Recognition technology of human body movement behavior in fitness exercise based on transfer learning

摘要

By the detection of human body movements in fitness exercise, the normative correction of human body movements in fitness exercise is carried out. Aiming at the problem that the current empirical analysis method is not accurate in detecting the movements of complex movements, a method for detecting human body movements in fitness exercise based on multi-scale feature decomposition is proposed. Firstly, the image collection and feature analysis of the complex action link behavior of the body-building movement are carried out under the multimedia vision. Then, Harris feature point detection and information screening method are used to enhance the collected image of the body-building movement, and the multi-scale morphological edge contour features of the image are extracted, thus realizing the feature point calibration and detection of the body-building movement behavior characteristics and improving the quantitative estimation accuracy of the movement behavior. Simulation results show that this method has a high probability of accurate detection and recognition, increases the extraction number of key information feature points of fitness human movement, and has a high guiding value for movement correction.