Introduction Augmented reality has already been used in sport broadcasting. It consists in overlaying graphics elements - speed data, swing plane - to the original video in order to offer richer and improved images to the TV audience, speakers in sport or sportive experts. We developed a new 3D tracking technique from image analysis using several synchronized video cameras and located directly on a golf field, in order to improve the content of golf tournament on TV. For some sports such as golf, motion capture in studio is not appropriate since the athlete adapts his play to the environment. This is the reason why developing a new motion capture system for outdoor purposes and without markers might be really useful. The GOLF-STREAM project provides data to better understand golf actions by presenting enhanced images, to better analyse and develop golf techniques, to coach and improve champion performances and to extract and visualise some pedagogical rules. The algorithm is a 3D articulated model-based tracker using a simplified anthropometric virtual 3D human. A generic model needs, at first, to be adjusted to the morphology to the filmed champion. An interactive positioning of the 3D model on the first frame of a video sequence is performed, then, the tracking algorithm automatically computes the evolution of the values of the set of degrees of freedom of the human body all along the swing motion. State of the art The human motion capture technique developed in the project GOLFSTREAM doesn't need any markers but uses a 3D human body model. In our early work, this algorithm was developed to track simple 3D rigid bodies. Only six degree of freedom (3 rotations angles and 3 translations) were computed for each video frame comparing the appearance of a parametric 3D model with images of a video sequence [2]. Gavrila [1] et al. obtained good results by tracking dancers. Like in traditional optical tracking systems, we position a set of video cameras all around the zone where we capture champions motion. Design of a synthetic puppet A synthetic puppet is used to mimic the motion of professional golfers. The quality of its design is a trade-off between the number of degrees of freedom to compute and the puppet ability to mimic accurately the swing motions. The DOF values are limited to certain intervals in order to force the puppet to only take realistic and stable postures. The synthetic bones influence the 3D skin envelop surrounding them according to some weight fixed at the design time. The 3D envelope moves and is deformed according to the bones positions when the puppet posture is modified. Morphological adjustment and pre-positioning of a synthetic puppet The generic puppet is then adjusted to the filmed champions morphology. The adjustment of the skeleton and the skin envelop is realised semi-automatically using the video images. Then, the puppet is overlaid on the first frame of the video sequence and its position is retrieved interactively. Tracking algorithm After the initialisation, texture of the synthetic puppet is extracted from the first frame of the different cameras. Then, a simulated annealing optimisation algorithm retrieve, for each video frame, the values of the degrees of freedom of the puppet skeleton in order to mimic the swing made by the golfer. The tracking process minimises a matching error between the appearance of the synthetic puppet and the video frame, following the skeleton hierarchy: the pelvis of the puppet is first matched to the image, then comes the position retrieval of the spine, clavicles, neck, arms and feet. Each trial requires a rendering computation and a pixel-to-pixel matching between the generated synthetic image and the real image. This tracking process keeps on going from frame to frame up to the end of the video sequence and allows the measurement of a golfer motion by mimicking him (see results on the image showing 4 views of the golfer) Conclusions : The type of tracking proposed is non-intrusive and can take place during a championship, and is able to capture the " real " sportive motions. The joint study of a player morphology and of his motion allows one to study differences between types of swings for a given morphology. All the environment parameters and psychological conditions are implicitly recorded at the time of the motion capture. Our system of gesture analysis will also be useful for several other athletic sports.
© Copyright 2002 Expertise in Elite sport. 2nd International Days of Sport Sciences, 12.-15. November 2002, INSEP, Paris (France). Veröffentlicht von INSEP. Alle Rechte vorbehalten.