Tracking multiple objects in real time

Abstract

This thesis documents the design and implementation of a real-time object tracking system capable of operating on any modern generic desktop PC including a Video-for-Windows image capture device. A secondary achievement of this thesis is a point tracking algorithm which works seamlessly (though not in real time) with the rest of the system. Multiple points within an object can be tracked. Before presenting details of the implementation, a general introduction and a theory section relative to this implementation is given. In particular, the concepts of the Hausdorff distance and RGB histogram matching are explained. Details and results of the system are presented, illustrating the preprocessing, object isolation, object tracking and histogram matching algorithms. An indication of the accuracy achieved is given by the two case studies included. Speeds of up to 2.5 frames per second were achieved with adequate object tracking accuracy. The area matching method documented herein are shown to be too slow to run in real-time. It is concluded that the multiple object tracking system described performs robustly and efficiently with an adequate accuracy. The feasibility of implementing real-time image processing applications with generic hardware is asserted.