Research On Visual Tracking Based On Spatio-temporal Context Learning

An anti-occlusion visual tracking algorithm based on spatio-temporal context and a scale adaptive tracking algorithm based on color naming space are proposed to solve the tracking drift problem of visual tracking when target occlusion and scale variability.About research of anti-occlusion visual tracking based on STC,a spatial context model is established using spatio-temporal relationship between target and local context region for spatio-temporal context learning,estimating the target position by calculating the probability of the confidence map,the estimated target location region is divided into block occlusion,and discussing the threshold and occlusion probability.If the occlusion probability is less than the set threshold,the target has no occlusion,if the occlusion probability is greater than the set threshold,the target occlusion degree is further determined,and the block matching and particle filtering methods are chosen to carry out the occlusion target tracking.About research of scale adaptive visual tracking based on STC,the image is mapped to the color naming space to extract the feature of the color attribute,and the confidence map is estimated by the spatio-temporal context learning to estimate the target position,then the T-frame target region is calibrated,finally,three different size tracking windows are set in the T+1-frame.The color histogram information of two frames was extracted and the similarity matching was used to determine the best tracking window of T+1-frame and output as the tracking result.In the experimental verification,the eight groups of image sequences with occlusion and scale variability were selected,compared with the current popular algorithm,and the performance of the algorithm and the comparison algorithm were evaluated by tracking the success rate,central error and the operating rate of the algorithm.The experimental results show that based on the original algorithm,the STC-AO algorithm can better adapt to the visual tracking of the target with local occlusion and global occlusion than,and the STC-SA algorithm can better deal with the visual tracking with scale variability,which improves the universality of the spatio-temporal context visual tracking under guaranteeing the efficiency of the algorithm.