Research on sit-up counting method and system based on human skeleton key point detection

This paper proposes a sit-up counting system based on bone key point detection, aiming to solve the subjectivity and inefficiency problems of traditional manual counting. The system uses deep learning algorithms to detect and identify key points of human bones, and by improving the network structure and introducing jump connections, it significantly improves the accuracy and efficiency of positioning key points of bones. The number of improved network parameters was reduced from 125.6M to 38.3M, significantly reducing model complexity; at the same time, the processing speed was increased from 36.5 FPS to 63.6 FPS, showing higher processing efficiency. Combining skeletal key point tracking and illegal motion detection, real-time and accurate counting is achieved, and the accuracy of sit-up detection reaches 98.57%. The system integrates real-time detection, data collection, display, storage and query functions, providing an efficient and objective counting solution for sports and testing.