An improved spatio-temporal context tracking algorithm combining LK optical flow
Zhongjian Li, Yaru Su, Wenke Ma · 2017
Recently, the issue of visual tracking has been brought into a focus of public attention. In this paper, we present the improved spatio-temporal context (STC) tracking algorithm combining LK optical flow. Original STC algorithm estimates the best object location by maximizing the confidence map that integrates the spatio-temporal context information. Nevertheless, original STC algorithm only considered positional relationship between the pixels in the context region and object, without considering relative motion and direction. To enhance the accuracy and stability of the tracking, improved STC algorithm applies the idea of partition processing to the context region. Furthermore, The LK optical flow based on Harris corner is utilized to obtain the corresponding weight matrix and integrate it into the Bayesian tracking frame. In the experimental phase, we use the Unmanned Aerial Vehicle (UAV) as a tracking object. Numerous comparative experiments demonstrate that the improved STC algorithm can achieve fast, accurate and continuous tracking for UAV scale and angle variation, rapid flight, complex background changes and occlusion. Overall, it is obvious that the robustness and accuracy of the improved STC algorithm are better than before by comparing the tracking overlap rate and success rate.