Multiple Object Tracking Based on Motion Estimation and Structural constraints

Abstract

To solve the time-consuming problem and the low efficiency of the global exhaustive searching in the object tracking, this paper propose a new search strategy based on motion estimation and structural constraints. First, the motion vector of one object is calculated, associating with the location of the object in the previous frame, its moving direction and scope are predicted in the current frame. Then, with the combination of structural constraints between objects, the accurate search direction and scope of the other targets can be determined. We choose five videos for the experiment to confirm the superiority of the search algorithm in this paper. For each video, all these measurements are averaged over all objects, over all frames, and over five separate runs of the tracker. Experimental results show that the new search method can narrow the search range and enhance the searching efficiency under the condition of no affect on the tracking accuracy, thus the complexity of the multi-object tracking algorithm will be reduced.