A dynamically selecting model to track a moving object

Abstract

Based on the Lucas-Kanade optical flow method, a dynamically selecting model is proposed in this paper to track a moving object. This model is composed of an object model, a consistency constraint model, and a random sampling model. Based on the current image frame, the object model is used to calculate the relevant feature points for the next frame. The random sampling model is used to resample the feature points from the current frame and the relevant points on the next frame. By considering the consistency constraints in speed, direction and the object stability, the consistency constraint model is used to check the consistency for the feature points obtained by the object model, remove some unfit feature points, and add some new feature points obtained by the random sampling model. A stable tracking trajectory would be obtained during the tracking process using the proposed method. Simulation experiments comparing with traditional methods are conducted in different situations, such as illumination change, partial occlusion, high-speed motion, and object image with noise. The results show that the proposed object tracking method has better performance, which includes partial non-rigid object and varying velocity running object in real-time environment with robustness.