Visual Tracking in Discrete Space by Particle Filtering

Abstract

An efficient visual tracking algorithm based on particle filtering is proposed. To reduce the time complexity, a procedure for preventing redundant image processings used to find a target object in an image sequence is introduced. This reduction is achieved by determining equivalence samples (called particles), which represent a state distribution, in the digital space. To evaluate the performance of the proposed algorithm, a noisy image sequence,consisting of 90 binary images with 320*240 pixels, is used.The result shows that the execution time decreases with increasing the number of particles and reduces by 32.9% compared with the conventional way when particles are 10000 in number.