Dataacquisition And Modeling Of Human Motion Based On Kinect

The combination of motion capture data and physical simulation provides an important technical approach for the research of human motion characteristics and the development of humanoid robots.Based on Kinect and ODE physics engine,this paper designs and implements a series of methods for human motion data capture,data conversion,motion modeling,motion reproduction of virtual humanoid robots and quantitative evaluation.Firstly,the basic work of human motion data capture,data conversion and data storage is completed.Based on human motion data collected by Kinect,a transformation model and algorithm from Kinect skeleton data to human posture data are constructed.The motion data are stored in BVH files to provides support for data analysis and data drive.Secondly,In the captured database,the PCA clustering algorithm is used to extract the key pose data with clear semantic and model them as motion nodes.Based on this,a motion graph structure is proposed and designed to realize the transition of motion sequences between different motion nodes.The transition creates a sequence of human motion that conforms to the application scene and improves the utilization of the motion capture data.Thirdly,a full-scale virtual humanoid robot is built based on ODE,which provides a verification platform for the simulation.Converting the motion capture data to a corresponding joint torque input to realize the motion reproduction of the virtual robot.And the real-time calculation and control of joint pose data to drive torque is realized,and the input of joint torque is controlled by PD control algorithm.Finally,by going through Kinect data acquisition,off-line BVH conversion,position and pose calculation,joint torque calculation,simulation and drive,the virtual humanoid robot achieves an accurate reproduction of human body action.The system is applied to the development of a prototype of a lightweight humanoid robot arm.Human motion is evaluated by defining motion similarity evaluation indicators and using interactive verification methods such as physical simulation.In summary,this paper completes the human motion data collection and modeling based on Kinect,drives the virtual robot to simulate the motion capture data,and establishes the evaluation index to quantitatively evaluate the human motion.The practice shows that this series of methods can accurately realize the learning and reproduction of human body movements(waving,walking,standing,etc.)by humanoid robot,and can be used in the process of learning,planning and bionic control in the development of humanoid robot.