Feature Extraction-Based Fitness Characteristics and Kinesiology of Wushu Sanda Athletes in University Analysis

Abstract

This paper designs a feature extraction-based physical characteristics and kinematic analysis of collegiate martial arts Sanda athletes by profoundly studying the model of athletes' action feature extraction and constructing physical characteristics and kinematic model of collegiate martial arts Sanda athletes. This paper provides feature extraction category support by describing the data types of clustered objects, calculating behavioral operator distances, identifying multiple behaviors occurring in collegiate martial arts Sanda tactics, and delineating the correct thresholds for operator identification. A comparison experiment is designed to verify that the proposed method can improve the accuracy of feature extraction for athletes' technical and tactical use and is more applicable to the comprehensive assessment of athletes' behaviors. The underlying features of the athletes' motion technique forming image features are analyzed, the image features are regionally divided, the feature sets of different levels of developing actions are obtained, and the image features are allowed to map to the corresponding feature dimension space. The AdaBoost algorithm filters out the feature data of the athletes' action images that contribute most to the intelligent visual analysis. It is used as training samples for training and recognition to complete the extraction of athlete action image features. The action times of the five different phases of side kick up, knee lift, kick strike, and strike: recovery is 0.2986 s, 0.1819 s, 0.1322 s, 0.0708 s, and 0.9986 s, respectively, which reflects that the time of different action phases of athletes with extra body weight is more related to the athletes' weight, strength, flexibility, and coordination.