UAV Tracking Based On Correlation Filter

As a hot issue in the field of computer vision,the development of visual tracking is in full swing.It has extensive application value and development prospects in agricultural monitoring,traffic monitoring,power inspection,film and television drama shooting and post-earthquake rescue.Problems such as illumination changes,local or global occlusion,and target deformation in visual tracking are classic problems in the field of target tracking.The target tracking of drones also causes viewpoint changes and camera movement due to low flying height and wide flight range.After the emergence of tracking difficulties,although domestic and foreign researchers have been working on a more robust and accurate tracking algorithm,the target tracking technology has also been greatly developed,but so far,there is still no one can solve various tracking algorithm for the problem appears.Based on the classical correlation filtering algorithm in visual tracking,this paper proposes some improvements from the apparent features and robust image representation,and has obtained some research results.The main contributions are as follows:In order to solve the occlusion loss problem in the target tracking process of the UAV,this paper proposes a multi-feature re-detection target-related filter tracking method for multi-feature re-detection based on the classical correlation filtering algorithm KCF.Multi-feature target fusion and re-detection when the target is occluded to mitigate the influence of local or global occlusion,so as to improve the tracking accuracy and robustness of the tracker.Illumination change is one of the problems often encountered in the process of UAV target tracking.In order to alleviate the influence of illumination changes on UAV target tracking,this paper proposes a quaternion Fourier based on correlation filtering algorithm.The saliency detection of the UAV target-related filter tracking method reduces tracking consumption by processing adjacent frame targets.This simple and effective method is used to detect the target's effective tracking by quaternion Fourier saliency detection to ensure the robustness and stability of the tracker.