Long-term stable target tracking algorithm based on improved Staple

Staple algorithm has been widely proven to be an efficient target tracking method, but it is still easy to fail in some tracking scenarios such as target deformation, motion blur, and complete occlusion. This paper makes improvements to the above problems. First of all, we use a binary mask based on spatial reliability to enhance the target information in Staple's color features, which improves the tracking accuracy of the algorithm in complex scenes. Secondly, we propose a response graph evaluation index based on secondary detection, that is, the least square filter is used for convolution at the original response peak to obtain a more accurate tracking state judgment. Finally, if the current state is judged to be a failure, we use a particle filter-based motion estimation method to relocate the target, thereby improving the algorithm's tracking success rate when the target is occluded. The test results on the OTB2015 data set show that the overall accuracy of the algorithm in this paper has reached 80%, and the overall success rate has reached 73.2%, which proves the long-term stable tracking performance of the algorithm.