Semi-automated 3D Reconstruction of Volleyball Players for Physical Load Analysis

Abstract

Despite volleyball not being a contact sport, the sport's high-intensity movements - such as jumps, dives, and sprints - pose injury risks to players. In recent years, the surge in computer vision applications within sports has offered robust analysis tools for both coaches and players. However, the existing applications in volleyball still lack comprehensive 3D-based metrics in a non-laboratory setting. This work addresses this gap by developing a semi-automated algorithm specifically designed to reconstruct volleyball players in 3D, using a calibrated multi-camera set-up able to capture synchronized videos. The primary goal of this research is to equip coaches with a valuable tool for assessing individual players' physical loads enabling them to make more informed decisions or adjust their training program to prevent injuries. Detailed analyses of player jumps, dives, and trajectories within rallies provide insights into the algorithm's performance and practical application in volleyball settings. Our findings reveal promising advancements in jump and dive detection, as well as jump height estimation, as our estimates are closer to the ground truth than those of the widely employed Vert sensors. Moreover, our findings highlight the significance of a 3D player reconstruction algorithm, concluding that the implications go far beyond the explored analyses in this work as there are numerous other potentially intriguing avenues left unexplored.