Research Of Pedestrian Detection Algorithm For Reception Robot

With the rapid development of science and technology,artificial intelligence technology is widely used in various fields.The development of artificial intelligence technology has broadened the research ideas of reception robot system,and extended the application field of reception robot.As an important part of service robot,reception robot has wide application prospect and important research value.Vision is the main way of human access to external information,and robots are similar to human beings.In order to make reception robot provide better service for humans,it is very crucial for them to detect the characters through vision.Therefore this paper mainly studies the pedestrian detection algorithm in the reception robot vision system:(1)This paper studied the method of pedestrian detection for reception robot,analyzed the domestic and foreign research results of reception robot and pedestrian detection.Through studying popular pedestrian detection algorithms,the difficulties and key points of the pedestrian detection task in the real working scene are specified.Since the pedestrian detection of reception robots are featured with unfixed targets,unfixed background and real-time change of foreground scale,this paper uses HOG feature for pedestrian feature extraction,and uses Support Vector Machine(SVM)to classify pedestrians.(2)In this paper,the HOG features and SVM were researched,and the shortcomings of the traditional HOGSVM algorithm in pedestrian detection were pointed out.By analyzing the influence of cellsize,sample scale,sample shape and occlusion on the recognition rate,the cellsize parameter optimization and rules of setting sample parameter are given to satisfy the automatic setting of various parameters in different situations.(3)This paper studied the problem that the recognition rate is affected by the uncertainty of target scale.Then a multi-scale and multi-classifier parallel detection algorithm is proposed,which makes the target can be detected at different scales.(4)The proposed algorithm was validated in simulation and experiments by MATLAB.The experimental results showed that the multi-scale and multi-classifier parallel algorithm has higher detection accuracy in the environment with complex changes such as clothing,illumination and scale.The advantages and disadvantages of the pedestrian detection algorithm for reception robot were studied in this paper.The factors that influence the pedestrian detection accuracy include cellsize,sample size,sample scale and occlusion in the application of real scene were studied and the author provided some references for the pedestrian detection problem of reception robot.A large number of experiment results in the real scene demonstrated that the multi-scale and multi-classifier parallel detection algorithm can detect the moving targets better in the real scene under the complex conditions such as movement changing,illumination changing,angle changing and scale changing,and the recognition rate increased about 20%.The algorithm has a great value for the pedestrian detection problem of the reception robot.