An adaptive visual tracking method based on multiple features

Object tracking naturally plays a key role in any visual surveillance system, and there are a number of tracking algorithms for different applications. Here we present an object tracking system based on the Multiple Hypothesis Testing approach. The main characteristic of our approach consists in the development of a probabilistic data association mechanism which makes use of multiple features about each observed objects in the scene. The appearance and disappearance of object is based on a hypothesis matrix. Each matrix element, represents the possibility that a given object at a certain time instant matches another object at a successive time instant. In practice the matching between objects is obtained by comparing Kalman predicted features and observed features between successive time steps. Therefore, our algorithm dynamically creates and destroys tracks on the basis of the hypothesis matrix.