Marketing Department

World Academy of Science, Engineering and Technology Vol:6 2012-08-21

Multiple Object Tracking using Particle Swarm Optimization

Chen-Chien Hsu, Guo-Tang Dai

Currently, several multiple object tracking algorithms are known to be available, including Kalman filter [8], particle filter (PF) [2], [3], [4], [6], and Mean Shift [18] etc. As a nonlinear time-series filter, PF is mainly used for position estimation of a target object. However, when the environment to be tracked becomes too diverse, tracking results will be affected and hence errors may occur. Furthermore, the disturbances introduced into the next generation by PF generally incur accumulated tracking errors, which inevitably affect the accuracy in tracking the target objects. In particular, when two or more objects come close to each other or overlap, multiple object tracking by particle filter based methods generally fails, because particles tend to move to regions of high posterior probability. Furthermore, when an object disappears or re-enters the screen from a different location, tracking of these multiple objects generally fails. As an attempt to solve this problem, this paper proposes a particle swarm optimization (PSO) based approach for multiple object tracking [5] based on histogram matching [7],[19],[23]. By obtaining the difference between the gray-level histogram within the search range of the video images and that of the target objects, fitness associated with each particle can be evaluated to evolutionally track the objects through the introduction of multiple swarms. With its capabilities of directed random search for global optimization, the PSO has provided a promising alternative to address the above-mentioned problems and difficulties. Experimental results show that the proposed PSO algorithm can quickly converge and hence be able to track multiple objects in real time. Especially, when objects move outside the search window and then re-enter again, re-tracking can be achieved thanks...