MOVING OBJECT TRACKING FROM VIDEO USING KALMAN FILTER ITS STUDY AND IMPLEMENTATION USING MATLAB

Abstract

Object tracking from video sequence is a fundamental task in many computer-vision applications. A common and widely used approach is to perform background subtraction for determining the moving object against the back ground. There are many challenges in, developing a good algorithm which can cater to various backgrounds and different situations. Changes in lighting conditions, movement of background objects, varying shade of object and nearby objects, shaking of camera etc causes noise while segmenting object from background. The low contrast of some of the object parts against background results in fragmentation of segmented object. The dynamic. background scenario poses big challenge for designing robust algorithms. In this dissertation I have developed an initial background model based on some initial frames of background and used statistical properties of the desired object to segment the object despite presence of noise [5, 6].Thereafter this initial background model is updated- every frame to achieve proper segmentation of the object. Kalman filter has been used to estimate the position of the desired object and to dynamically update the initial background model by predicting the position of the object in future frames. The objects moving in straight line with constant velocity have been modeled in this work. An algorithm to dynamically update the background has been proposed. The performance of proposed algorithm is then evaluated on practical videos for tracking of moving objects. The results of tracking of objects in these practical videos were found satisfactory. The algorithm is then modified to track two objects simultaneously in the given video, albeit, with fixed background. The results of tracking of two objects were also found satisfactory.