Research On UAV Tracking Algorithms Based On Adaptive Space Regularization

Video object tracking technology has been applied to all aspects of life,such as video surveillance,autonomous driving,drone tracking,etc.Nevertheless,due to the complexity of various video scenes and the different requirements for tracking in different industries,there is still no object tracking algorithm that can adapt to various complex scenes,such as object occlusion,scale change,similar object interference,and so on.Aiming at some problems existing in the UAV tracking algorithm under the relevant filtering framework,this thesis improves three aspects: objective function,template update method,and feature extraction.The main improvements are as follows:(1)Improve the objective function.An adaptive spatial regularization method is proposed.In this method,an adaptive space regular term is introduced into the objective function,so that the algorithm realizes the adaptive update of the filter coefficients by learning the effective space weights of specific objects and their appearance changes during the filter training process.At the same time,the method can use the alternating direction multiplier method ADMM for optimization,and each sub-problem has a closed solution.(2)Improvements in template updating.A new template update method is proposed to judge the reliability of templates during tracking.This method is different from the method of directly performing linear update without usually performing reliability judgment.In the actual tracking process,the template is updated only when it meets the update conditions,otherwise it is not updated.(3)Improvement in feature extraction.An object tracking method is proposed based on multi-feature fusion.During the training process of the filter,the hand-crafted feature HOG is extracted for the scale estimation of the object.The extracted deep features VGG-16,VGG-M,and handcrafted features HOG are fused with a certain strategy to estimate the position of the object.Finally,the scale and position information of the object is obtained.In the experimental part,this thesis conducts a large number of experimental analyses based on two benchmark test sets of UAV123@10fps and OTB-100.The results show that the tracking algorithm based on adaptive space regularization proposed in this paper effectively improves the performance of the tracking algorithm in challenging environments such as scale change,similar object interference,and object occlusion;The proposed adaptive spatial regularization tracking algorithm based on multi-feature fusion has better tracking performance compared with several mainstream algorithms in challenging environments such as fast motion,illumination changes,and background clutter.