Research On Visual Tracking Algorithm Based On Spatio-temporal Context

Visual tracking is one of the hot issues in the field of computer vision.Its main task is to estimate the location of the target in the image sequences.It has been widely applied to various fields of human life,such as human-computer interaction,medical diagnosis and so on,and has received high attention of scholars at home and abroad.So far,numerous excellent tracking algorithms have been proposed.Spatio-temporal context tracking algorithm(STC)is one of these algorithms and it uses the spatio-temporal relationship between the target and its local context to track the target and transforms the tracking process into a process of finding the maximum value in the confidence map.In addition,Fourier transform is used to speed up the computation.However,due to the influence of strong interference such as illumination changes and occlusion,the performance of tracking is still not ideal in practical applications.To achieve efficient and robust tracking,the major work and innovations are as following:(1)To solve the problem that the model is prone to updating incorrectly in STC,a tracking algorithm based on spatio-temporal context and adaptive features(ASTC)is proposed in this paper.First,ASTC judges whether the target is interfered by the peak sharpness of the confidence map.And then ASTC adopts the perception hash algorithm and the block discrimination method to judge the interference type that the target is subjected to.Finally,ASTC adaptively sets the learning parameter of the model.What's more,ASTC uses a scale filter to adaptively adjust target scale and improves the accuracy of target size.ASTC algorithm effectively suppresses the error accumulation.It has better tracking effect and enhances the robustness.(2)To solve the problem that STC uses low-level grayscale features and is unable to retrieve the target after losing it in the case of strong interference,we propose a tracking algorithm based on particle filter and significant context(PF_CSTC).PF_CSTC constructs a significant context prior model by using the color histogram to solve the disadvantage that STC uses the low-level grayscale features.Then the wavelet transform image matching algorithm is used to analyze whether the target is missing.If the target has been lost,the particle filter algorithm is used as a correction algorithm to retrieve it.The PF_CSTC enhances the robustness and improves the tracking effect.(3)In this paper,we conduct experiments on the public datasets and compare the two improved algorithms with many existing algorithms.Extensive experimental results show that ASTC and PF_CSTC algorithm outperform the original STC algorithm and other state-of-the-art algorithms.What's more,the average success rate of ASTC and PF_CSTC is nearly 30% and 25% respectively higher than that of STC,and the central coordinate errors are decreased by 18.6 pixels and 36.4 pixels.In general,ASTC and PF_CSTC algorithm have better tracking effect and robustness.