Predictive Modelling of Ball Motion Trajectories Through Computer Vision Analysis

This paper presents a computer vision system for real-time object detection and trajectory prediction with significant applications in data analytics for sports and performance analysis. Modern performance evaluation increasingly relies on quantitative motion data to extract meaningful insights about movement patterns, biomechanical efficiency, and tactical optimization. However, existing methods typically require complex hardware or advanced models, limiting accessibility for widespread analytical applications. To address this gap, a lightweight system was developed by integrating deep learning object detectors with Kalman filter-based motion prediction algorithms. The object detection model was trained on 889 labeled images to detect moving objects with precision, while the Kalman filter estimated velocity and predicted future trajectories using a constant acceleration model, generating rich temporal datasets suitable for comprehensive motion analysis. Testing on both simulation and live video streams demonstrated the system’s capability to produce analytical-grade data with a Mean Absolute Error (MAE) of 9.25 pixels and Root Mean Squared Error (RMSE) of 55.30 pixels. Despite initial cold-start errors, the system consistently adapted after several frames, maintaining data quality standards required for performance analytics. The results indicate that accurate motion trajectory prediction and comprehensive data generation can be achieved using cost-effective single-camera setups, democratizing access to sophisticated motion analysis tools previously available only in high-end facilities and enabling data-driven performance optimization across sports training, biomechanical research, and educational applications.