MOVING FOREGROUND OBJECT DETECTION AND TRACKING IN VISUAL SURVEILLANCE SYSTEM

Abstract

ABSTRACT Moving object tracking has been an interesting research field since decades due to its wide applications in the field of security, surveillance, medicine, entertainment etc. A lot of research work has been done previously for object tracking when the camera remains stationary. But, now a days due to increasing popularity of hand held cameras and mobile phone cameras, there is a great demand for moving object tracking when the camera is moving. Unfortunately, none of the methods developed for object tracking in stationary camera conditions, are applicable for object detection or tracking when the camera is not stationary or in other words when camera is moving. This generates a great need for the development of new robust methods to track moving objects, when the camera is moving. In this thesis work, a new method is proposed for moving object tracking in moving camera scenario. In the method proposed, first background is estimated and then background subtraction is performed. In this method, first of all geometric transformation is calculated between every two successive consecutive frames, and then we transform frame (t+1) in the plane of frame t using this geometric transformation, thus cancelling the camera motion. For determining the object motion, we have calculated motion measure (MM) and reliability measure (RM) for each pixel. This motion cue and the cue obtained from the difference of background and current frame are combined to get foreground and background mask. New background is modeled for the current frame using this foreground and background mask and then background subtraction is performed for final foreground and background labeling. The proposed method is able to track the object present even in the first frame of the video. This method also handles fast moving background. Experimental results show that, the results are reasonably good compared to most of the other existing methods.