Research And Application Of Service Robot Action And Intention Recognition System In Home Environment

With the rapid advancement of robot technology and the continuous improvement of economic living standards,it has become an inevitable trend for robots to enter human daily life.Based on the application requirements of home service robots and the related technologies of robots,the paper systematically studies the human activity and intention recognition technology and system applications in the home environment.This paper first elaborates the research and technical status of the human activity and intention recognition system of service robots in the home environment.Then,it analyzes the home environment activity and intention recognition system functional requirements,completes the system overall framework design and relevant functional software framework design,selects the software and hardware platform,and describes the basic operating principles of the system combined with specific examples.The paper elaborated three main modules of the system function software: human activity detection and recognition module,object detection and recognition module in the home environment and family intention recognition module based on activity and object.Firstly,dynamic bayesian mixture model is introduced to recognize human activity,which is based on the basic classifier and calculates the confidence of each classifier through the posterior probability,and dynamically adjusts the weight of each basic classifier to improve the recognition accuracy.By obtaining human activity image through Xtion depth vision sensor and human bone points through Open NI,the model calculates the position distance,angle and speed information as the input of the classifier to recognize human activity.At the same time,based on SIFT feature and depth information and introducing grid motion statistics to improve matching efficiency,preliminary recognition and location of objects are carried out.The extracted potential objects are classified by support vector machine through histogram feature vector of color domain.Then,the Markov decision process is introduced to identify and speculate the target person's intention behavior,which is combined with the target person's home living habits information.By fusing the obtained human activity information and home object information,researching and designing the intention speculation reward function,this method is able to speculatively identify potential probable intentional behaviors and adjust function values based on past speculative information to improve the accuracy of intentional recognition.Finally,combined with the actual requirements of the home service robots,this paper selects the delivery service task in the home environment to design a series of experiments about activity and intention recognition,task generation and confirmation,and task execution,which verifies the feasibility and effectiveness of the system.