Real Time Multiple Object Tracking and Occlusion Reasoning Using Adaptive Kalman Filters

Object tracking in image sequences is one of the fundamental steps in designing intelligent surveillance systems. The fact that Multiple Object Tracking (MOT) algorithms requires occlusion reasoning and data association, makes design of these algorithms much more complicated than Single Object Tracking (SOT) algorithms. A new method for real time MOT is introduced in this paper to efficiently solve the occlusion issue. Background subtraction has been employed for detecting objects in this method. In order to computing data association between object in current frame with previous tracks, a new distance function is introduced for implementing General Nearest Neighbor (GNN) method. In the case in which objects are in a distance, Kalman filter with constant measurement noise covariance has been used for tracking objects however when occlusion happens, measurement noise covariance will be adapted by result of a local template matching in which correlation coefficients method has been employed. Experimental results confirm the efficiency and robustness of proposed method for MOT and occlusion reasoning.