Aerobics action recognition based on graph neural network and D-S evidence reasoning

Aiming at the problem that the existing methods of aerobics action recognition cannot dig deep information such as movement and body features, a novel aerobics action recognition based on graph neural network and D-S evidence reasoning is proposed. This algorithm uses the aerobics movement history and body information, combines with the advantages of multi-feature fusion and graph neural network, through modeling the relationship between body and movement, to obtain the influence degree of body information on different types of aerobics movements, as well as the long-term and short-term nature of historical movements. By combining the least square support vector machine and D-S evidence theory, the motion acceleration data and image coordinate data are fused to input the fused motion data. In the experiment, the performance of this new method and other advanced methods in aerobics movement recognition task is compared. The results show that the new method has obvious improvement in accuracy, precision, recall and F-value, which proves the effectiveness of the proposed algorithm.