Traffic Police Command Action Recognition Method Based On Multi Feature Comprehensive Analysis

With the development of society and technology,traffic congestion has become a major issue in urban development.In order to solve the problem of urban traffic congestion,the appearance of driverless car brings new dawn.With the help of intelligent assistant driving system,driverless vehicle can effectively solve the issue of traffic congestion caused by most people.In the domain of rapid development of artificial intelligence and big data,human motion recognition technology based on machine vision has been widely used.However,it is a remarkable research on applying this technology to the driverless system and identifying traffic police command action with potential value.Based on this background,this paper mainly studies the traffic police command action recognition under the actual road conditions.In order to locate traffic police under the complex background of the actual road conditions,a variety of different color threshold voting methods are used to detect the location of traffic police.The color of different areas in the face and uniform of the traffic police is extracted and voted uniformly,respectively.The area of the traffic police is determined according to the rate of votes.Then in order to improve the maximum connected area,the image is filled with hole filling technology.The detection of the traffic police's moving area is realized by finding the location of the center of gravity of the traffic police.Finally,the original image sequence is cut and recombined based on the moving area.The new image sequence of traffic police reduces the interference in the complex background environment.In view of the actual road condition,the automatic driving system needs to identify eight commanding actions from four different orientations.The thirty-two traffic police command actions are modeled by the width of the traffic body,the height of the human body,the position of two arms,the rotation of the head and the number of moving steps,respectively.In addition,a comprehensive analysis method of the curve of the width and height of the human body is used to realize the motion segmentation of multiple command actions in the video.For each kind of video segment of traffic police command action,region growing,frame difference method,human skin color detection and other image processing methods are used to extract motion features.In addition,five feature change curves are obtained.Then,based on the training sample set,the curve features of each characteristic curve are extracted.In addition,the initial sample template is established with the model.After it,to build a more reasonable final sample template,we set the weight empirical value of each action's characteristic value through the experimental method according to the characteristics of each action.Finally,to realize the traffic police command action recognition,inputting the test video to match the template.