Research On Gymnastic Action Recognition Methods Based On Multiple Sensors

Standard gymnastic exercises can play a role in strengthening the body,but nonstandard gymnastic exercises can lead to poor exercise results.Therefore,effective identification and evaluation of gymnastic movements are the research focus of scientific training of gymnastics.At present,many universities and research institutions have achieved certain research results in the field of motion recognition,but the research on the combination of gymnastic movements and motion recognition technology is still in its infancy.Due to the complexity of gymnastics movements and the occlusion in the movement transformation process,the recognition progress has been slow.Based on full consideration of the above issues and detailed research and analysis on motion recognition,a set of gymnastic movement recognition system based on multiple sensors is designed to collect,record,and analyze gymnastic movement data in real time.The main research contents of this article are as follows:The three-axis acceleration and angular velocity raw data of the eleven nodes of the human body were collected by the gymnastic action recognition system,the raw data was preprocessed,and the sliding mean filtering method was used to eliminate glitches and other disturbances in the data curve.Based on the action decomposition of broadcast gymnastics as the standard and the experimental video as a reference,the preprocessed data was segmented.According to the characteristics of gymnastics and data segmentation,eight parameters such as standard deviation,information entropy,and mean square error were calculated as the classification characteristics of gymnastics data.Compared with machine learning algorithms such as K-nearest neighbors,Naive Bayes,and decision trees,the SVM algorithm has a higher recognition rate for gymnastic movement recognition.Aiming at the recognition effect of SVM algorithm mainly depends on the choice of kernel function and its low recognition rate,this paper proposes a gymnastic action recognition algorithm based on voting classification,and verifies that the algorithm has better recognition effect than SVM algorithm.By setting different stretching angles,kick heights,and audio double speeds as reference indicators,the purpose of quantitative evaluation of gymnastic movements is achieved.The experimental results show that compared with the SVM algorithm,the voting classification algorithm has better recognition effect on six kinds of gymnastics.The average recognition rate of three-axis acceleration each node is 96%by the voting classification,the average recognition rate of three-axis angular velocity each node is 96%by the voting classification.Based on the voting classification,the average recognition rate of three-axis acceleration and three-axis angular velocity are 98%,indicating that the algorithm has a higher recognition rate.