Basketball players' step action recognition based on two-model convolutional neural network

With the boom of national sports, basketball is widely popular as a competitive sport. In basketball, footwork is particularly important. The transition of footwork creates a corresponding movement trajectory, and the analysis of both helps to improve the skill level of players. In order to identify the player's stride, this study uses a smart insole containing sensors to extract the angular velocity and acceleration values of the stride and perform normalized preprocessing, and then operates the values according to the proposed Double Model Convolutional Neural Network (DMCNN) algorithm, and the advanced features of the steps are extracted to complete the recognition of the steps. Also, this study depicts the player's motion trajectory based on the quadratic method and analyzes it. The experimental results show that the DMCNN algorithm is 100% accurate in recognizing all five steps and converges to a stable state at 28 iterations with a final stable fitness value of 0.15. The velocity and acceleration values transmitted by the sensors are highly consistent with the real values, and the trajectories depicted basically coincide with the real paths. Therefore, this study proves the effectiveness of the DMCNN algorithm in step recognition and the accuracy of the depicted motion trajectory.