Research On A Fusion Strategy Of Visual Memory Mechanism For Object Tracking

With rapid development of computer vision and artificial intelligence,People's lifestyles,as well as cities are becoming more and more intelligent.Recently,since intelligent video surveillance plays important roles in all kinds of smart city services,visual object tracking attracts more attention.However,current object tracking algorithms still face various challenges in practical applications because of the complex scenarios,such as target occlusion,illumination changes,fast motion,and changes in posture.In addition,the real-time requirement of the object tracking in practical applications is also an unsolved problem now.Therefore,how we solve the above challenges has become an important research topic in visual object tracking.The current object tracking algorithm still has a certain gap in performance with the human visual system.Therefore,this paper proposes a visual memory mechanism and introduces it into object tracking to improve the performance of object tracking algorithms by studying the mechanism of human visual system.The main research contents of this paper are as follows:(1)By simulating the object tracking process of human vision and analyzing the mechanism of human vision in dealing with object tracking in various complex scenes,the visual memory mechanism model is constructed and the related concepts of visual memory mechanism is proposed.(2)By integrating the constructed human visual memory mechanism,an object tracking algorithm framework is constructed.Then,the typical representatives of correlation filters and Tracking-by-detection tracking algorithms are selected to integrate the visual memory mechanism.(3)On the current mainstream visual object tracking algorithm evaluation datasets OTB2013 and OTB2015,two improved algorithms that incorporate the visual memory mechanism strategy are evaluated with current algorithms.The effectiveness of the proposed object tracking strategy which integrates visual memory mechanism is verified by qualitative and quantitative experimental analysis.