Research On Human-robot Intelligent Interaction Technology Based On Visual Perception Of Service Robot

With the aging of the population,the increase of labor cost and the upgrade of service demand,robots are gradually coming into people's life,and the interaction scenes between human and robots are becoming more and more abundant.In addition,now the computing power is getting stronger and stronger,and the data scale has accumulated to a certain scale.These technological foundations have promoted the breakthrough of artificial intelligence technology,and artificial intelligence technology has also been applied in human-robot interaction on a large scale.As an important technology in the field of artificial intelligence,computer vision plays an important role in human-robot interaction.Therefore,this paper studies the intelligent human-machine interaction technology of service robot based on vision,and realizes that the robot can understand the human state through vision,and serve people according to the human state.The main research contents of this paper include:(1)Object detection based on vision.Object detection is based on the premise of visual human-robot interaction.In this paper,Mobilenet-SSD is selected as the combined object detection method,which has the advantages of lightweight,real-time and high efficiency.In addition,it provides a method of YOLO-LITE that can complete real-time detection without GPU.The two methods both have their own advantages and can be selected according to the actual situation.(2)Human behavior recognition based on vision.Human behavior recognition is the base of visual human-robot interaction.For this part,SSD detector is firstly used to detect human body,then SORT tracker is used to track human body,and finally human behavior recognition is conducted by using the model obtained from C3D learning spatio-temporal characteristics.In addition,a method of human posture detection is provided to obtain more human posture information.(3)Recognise human status based on visual features.Human intention is a high-level feature.In this paper,a word bag model is established to realize the mapping from low-level visual features to high-level features,and then the state recognition of the accomplisher.Firstly,clustering algorithm is used to construct word bag codebook,and then support vector machine classifier is used to obtain human state.(4)Human-robot interaction based on NAO robot.In this paper,a human-robot interaction system is designed by using NAO,a humanoid robot platform,and a robot knowledge base based on application scenarios is established.The verification of human-robot interaction based on vision is completed,and the robot can understand the state of human beings through vision,so as to complete the interaction with human beings.