Research On Object Tracking Based On Spatio-Temporal Context Algorithm

Target tracking is one of the most popular research topics in the field of computer vision.It has a wide range of applications and future prospects in such as intelligent visual surveillance,human-computer interaction,intelligent transportation,video indexing,medical diagnosis,image compression and vehicle navigation and so on.In recent years,although many excellent target tracking algorithms have been proposed in the field of target tracking,designing a target tracking algorithm that can adapt to any complicated scene remains a huge challenge.In this thesis,we analyze the basic theory of target tracking algorithm at home and abroad,and focus on the target tracking algorithm of spatio-temporal context.The advantage of the target tracking algorithm of spatio-temporal context is that the temporal contextual information of the object and the spatial contextual information of the object are used together to track the target more accurately in the face of illumination changes and target rotation,and the algorithm is real-time and robust.However,when the target occludes and the target scale changes continuously in a complex background,the target and its local context area feature information will change one after another,which causes tracking drift and tracking failure.As target tracking via spatio-temporal context learning easily fails to track target stably when target is in the occlusion condition and complex background,a target tracking algorithm for spatio-temporal context based on particle filter is proposed.By setting experimental parameters,the rectangle area of the first frame target is selected automatically.During following frame tracking,the Bhattacharyya coefficient is used as a judgement criterion for occlusion.When the target is occluded,the particle filter is used to estimate and predict position and trajectory of target in the subsequent frame,and the target can be traced accurately.The experimental results show that the proposed algorithm can not only be applied to target tracking under complex conditions such as illumination changes,target rotation and background disturbance,but also has robustness to target occlusion and satisfies the requirements of real-time tracking.Aiming at the problem that the tracking via spatio-temporal context has low tracking accuracy due to the changing of the target scale,an adaptive scale target tracking algorithm for spatio-temporal context is proposed.Firstly,the color histogram and HOG feature of target template are extracted.Then,learning online by using the spatio-temporal context model to obtain the most probable location of the confidence map.Finally,according to the updated scale method adaptive track the subsequent frame target to achieve the best tracking effect.The experimental results show that the proposed algorithm not only shows advantages in illumination change,background interference,target rotation,but also performs best when the target scale changes constantly and achieves the target robust tracking.