Research On Video Target Tracking Algorithms In Complex Scenes

As a basic problem in the field of computer vision,target tracking has gained widespread attention in the past decade.At present,it has a large number of applications in many fields such as video surveillance and human-computer interaction.Its core problem is to estimate the position and scale of the target in the unknown state of the subsequent frame after the bounding box of the target object of the first frame is given,thereby realizing the process of target tracking.However,in realistic complex scenes,there are many inestimable interference factors,such as camera shake,out-of-field,severe occlusion,deformation,fast motion,and other internal or external interference.Therefore,there is still a long way to go to design a target tracking algorithm that is accurate,robust,and resistant to multiple interference factors.In this paper,the problem of improving the accuracy and robustness of the video target tracking algorithm in complex scenarios is deeply studied and analyzed.The main research results are as follows:(1)To address the problem of target deformation and poor robustness of tracking in complex environment,a correlation filter tracking algorithm fusing image saliency via sparse reconstruction is proposed.In the process of target tracking,background template is extracted by superpixel segmentation.Target color correlation is obtained based on sparse reconstruction.Then the correlation filter detection score is combined with the target color detection score for accurate tracking.Template update speed is adjusted by the peak sidelobe ratio which is based on the fused detection score.Meanwhile,a center prior is established to correct the sparse reconstruction based saliency map.The proposed target tracking framework can adapt to deformation,illumination and other complexities.Experiments show that this algorithm is superior to other state-of-art tracking algorithms in terms of accuracy and robustness.(2)Aiming at the failure of existing hierarchical convolutional features for visual tracking algorithm in complex environments,an adaptive object tracking algorithm based on spatial attention mechanism is proposed.According to the color histogram of the current frame,the spatial attention mechanism is established based on the Bayesian classifier.After extracting multi-layer convolutional features in VGGNet19,the spatial attention map is fused with convolutional features respectively to construct more robust target apparent models.The response is obtained by using the correlation filter,and the final response is achieved by the weighted summation criterion.The adaptive update of the filter template is implemented by using the frame difference method to adjust the learning rate in the tracking process.The experimental results show that the tracking accuracy and robustness of the proposed algorithm are better than the existing state-of-the-art tracking algorithms in most complex environments.(3)Aiming to solve the problems that the redundant channel in depth features affects the tracking speed and accuracy,and the single feature can not adapt to all complex scenes,a multi-feature combined target tracking algorithm based on channel clipping is proposed.Firstly,the traditional manual features and depth features are combined for tracking.Secondly,a channel clipping strategy is designed by comparing the deep feature means of each channel in the target and searching areas.Then,the depth feature is updated by the proposed interval updating strategy,and the traditional feature filter template is updated by the average peak correlation energy.Compared with the 10 algorithms on the OTB2013 and OTB2015 datasets,the results show that the proposed algorithm achieves better results in tracking accuracy and success rate.