The Computer Vision Simulation of Athlete’s Wrong Actions Recognition Model Based on Artificial Intelligence

At present, in basketball teaching in China, the traditional basketball training method is for coaches to communicate face-to-face with athletes, observe their basketball movements, and judge the correctness of the movements based on the coach’s personal experience. However, this method mainly relies on the subjective judgment of the coach and lacks objective evaluation of athletes, making it impossible to objectively evaluate their performance. This article mainly studies an athlete’s incorrect action recognition model based on artificial intelligence algorithms and computer vision, and constructs an athlete’s incorrect action recognition model based on a dual channel 3D convolutional neural network (CNN). In this project, spatial attention mechanism (SA) was introduced into 3D CNN. By using inter frame difference information that can represent significant changes in athletes’ motion status, and combining it with grayscale video data, accurate recognition of athletes’ incorrect actions was achieved. The simulation results show that as the number of basketball technical errors increases, the recognition accuracy of this method decreases slowly. When the number of basketball technical errors reaches 400, the accuracy of action recognition is still as high as 87.552%. This indicates that this method can control the error rate within a reasonable range, improve the ability to identify basketball technical errors, and provide strong support for basketball teaching. In addition, the experimental results of this method also include various other achievements in performance calculation, further verifying its superiority in identifying basketball technical errors.