Adaptive Multiple Kernels with SIR-Particle Filter Based Multi Human Tracking for Occluded Environment

This paper proposes a new technique to build a fully automatic tracking system which handles occlusion problem in a complex environment. In multiple human tracking, handling of occlusion is the challenging issue. When occlusion occurs, kernel based tracking was proven to be the promising approach. Hence, to overcome the occlusion problem the human body was considered to have multiple kernels. In this paper, SIR-Particle filter tracking was embedded with multiple kernels that build a fully automatic tracking system. The accuracy of the tracking system was evaluated by using Multiple Object Tracking Accuracy (MOTA) metric. Our tracking system was experimented using PETS benchmark dataset and found that the accuracy was computed as 97%. Keywords-Human tracking, MOTA, Multiple kernels, Occlusion, SIR-Particle filter