Application of Computer Vision and Deep Learning in Swimming Action Recognition and Evaluation

China’s position in the world swimming industry is becoming increasingly high, which is also an important achievement in the swimming industry. Improving the strength of athletes is urgent. This article used a combination of sliding window technology and mixed Gaussian model to model the underwater video background of a swimming pool. A video monitoring model based on computer vision and deep learning was used to recognize the actions of swimmers. This article extracted breaststroke motion data from various stages, as well as motion data from freestyle, butterfly stroke, and different stages of breaststroke, and conducted classification and recognition work. This article evaluated the recognition accuracy of different stages of breaststroke and the specific stage actions that were easily misclassified as freestyle and butterfly stroke, in order to achieve overall recognition and stage recognition of swimming posture and expand the application range of recognition models in swimming motion recognition and evaluation. Swimming motion recognition found that in butterfly stroke, the maximum pitch angle between the head and hips was 110°, and the maximum pitch angle between the head and hips was 42°. This article provided a sports guidance system suitable for swimming, which helped empower skills for training and enables most athletes to grow rapidly.