Object Tracking on Satellite Videos: A Correlation Filter-Based Tracking Method With Trajectory Correction by Kalman Filter

Object tracking toward satellite videos faces various challenges, such as small size of the moving object, few texture, background similarities, etc. In this article, we propose a high-speed correlation filter (CF)-based tracker for object tracking on satellite videos. It takes advantage of the global motion characteristics of the moving target in satellite videos to constrain the tracking process, which is achieved by applying a Kalman filter (KF) to correct the tracking trajectory of the moving target. Thus, our tracker is named CFKF. Besides, a tracking confidence module is designed to pass information from the CF-based position detector to the KF-based trajectory corrector, and a constant optimized model update frequency is studied to speed up the tracker, as well as improve its performance. Furthermore, the target's orientation during the tracking process can be obtained by utilizing an orientation detector based on slope calculation. Experiments conducted on the satellite video datasets demonstrate that our tracker CFKF outperforms the other representative CF-based tracking methods in terms of accuracy and robustness and is also fast in speed.

I. Introduction

Object tracking aims at automatically obtaining the target's position in the whole video, given the ground-truth (GT) of the first frame. It is one of the most heated topics in computer vision area due to its wide range of applications, such as automated surveillance, human–computer interactions, and motion-based recognition [47].