is n×n. Matrix A is optional and
relates the control input (if any) bk∈akto the state ak. Finally, the n×n matrix, relates the measurement tk to the state ak-1.
The Kalman filter maintains the following two estimates of the state:

1. which is an estimate of the state at time-step, given knowledge of the process up to step k-1. It is an a
priori state estimate at time-step 1

2. Which is an estimate of the process state at time-step k given the measurement tk. It is a Posteriori estimate
of the state at time-step k.

Below mentioned Error Covariance matrices are indeed looked after.

1. which is the a priori estimate error covariance of .

2. which is the a posteriori estimate error covariance of .

There are two possible steps in which Kalman filter operates

The First one is the prediction of the next state estimate(k|k− 1) using the previous one. The second is the
correction of that estimate using the measurement, to obtain(k|k). Initially, (1|1) and E(1|1) are considered known.
To maintain those estimates, the following operations take place. In the prediction step:

1. State prediction:

2. Error covariance prediction:

In the correction step

1. Measurement prediction:

2. Residual:

3. Measurement prediction covariance:

4. Kalman gain:

5. State update:

6. Error covariance update:

www.tjprc.org [email protected]
4 Anita Kulkarni & Elizabeth Rani

Kalman filter is initialized by defining the state transition matrix A, The state measurement matrix 4, both
covariance matrices and always this is used to feed the filter with a measurement; k [10]. Multiple objects can be tracked
easily from the video dataset.

SIMULATION RESULTS

CONCLUSIONS

In this paper, we presented a tracking method for processing video data in order to perform tracking by a machine
vision system. It summarizes as multiple objects can be tracked simultaneously. It can control some problem of multi
object tracking, such as appearance and disappearance of objects, and missing of an object. Amongst the methods
reviewed, improved optical flow algorithm is found to be more promising as it gives better accuracy in less computation
time. The present work can be extended further by using more tracking algorithms and comparing their performance
accordingly to achieve more accuracy. Also we will try to test our algorithm on real time video frames.

Impact Factor (JCC): 6.7847 NAAS Rating: 2.92

 KALMAN Filter Based Multiple Object Tracking System 5

REFERENCES

1. ShuTian, Fei Yuan, Gui-Song Xia Multi-objecttracking with inter-feedback between detection and trackingNeurocomputing,
Volume 171, 1 January 2016, Pages 768-780

2. Zhiyu Zhou, Dichong Wu, Zefei Zhu Objecttracking based on Kalman particle filter with LSSVR Optik - International Journal
for Light and Electron Optics, Volume 127, Issue 2, January 2016, Pages 613-619

3. AslaniSepehr and NasabHomayounMahdavi, Optical Flow Based Moving Object Detection and Tracking for
TrafficSurveillance, International Journal of Electrical, Computer, Electronics and Communication Engineering, 2013, Vol 7
No 9.

4. J Berclaz, F Fleuret, E Turetken and P Fua, Multiple Object Tracking using K-Shortest Paths Optimization, IEEETransactions
on Pattern Analysis and Machine Intelligence, 2011, p. 1806-1819.

5. Y Zhai, MB Yeary, S Cheng and N Kehtarnavaz, An object-tracking algorithm based on multiple-model particle filteringwith
state partitioning, IEEE Transactions on Instrumentation and Measurement, 2009, p. 1797-1809.

6. BD Lucas and T Kanade, An Iterative Image Registration Technique with an Application to Stereo Vision, Proceeding
ofImaging Understanding Workshop, 1981, p. 121-130.

7. A Agarwal, S Biswas, S Singh, S Sural and AK Majundar, Object Tracking using Background Subtraction and
MotionEstimation in MPEG Videos, Springer-Verlag Berlin Heidelberg, ACCV, 2006, p.121-130.

8. Qiaona Pei, Moving Objects Detection and Tracking Technology based Optical Flow, North China University of
Technology,2009, p. 11-14.

9. ZervosMichalis, Fua Pascal and ShitritHoresh Ben, Real Time Multi-Object Tracking using Multiple Cameras,
SemesterProject, School of Computer and Communication Sciences, 2012.

10. Henrique Morimitsu, Isabelle Bloch, RobertoM. Cesar-JrExploring structure for long-term tracking of multiple objects in
sports videosComputer Vision and Image Understanding, Volume 159,June 2017, Pages 89-104

www.tjprc.org [email protected]