Research And Implementation Of Human Behavior Analysis Based On Spatial-temporal Relationship

The development of transportation brings convenience to people,however,traffic accidents caused by illegal driving also happen frequently.With the development of intelligent monitoring,more and more buses have installed monitoring to detect the illegal driving.But it is time-consuming,labor-intensive and difficult to fully process large volume data by manual screening of surveillance video.Facing the low efficiency and low accuracy of manual and traditional algorithms to deal with large-scale data,intelligent methods for driver behavior recognition have been proposed and become increasingly important.Therefore,it is of great significance to study the behavior recognition based on surveillance video.This thesis mainly focuses on the problems of behavior analysis methods based on spatial-temporal relation and proposes a behavior recognition model based on the long-time spatial-temporal relationship.The research content of this thesis is as follows:(1)A behavior recognition method based on spatial-temporal features of skeleton image representation is proposed.Long-time behavior recognition faces the difficulty of fully exploiting human skeleton spatial-temporal information.To address this problem,a new method based on existing algorithms is proposed to improve the skeleton’s image representation method and the feature extraction method,and then the behavior recognition model is constructed.(2)A behavior recognition method based on the spatial-temporal relationship(STRM)is proposed.Combined with the new proposed behavior recognition method,a feature extraction method for high-rise structures is added as an assistant.In this feature extraction method,the human skeleton was divided based on connection and the features of these parts are used to build the graph to describe the behavior.The feature of the graph is extracted to assist the behavior recognition task.By this way,the STRM can recognize behavior by the part information in the long-time skeleton and the small part high-level structure information.It takes into account both the characteristics of human body and the coordination between small parts.(3)An illegal driving behavior detection system is implemented.In order to verify the effectiveness of the STRM,this thesis designs and completes the illegal driving behavior detection system in the bus scene and achieves the detection of drivers’ illegal driving behaviors.