Fitness Training Assistance System Based on Mediapipe Model and KNN Algorithm

Fitness exercise is essential to improve the physical literacy of all people, however, the cost required for fitness is expensive and the lack of professional guidance for individuals may lead to the risk of injury. To address this issue, this paper proposes a fitness training assistance system based on machine learning algorithms. The system is capable of classifying and counting multiple training movements. First, the MediaPipe model is used to realize the accurate recognition of human posture and extract key point features. Then, by analyzing the angles formed by the coordinates of these key points, the counting function of the movements was realized. Next, the KNN algorithm is combined to classify and recognize the training actions. Finally, the PCA dimensionality reduction method and cross-validation technique were used to post-process the data, resulting in an overall model accuracy of 99%.