Object Search Strategy in Tracking Algorithms

Abstract

The demand for real-time video surveillance systems is increasing rapidly. The purpose of these systems includes surveillance as well as monitoring and controlling the events. Today there are several real-time computer vision applications based on image understanding which emulate the human vision and intelligence. These machines include object tracking as their primary task. Object tracking refers to estimating the trajectory of an object of interest in a video. A tracking system works on the principle of video processing algorithms. Video processing includes a huge amount of data to be processed and this fact dictates while implementing the algorithms on any hardware. However, the problems becomes challenging due to unexpected motion of the object, scene appearance change, object appearance change, structures of objects that are not rigid. Besides this full and partial occlusions and motion of the camera also pose challenges. Current tracking algorithms treat this problem as a classification task and use online learning algorithms to update the object model. Here, we explore the data redundancy in the sampling techniques and develop a highly structured kernel. This kernel acquires a circulant structure which is extremely easy to manipulate. Also, we take it further by using mean shift density algorithm and optical flow by Lucas Kanade method which gives us a heavy improvement in the results.