Research on Real-Time Tracking and Standardized Evaluation Algorithms for Swimming Posture Integrating Computer Vision

Abstract

In order to overcome the problem of high-precision, low-latency capture and quantitative assessment of human posture in swimming in dynamic underwater conditions, the paper suggests a computer vision-based framework of real-time tracking and standardized assessment of swimming postures. To focus on the identification of key points on the torso and limbs of swimmers, first, an underwater target detection network that relies on the multi-scale feature extraction and attention mechanisms will be built. Second, a spatiotemporal context fusion and 3D motion reconstruction module is developed that integrates temporal smoothing methods and camera projection geometry constraints to recreate the 3D skeleton sequence of swimming motions. Lastly, a universal scoring system founded on kinematic parameters and motion template matching is developed to determine the quality of motion quantitatively based on the joint angles, limb symmetry and stroke frequency. The experiments are confirmed using a self-constructed underwater swimming video dataset that contains the samples of four strokes of swimming in varying conditions of light and water quality. Findings indicate that the developed technique is more efficient than other comparative techniques in key point detection performance, processing speed in real-time, and consistency in evaluation. This approach is effective in providing technical assistance in training and injury prevention in swimming that is intelligent. The pose-estimation framework can be integrated with wearable motion sensors for underwater or poolside training monitoring.