Study On Pedestrian Perception System Of Security Patrol Robot

As a kind of robot,the security patrol robot is designed to assist video monitoring and security personnel to carry out daily security tasks.It's movement characteristic can greatly reduce blind spots of inspection and improve the quality of security.With the rapid development of artificial intelligence and computer vision technology in recent years,the intelligent perception technology of security patrol robot has put forward higher requirements.In order to get location and identification of the pedestrian target,the pedestrian perception system of the security patrol robot is constructed.In this paper,the pedestrian detection and pedestrian tracking and pedestrian identification are studied and applied in security patrol robot.The main contents of this paper are as follows:(1)Study the pedestrian detection technology in security patrol robot.In order to meet the real-time and accurate requirements of pedestrian detection in security patrol robot,YOLOv3 target detection algorithm is used to realize pedestrian detection during security patrol robot patrolling.For improving pedestrian detection speed,MobileNet is used to replace feature extraction network in YOLOv3.Beside,in order to make YOLOv3 more suitable for pedestrian detection,this paper makes some adjustments to the initial value of the anchors in YOLOv3.(2)Study the multi-pedestrian target tracking in security patrol robot.In this paper,a multi-pedestrian target tracker of security patrol robot is designed by adding DeepSORT algorithm in pedestrian detection to get pedestrian tracks.Compared with pedestrian detection,Deep-SORT algorithm can utilize the correlation information between image sequence to estimate the position of the lost pedestrian.(3)Study pedestrian identification based on face in security patrol robot.Firstly,Haar-like feature and Adaboost algorithm are used to realize face detection of pedestrian in tracks.Secondly,Norm LDA is proposed to extract face features based on Fisher LDA.Thirdly,K-Subspace is proposed to cooperate with Norm LDA realizing the design of face recognition.Finally,the experiments are designed to demonstrates the effectiveness of face detection algorithm and face recognition model.(4)Build pedestrian visual perception system of security patrol robot.Firstly,the pedestrian visual perception system of security patrol robot is realized based on ROS,including hardware platform,software platform and software function implementation.Secondly,the workflow of the pedestrian visual perception system is defined and implemented according to the practical application scenario during patrolling.Finally,the experiment shows that the pedestrian perception system designed in this paper can be used in security patrol robot.