Research On Human Motion Analysis Algorithm In Motion Scene

With the development and application of computer technology,artificial intelligence technology has been widely used in sports training.As a whole-body exercise,skipping rope plays an increasingly important role in daily life,and has been included in the project of the high school entry examination.At present,in the skipping rope teaching,the lack of professional sports teachers leads to inefficient training of students;the skipping rope monitoring equipment is bulky and expensive,and the cost and energy of manual statistics are too large.A method to automatically and quickly analyze whether the movements in the process of skipping conform to the norms,and give correct guidance and training plan have become the key to improve the performance of skipping rope test.The existing vision-based human motion behavior recognition and analysis algorithms have high complexity,poor robustness,and heavy calculation burden,and cannot be real-time.In addition,due to the lack of professional human movement analysis personnel,the research on human movement analysis and exercise quality evaluation needs further exploration.Therefore,in the process of human body movement,it is of great significance to implement a motion analysis method with high robustness and stable time cost for motion analysis and motion quality evaluation during the motion.This thesis addresses the problem of motion analysis during the skipping process.First,the key point coordinates in the skipping process are obtained through a 2D posture estimation algorithm,and the coordinates are pre-processed to obtain a robust data sequence.Then,the problem of motion analysis in the process of skipping is transformed into a multi-label classification problem.A new multi-label classification model ALSTM-LSTM is proposed.Finally,a skipping motion analysis system for mobile phone is designed and developed.The main research contents are as follows:(1)A deep learning framework is proposed to obtain the coordinates of key points in the process of skipping.The framework is based on Open Pose.In the aspect of feature extraction,in order to reduce the parameters of the model and improve the efficiency of recognition,lightweight network model Mobile Net V2 is used to replace VGG19.For the network optimization,in order to further improve the accuracy and generalization ability of the Open Pose model,weights and penalties are introduced in the loss function,and the validity of the model is verified on the MPII data set.Finally Through different type phones,video streaming data of one minute skipping of athletes is obtained,and the body's key point coordinates is collected by improving Open Pose method,the obtained coordinates is pre-processed,i.e.,Cartesian coordinate system is set up based on the center of gravity of the coordinates of the three key points Of the neck,left hip,and right hip,and update all coordinates to get the human body key point sequence.(2)Based on the obtained key point coordinates,this thesis converts the six types of limb movement standards in the skipping process into multi-label classification problems,and proposes a multi-label classification model ALSTM-LSTM according to the algorithm adaptive method in multi-label learning.The experimental results obtained 95.1% accuracy,higher than other deep learning models,and the evaluation indicators such as accuracy,recall rate,and F1-score are also optimal.(3)The intelligent rope skipping teaching system of android terminal is designed and developed by using the network model.The system mainly analyzes the problems existing in the process of rope skipping,which can help the rope skipper to correct the wrong actions in the process of rope skipping and improve the students' rope skipping performance.