Multi-object tracking based on multi-feature joint matching

In order to solve the occlusion problem in multi-object tracking for the complex background of a video image, an approach for the multi-object tracking based on multi-feature joint matching is presented. First, the adaptive Gaussian mixture background model is used for reconstructing and updating the background to achieve the background subtraction of current frame and multi-object detection. Then, the joint matching tracking is developed based on matching color characteristics, positions and objects velocities. Finally, the objects in the scene are divided into entering, exiting, temporarily disappear of the object, the re-emergence of the object and the merge and split of the object, and the predicted position and the occlusion factor of the object are used to improve the accuracy of multi-feature joint matching. Experimental results indicate that the similarities of the same object and the different objects are 0.94971 and 0.50573 respectively in the tracking with the proposed approach, which is better than that of matching with the color characteristics. Furthermore, the similarity of object is 0.97283 after occlusion. The approach is satisfactory for real-time tracking of multi-object with appearance similarity in a complex environment.