CSE 190a Project Proposal: Converting 2D golf swing sequences into 3D models

Innovations in computer graphics and computer vision have been applied in various golf-related applications – motion capture systems to provide detailed swing analysis, video systems to supplement golf instruction, highprecision launch monitors to track club and ball physics, and state-of-the-art graphics techniques to develop golf video games that continue to look more and more realistic. Recently, EA Sports has added a feature to their golf video game to make the user experience especially unique: the ability to upload images of one’s own face to map onto 3D characters in the game. We would like to extend this idea by developing a system that takes a 2D video sequence of a golf swing and translates it onto a 3D character model, which, if successful, could reasonably be incorporated into future versions of the game. Motion capture can be used to create very precise 3D models of a golfer’s swing, but it is of course only available to people with access to such studios. It is very accessible and common, however, for golfers to record video of their swings using traditional video cameras or digital cameras. Converting the latter format – a 2D sequence of images – into the former – a 3D model – would approximate the precision of motion capture without the barrier to entry. We will build a software application that takes as input video of a golf swing from several vantage points and produces a 3D model that approximates the positions of the golfer in the input video. To assist in the conversion process and to achieve more precise results, the user will be asked to calibrate the input swing sequence by identifying key positions in the swing (e.g. the locations of the golfer’s shoulders, elbows, and hands). These clues will then be used to map the golfer’s positions onto a 3D model.