Research And Application Of Human Detection And Tracking Technology Based On Attribute And Grammar

With the development of mobile Internet and machine vision technology,dynamic human detection and tracking technology has become a research hotspot,which faces many technical challenges.For example,there are differences in clothing for different human targets;even for specific individuals,because of the uncertainty of human motion and the diversity of human posture,the shapes of clothing are different at different times;uncertainties(poor lighting,target occlusion,background clutter,shadow,motion blur,etc.)can affect people.The results of human body detection and tracking make it difficult for the current human body detection and tracking algorithm to meet the real-time and accurate requirements of the monitoring system.In view of the above challenges faced by dynamic human body detection and tracking,the following work has been carried out:(1)Aiming at the problems of human body shape change(such as non-rigid deformation of human body,dress change,etc.)and target occlusion,the dependence relationship and component attributes of human body shape components and components are modeled and described by introducing Attribute and Or Grammar(Attribute and Or Grammar),so as to provide a basis for extracting dynamic human body features.Support.(2)A human detection algorithm based on A-AOG is proposed.The algorithm consists of a human component region Proposal Network(HC-RPN),an Attribute and Position Predicte Network(A-PPN),and a human form-based maximum suppression algorithm HF-NMS(Human Form-Non Maximum Suppression).? The algorithm optimizes the human body detection algorithm according to the human body morphological characteristics,so as to solve the problem of missing detection caused by human occlusion and motion blurring.(3)A dynamic human tracking framework based on A-AOG is proposed.The detection algorithm and tracking algorithm are multi-modal adjusted and fused to solve the problem of target loss in long-time target tracking.In addition,a human attribute matching algorithm is constructed to solve the problem of target drift caused by occlusion.(4)The above A-AOG human detection technology and tracking framework are applied to the bus passenger flow automatic statistics system.The system realizes the automatic statistics of passengers on and off buses by analyzing the video of passengers on and off buses taken by bus cameras.The system experiment proves that the accuracy of the bus passenger automatic statistics system reaches 93.5%.It proves that the dynamic human body detection and tracking technology proposed in this paper can adapt to complex scenes well.It has strong robustness to complex scenes such as human shape change and human body occlusion.It has good application value.