Research On Human-computer Interaction And Motion Simulation Technology Based On Nao Robot

With the continuous development of robotics,human-computer interaction,as a branch of robotics technology,has also been greatly developed,and its integration with aerospace,medical education and services is becoming more and more high.By letting robots learn and understand human behavior,not only can the robot's intelligence level be increased,but in the near future there is hope to replace humans in a dangerous environment,which is of great research value.This paper is based on the development of human-computer interaction technology based on Nao robot.Using Kinect sensor to realize human motion capture,combined with extended Kalman filter and inverse kinematics to improve the effect of motion imitation,and then construct a convolutional neural network model for human motion recognition,and finally achieve a simultaneous human body imitation and motion recognition.The functional human-computer interaction system,the specific research content is as follows:(1)An improved Nao robot three-dimensional coordinate transformation algorithm is proposed to convert the coordinate data in three-dimensional space into the corresponding Nao robot joint angle.The traditional Nao robot space coordinate conversion algorithm inversely pushes the ? value by calculating the cos? value of the angle between the two joint vectors.This method has large calculation errors and is susceptible to external environmental noise interference.Aiming at these problems,this paper proposes an improved spatial transformation algorithm.The algorithm takes the extended Kalman filter as the basic framework and divides the process of spatial coordinate transformation into two steps: measurement and estimation.Firstly,the inverse kinematics and Newton iteration method are used to solve the problem of the joint coordinates of the known robot joints,and then the joint data obtained by the solution is input into the extended Kalman filter framework to establish the extended Kalman filter.Measurement model and prediction model.The output of the final prediction model is the robot joint angle data obtained after the space coordinate transformation algorithm.The algorithm improves the accuracy and anti-interference of the space conversion algorithm.(2)An improved VGGNet network model based on parallel network is proposed for Nao robot human body motion recognition.The traditional VGGNet network model improves the recognition accuracy by increasing the number of network layers.As the number of layers of the convolutional network deepens,both the parameter quantity and the training difficulty will increase greatly,and there may be a recognition accuracy rate that will decrease and “deteriorate” Phenomenon,etc.,through the feature fusion of two small networks,the final recognition accuracy is improved under the premise of minimizing the increase of the number of layers.(3)A human-computer interaction system based on Nao robot was built and implemented.Combine Kinect to write the motion capture upper computer program,track the skeleton structure of the action demonstrator located in front of Kinect and obtain the three-dimensional coordinate information of the human joint key points,and then output the human skeleton image to the screen through the OpenCV library.For the three-dimensional spatial coordinate data of human joints,the space transformation correlation algorithm of this paper is used to transform the three-dimensional space coordinates into the angle data of the Nao robot joint motion.Through the angle data,the robot motion is driven to realize the imitation of human motion.In order to improve the experience of the human-computer interaction system based on Nao robot,the human motion recognition function is added on the basis of human motion imitation.The convolutional neural network is used to identify the human skeleton image acquired by Kinect,so that the Nao robot can recognize the type of action made by the action demonstrator.