Object Tracking in Satellite Videos Based on Improved Correlation Filters

Abstract

With the continuous progress of the society, video satellite has been paid more and more attention. As a new type of earth observation satellite, it can observe the certain area and obtain more and more dynamic information than traditional satellites. In this paper, we propose a novel algorithm based on the improved kernel correlation filters (KCF) to track the object in satellite videos. The improvements are as follows: 1) fusing the different features of the object, whose purpose is to describe the object information more effectively and reduce the impact of moving scene changes in object tracking and 2) proposing a motion position compensation algorithm through combines Kalman filter and motion trajectory. Its purpose is to improve the effectiveness of object tracking and avoid tracking error when used alone. What is more important is that it can also solve the problem of tracking failure when the object is partially or completely occluded and 3) extracting the local object region for normalized cross-correlation matching, and its function is to improve the success rate and accuracy of object tracking. The experimental results show that our algorithm can more effectively track the moving object in satellite video with high accuracy.