An improved spatio-temporal context tracking algorithm

Spatio-temporal context (STC) algorithm transforms the tracking process into a series of processes to find the extremum of the confidence map and fully uses the density context information around the target, which makes the algorithm rapidity and robustness. However, STC cannot deal with fast motion, motion blur and the rapid change of scale, which will cause the spatial model update error and result in the failure of the algorithm to accurately extract the target area. To deal with the problem, an improved spatio-temporal context algorithm is proposed in this paper. Firstly, the position prediction based on the target motion vector is introduced, the motion information of the target is fully taken into account to improve the accuracy of the STC algorithm in extracting the target current position. Secondly, the scale correlation filter is used to improve the STC algorithm, so that the algorithm can accurately and completely extract the target area. Finally, experiment results on public data set are provided to show the effectiveness and robustness of our proposed algorithm.