3D Visual Recognition Algorithm of Action Based on Behavior Tracking and Feature Fusion According SimpleMKL Algorithm

作者：Wen, Ding^*; Wenhui, Ma

来源：WIRELESS PERSONAL COMMUNICATIONS, 2023.

DOI：10.1007/s11277-023-10511-3

• SimpleMKL algorithm
• Behavior tracking
• Feature fusion
• Tangent space mapping

There are some problems in the traditional recognition methods of sports wrong action, such as the low accuracy of 3D visual recognition and the poor recognition effect. Therefore, this paper proposes a 3D visual recognition algorithm of sports wrong action based on behavior tracking and feature fusion. Taking the Euclidean group SE as the motion feature to express the action, the feature extraction of sports wrong action is realized according to the tangent space mapping; The background subtraction method is used to remove the 3D visual background noise of sports wrong actions, and the Gaussian distribution matching is used to complete the classification of 3D visual features; According to the SimpleMKL algorithm, the optimal parameters of the objective function of sports wrong action are solved. Combined with the results, the 3D visual recognition of sports wrong action is realized by using the behavior tracking method. The experimental results show that the proposed method can improve the accuracy of error action recognition, and has a good effect of motion error action recognition.