Design And Implementation Of Motion Capture And Recognition System Based On Embedded System

The technology of human motion capture and recognition is a research hotspot in recent years.The human motion capture system has been applied in the fields of security,medical checkup,communication and animation,etc.In recent years,the technology has also been widely applied in the filed of robotics with the development of robot.In this paper,the human motion capture system is to achieve human-computer interaction for humanoid robot platform.Human motion capture systems are mainly base on computer vision or MEMS sensor.The system of human motion capture achieved in this paper is based on MEMS sensor.As for the motion recognition,some researchers recognize gesture by extracting motion features from dynamic motion images.However,due to the diversity and complexity of human motion,dynamic motion recognition based on 2D images has some limitations.In this paper,we achieve motion recognition using 3D motion trajectory.The information provided by 3D trajectory is more comprehensive,making the motion recognition more robust.Meanwhile,we implement the above two systems on the same embedded platform,which can complete motion capture and recognition in real time.The motion capture and recognition system achieved on embedded platforms can widely used in many situations.First of all,combining with specific requirements,we analyze the types of error in motion capture system.The algorithms for nine-axis inertial motion sensor error processing is set up and this make the results more accurate.Secondly,to display and store human motion data more conveniently,we design the method of converting the position of joints to BVH file format.Moreover,this paper introduces the method of displaying BVH file format.Thirdly,we propose a trajectory segmentation framework with the ability of segmenting trajectories quickly.we choose IID as the 3D motion trajectories representation in this paper,it is robust to local noise and can be calculated easily.Fourthly,for faster online recognition speed on embedded systems,we introduce a specific transformation of data dependence by using prefix computations.Further,we propose an efficient parallel DTW algorithm.Finally,we introduce the hardware and software of the system and the system is tested in the real environment.The results show that the designed system has the function of achieving the initial goal.