Target tracking with composite linear filters on noisy scenes

Abstract

A tracking system using a bank of adaptive linear filters is proposed. Tracking is carried out by means of multiple target detections. The linear filters are designed from multiple views of a target using synthetic discriminant functions. For each view an optimum filter is derived from noisy reference image and disjoint background model. An iterative algorithm is used to improve the performance of the synthesized filters. The number of filters in the bank can be controlled to guarantee a prescribed tracking accuracy. Computer simulation results show that the proposed algorithm is able to precisely track a target.