Research On Fast Detection And Recognition Of Human Body Based On Machine Vision

Human society is entering the era of robots from the information,Robotics has been widely used in industrial production and gradually played an important role in the service domains,Service robots have developed rapidly.There are many service robots that provide service in human-computer collaborative environment.Those service robots need to detect and locate human body,then approach people and service people.It can be seen that detecting and recognizing human targets quickly and efficiently is an important function of intelligent service robots.If the problem of rapid detecting and recognizing of human targets is solved,the service content of service robots will be significantly increased,and service quality and efficiency will also be significantly improved.However,human targets detection and recognition rate is affected by many factors,such as light intensity,target scale,spatial background.Using machine vision technology to effectively detect and identify human targets is still a challenging topic.The background of this paper is human-computer interactive service.It focuses on the research of human objects detection,recognition based on machine vision,exploring feasible visual guidance technology for man-machine interactive service robots.The dissertation systematically researched literature both on visual guidance robots and human objects recognition.To meet the typical demands of man-machine interactive service robots for human objects detection,recognition and positioning,a scheme is proposed,which composes of human objects detection under multi-source images combined with thermal infrared and visible light,identification and positioning.What's more,to improve the speed of collecting and processing images multi-source,a hardware platform based on Xilinx Zynq series heterogeneous multi-core processors has been successfully designed.The key technologies of this paper include human objects rapid detection based on multi-source image,recognition based on infrared gait features,and monocular visual positioning.The advantages are as follows.Firstly,multi-source image and SVM are combined in series to detect human head and upright human.It is rapid,well-adapted and has high robustness.Secondly,the infrared gait feature recognition method fusing on score level is proposed,which is based on feature vectors composed of multi-region and skeleton dynamic features.Also,the visible light face recognition technology based on OpenFace is involved to improve the adaptability of personnel identification for robots in the service scene significantly.Thirdly,positioning of the human objects in monocular visual positioning has the advantage of simple structure.This paper describes the design and implementation process of the hardware platform and software and shows the results of experiments.Firstly,in simple scenes without obstructions,the average speed of the recognition is up to 32ms by detecting the vertical features of the upright human body directly in the self-built database and the detection rate is 91%in the database of 10^-3FPPW.Secondly,in complex scenes,the average speed is 19ms by detecting head features from multi-source images and the detection rate is up to 90%in 10^-3 FPPW.Finally,the face recognition rate is 97%in the LFW database and the gait recognition rate is up to 86%in the self-built database.The experimental results show that the scheme proposed in this paper includes multi-source human body objects detection combined with visible light and thermal infrared,recognition and positioning is rapid and valid in indoor scenes.The research results provide an effective method for human objects detection and recognition for human-robot interactive service robots.Last but not the least,it has a good promotion effect for the practical application of visual guidance service robots.