Unlocking the Future of Injury Prevention: AI-Powered Pose Estimation in Real-Time linked with Biomechanical Simulation

This preliminary study utilizes an AI-based approach for real-time pose estimation to analyze squats, linked with biomechanical simulation to improve personalized injury prevention. The method consists of two main steps: real-time pose estimation using dual-camera systems and biomechanical simulations to assess body loads. Pose estimation is achieved by tracking key-points (hip, knee, ankle, foot tip) through RGB images using RTMPose model. Twelve participants (6 male, 6 female) performed squats that were recorded from lateral and frontal perspective. The results demonstrate the accuracy of the pose estimation, with mean absolute errors (MAE) ranging from 3.49° to 4.15° for different views and joints. The biomechanical simulation of the knee angle shows satisfactory agreement with the ground true data. In addition, muscle force distribution analysis revealed accurate simulation results when compared to existing literature. The presented method offers real-time feedback on squat execution, enabling users to adjust their posture based on kinematic data and musculoskeletal load analysis. This approach is a step toward personalized and real-time biomechanical assessment.