Qualitative 3D Reconstruction of Football Players using Motion Graphs

This project explores the use of motion graphs as a means of qualitative 3D reconstruction of football players, based on video footage from real games. A motion graph was built from clips of recorded motion capture data, featuring a professional football player, and synthesises motion by stitching these clips together in various combinations. A scenario is assumed where the motion of the players, is sometimes complicated and requires manual reconstruction, but otherwise loosely specified and suitable for qualitative reconstruction. The latter could then be generated by the motion graph. The system was first evaluated through path synthesis which is the process of constraining the motion to a user specified path in the ground plane. The results were limited by the nature of the motion data, but allowed for the selection of a particularly useful data subset. A modest extension to the path synthesis technique, was adopted, using boundary constraints to better accommodate the reconstruction problem. Although currently computationally nonviable, the boundary constraints approach had results that hint of its usefulness as a complement to a manual reconstruction method. Further investigations should use better suited motion data, and aim at refining the boundary constraints method.