A Small Object Tracking Method in Satellite Videos Based on Improved Kernel Correlation Filter

The video satellite enjoys the ability that it can stare at the interesting area and record object information in the form of continuous video sequences, thus making it have significant advantages in change detection, object tracking and other ap-plications. However, objects in satellite videos generally account for a small proportion of the images and are easily affected by occlusion. As a result, most of the current algorithms can not achieve satisfactory results in satellite videos object tracking. To solve these problems, this paper proposes a satellite video object tracking method based on improved kernel correlation filter. The main innovations of our method are as follows: 1) The average peak correlation energy and peak value of the response map are used to judge whether the object is occluded or not, which provide the criterion for adjusting the update policy of correlation filter template; 2) The Kalman filter is utilized to estimate the motion status of the object and accurately predict the position when the object is occluded. Based on the improvement, the proposed method effectively solves the problem of tracking failure when the moving object is partially or completely occluded. The experimental results show that the method can track the moving objects in satellite videos with 97.7% accuracy, and the speed reaches 2000 frames per second.