Object Tracking with Dynamic Template Update and Occlusion Detec

Abstract

The objective of this paper is to track moving objects using dynamic template initializations and updates, and to identify tracking events in videos such as occlusions and merging of motion regions. The proposed tracking method is based on spatiotemporal texture motion regions, image alignment, and minimum cost estimation based template selection. The dynamic template update is based on the detection of events in videos. The proposed method has been experimentally evaluated on color and thermal infrared videos