Twin network long-range tracking method for enhancing positioning capability

The invention provides a twin network long-range tracking method for enhancing positioning capability, which relates to the technical field of target tracking, and comprises the following steps: S1, obtaining a first frame of a target video, and cutting a labeled region of the first frame to obtain a target template; s2, performing feature extraction on the target template by using a target extraction branch of a twin network to obtain a target feature map; s3, obtaining a search region feature map; s4, sending the target feature map and the search region feature map into a region suggestion network to generate a series of regression frames and corresponding target scores; s5, selecting a regression frame region corresponding to the highest target score, and putting the regression frame region into a feature enhancement verifier to calculate a verification score; and S6, re-detecting the global picture in sequence by using a sliding window, and continuing to track the detected target. The algorithm is based o