Real-Time Human Motion Detection and Analysis Using Lightweight YOLOV8N

This paper investigates real-time human motion detection leveraging computer vision and deep learning technologies limitations of conventional sports training methodologies. A dataset for sports motion keypoint detection is meticulously constructed by collecting and manually annotating images of exercises such as pull-ups, sit-ups, and bar dips. To facilitate flexible implementation in environments with limited resources, the YOLOv8n-pose keypoint detection model is optimized for lightweight deployment, achieving a substantial reduction in parameters and computational demands while maintaining comparable performance to the baseline model. Furthermore, a sub- movement-based motion standardization analysis algorithm is employed to evaluate movement standardization and automatically count compliant actions in real-time. The algorithm designs evaluation parameters based on standard movement requirements and decomposes a complete action into multiple sub- movements for precise analysis. Experimental results demonstrate that the lightweight keypoint detection model ensures efficient deployment in diverse environments, and the accuracy of the motion standardization analysis algorithm reached 98.1 %, meeting the requirements of real-time sports motion detection.