AART: Athlete Action Recognition and Tracking

Abstract

medical imaging systems. However, conventional motion capture approaches are often expensive, computationally intensive, and unsuitable for real-time applications in dynamic and noisy environments. This paper proposes an Adaptive Algebraic Reconstruction Technique (AART)-based athlete action recognition and tracking system implemented as a lightweight desktop application. The framework integrates state-of-the-art deep learning and computer vision models, including YOLOv3 for real-time object detection, OpenPose for skeletal keypoint extraction, and Long Short-Term Memory (LSTM) networks for temporal motion sequence modeling. The AART mechanism employs adaptive step-size control to enhance reconstruction accuracy, convergence rate, and robustness against challenges such as occlusion, motion blur, and low-resolution inputs. The system supports both offline video processing and real-time webcam-based tracking, providing visualization of skeletal motion through an interactive graphical user interface. Experimental evaluation demonstrates improved accuracy, reduced tracking error, and lower computational overhead compared to traditional methods. The proposed approach offers a scalable and efficient solution for applications in athlete performance analysis, injury prevention, coaching assistance, and intelligent sports analytics.