A Pull-up Counting Algorithm Based on Human Key Point Detection and Image Segmentation

Pull-ups are an efficient exercise aimed at strengthening the muscles of the upper limbs, chest, and back. Through the synergistic action of these muscle groups, pull-ups achieve vertical displacement of the body on a fixed horizontal bar. As an important component of physical fitness assessment, pull-ups are not only crucial for enhancing personal constitution and overall physical fitness levels but also serve as a standard for measuring individual muscular endurance and strength qualities. To promote the intelligent transformation of physical education and testing processes, this paper proposes an innovative pull-up counting algorithm. The algorithm integrates key point detection and image segmentation techniques, utilizing deep learning models to achieve precise capture of human posture in sports videos. By effectively segmenting the pull-up bar area and performing line detection, combined with position change analysis, it realizes accurate counting of pull-up motions. Experimental results show that this algorithm possesses excellent environmental adaptability and robustness, maintaining high counting accuracy under varying lighting conditions and other interfering factors. On the 400ppd dataset, the average accuracy rate reached 96.5%. The experimental results demonstrate that the algorithm has outstanding environmental adaptability and robustness, ensuring high counting accuracy even under changing lighting conditions and other disturbances.