Research And Implementation Of Gymnastic Action Recognition System Based On Kinect

The gymnastic action recognition system based on Kinect equipment,which uses Kinect to grasp the body contour and body position of athletes to determine the standard degree of athletes' movements accurately.Combined with the characteristics of sports activities,it provides effective solutions for sports training and improves the sports level of athletes.In this paper,we use the methods of data acquisition,data processing and feature extraction,and human posture recognition to assist action learning.We use Kinect to collect data and obtain bone information and position information of bone nodes in real time,and then reverse redirect the generated data stream to the animation model of the system to drive the Kinect application platform By comparing and analyzing the data of standard action,the standard degree of user's action is calculated in real time and the dislocation information is fed back to help users to find wrong actions easily as well as improve the efficiency and efficiency of learning actions effect.The main work of this paper is as follows:The system uses bone skin technology to build the character model,which renders and demonstrates the 3D character animation on unity platform.In the calculation of skin changes in human model movement,the biquaternion linear skin algorithm is used instead of the default linear skinning algorithm of the system,which makes the details of the motion of the character model more delicate;HMM(hidden Markovmodel)and ANN(Artificial Neural Network)are used to improve the detection algorithm of human body motion.Clustering based on static k-means algorithm is used to optimize the sports data collected by Kinect equipment,which significantly reduces the calculation error,combined with DTW algorithm to solve the problem of gymnastic posture feature sequence misalignment,the project improves the performance of the algorithm and the accuracy of tracking the human actions,reduces the load and makes more reasonable use of hardware resources;Aiming at the recognition of gymnastic movement by Kinect in real life,combined with the idea of dynamic time warping,a gymnastic movement recognition algorithm based on DTW is designed,which can effectively solve the matching problem of gymnastic movement sequence corresponding frame on the time axis;In order to better help gymnasts master the accuracy of gymnastic movements,the feature extraction of human skeleton data in the two dimensions of angle and speed is carried out respectively.The comparison between the gymnastic movement data of the trainer and the data of the standard gymnastic movement database is realized,and the score is obtained according to the action standard evaluation,so that the trainer can choose different gymnastic training items,At any time,the required action learning,and the dislocation information real-time feedback to the trainer,correct the error.The experimental results show that the system can better assist the trainer to learn gymnastics.