Spatial-temporal regularized UAV correlation filter based on adaptive filter template

Abstract

Due to its quick computation, the correlation filter (CF) is frequently employed in object-tracking activities. The occlusion and deformation of the tracking object during the CF tracking process will pollute the filter, which will cause the filter’s performance to degrade. This is especially obvious in unmanned aerial vehicle (UAV) tracking and increased interfered sequences will further aggravate the problem. In response to this problem, we use ARCF as the baseline and add a spatial-temporal regularization term based on the adaptive template. We also introduce a historical filter template to generate a more global adaptive template. The experimental results illustrate that our tracker has the most sophisticated capabilities compared to the other 8 SOTA trackers on the DTB70 benchmark.