Automated Evaluation System for Pull-Up Exercises in Physical Fitness Testing Scenarios

With the cross-fusion development of artificial intelligence and computer vision in various fields, different industries are incorporating computer vision technology to achieve intelligence and automation. Especially in large-scale scenario tasks, unsupervised approaches can reduce labor and time costs. Physical fitness testing, as a necessary assessment subject for physical fitness, involves a large number of personnel and a relatively scarce number of professional instructors. This not only results in high manpower and time costs but also renders the assessment task meaningless without professional evaluation. In this context, this paper presents an intelligent analysis system based on computer vision that uses the pull-up exercise in physical fitness activities as an example. By utilizing the body pose coordinates feedback from pose estimation, combined with the proposed graph convolutional action evaluation network, the system can achieve segmentation of continuous actions and quality evaluation of segmented sub-actions. The test results demonstrate a high correlation between the proposed method’s evaluation and the results given by professionals. This system can also be applied to intelligent analysis tasks in other physical fitness exercises.