An Efficient Object Tracking Algorithm based on Dynamic Particle Filter

Abstract

The tracking problem of the non-rigid object has two difficulties. First, the object can change his scale, form or be partially occluded with other objects and, second, the location of the object can vary in the non-linear manner during the time. The paper considers the method, which solves given problem of the object tracking in the complex background. The solution of the mentioned problem is solved using dynamic particle filter (DPF), where the amount of the particles depends on the changing rate of the object location. The algorithm of dynamic particle filter was formalized using formal DPLA (dynamic PLA) notation. The object tracking is done according his color features (HSV histogram), which are independent from the object scale and/or form. Proposed method is demonstrated in the context of tracking real objects in the video data. The results shows, that the object tracking rate depends on the amount of generated particles and the tracking error. Ill. 8, bibl. 15 (in English; summaries in English, Russian and Lithuanian).