Research On The Key Technology Of Target Anti-occlusion Tracking Based On Multi-algorithm Fusion Of Spatio-temporal Context

Target tracking technology is one of the key research topics in the field of computer vision,there are numerous environmental factors are uncontrollable during actual exercise,tracking techniques of challenging conditions such as global occlusion and sudden light changes still need constant exploration and development,solving occlusion alone involves many disciplines such as machine learning,pattern recognition,and deep learning.Currently able to track technology effectively still researching and improving.The STC algorithm is one of the fastest and most stable trackers in the tracking algorithm,however,there has not been a specific tracking recovery mechanism for occlusion,tracking drift often occurs.TLD framework is the earliest and most effective long-term single-target tracking algorithm,however,in the case of rigid deformation,global occlusion,cannot effectively track goals,especially in the process of occlusion of the target,the output window is over-tuned.This article addresses the above issues and proposes the following solutions:(1)An adaptive learning spatio-temporal context algorithm is proposed,solve the problem that tracking's window not change for a long time lead to learning space context model does not have a targeted problem when the target scale change,using the characteristics of the SIFT algorithm,adjustment window after mismatching ad matching acceleration,at the same time,it also improved the learning mechanism of the space-time model.Specific learning of target information helps to improve the robustness of the tracking algorithm.(2)The algorithm mentioned in(1)improves the occlusion capability,However,there is no specific anti-blocking mechanism.Therefore,a temporal and spatial context anti-occlusion tracking algorithm for adaptive target change is proposed.Integrate the detector of the TLD framework into the STC algorithm,presenting a sports similar's method to solve the problem of excessive window adjustment,at the same time,it has also added an adaptive learning processing method.(3)Trackers of traditional TLD algorithms are susceptible to light and require that the target motion has a smooth,that is,the median flow tracker of the traditional TLD algorithm is not as good as the STC algorithm.So presents a TLD object tracking algorithm based on spatio-temporal context similarity algorithm(TLD-STC),the validity of the tracker does not only adopt conservative similarity,also proposed motion similarity and target space-time model,solve the problem that after the resolver fails,the detector detects multiple calculation results and does not output problems,integrate the spatiotemporal context into the TLD framework.Through experimental evaluation,the improved algorithm presented in this paper solves the lack of anti-blocking ability of STC tracker and adding the idea of spacetime context to the TLD framework,improve the stability of its tracking and antijamming.