Development of a motion capture system based on the retrieval in a human motion database on an eigenspace

运动捕捉是一种对人体运动进行三维建模的技术，该系统被用于制作数字电影、数字视频游戏等中的角色。但现有的运动捕捉系统需要摄像机标定才能恢复运动人体上的标志物，这阻碍了运动捕捉系统的广泛应用，本研究项目旨在开发一种既不使用摄像机标定又不使用标志物的新型运动捕捉技术。它是通过基于特征空间技术对各种人体运动视频图像建立运动数据库，并在数据库中给出树形结构来加快检索速度来实现的。一旦摄像机拍摄到未知的人体运动的图像，就在运动数据库中高速搜索与所捕捉的运动最相似的运动。发现的运动给出了未知人体运动的3-D模型。实验结果表明，该系统的性能令人满意。该系统只需拍摄运动中的人在现场的视频图像。然后，仅通过视频流以3-D方式恢复运动。这意味着，与现有的运动捕捉系统相比，所提出的运动捕捉系统可以应用于各种人体运动。然后，人们可以预期运动捕获系统在人类运动恢复/分析是主要关注的各种社区中得到更广泛的使用。

Motion capture is a technique for 3-D modeling of human motions and the system is employed for making the characters in digital films, digital video games, and so on. But the existent motion capture systems need camera calibration for 3-D position recovery of the markers attached on the body of a person in motion, which prevents the motion capture systems from their wider use.This research project is to develop a novel motion capture technique which employs neither camera calibration nor markers. It is realized by making a motion database from various human motion video images based on an eigenspace technique and by giving a tree structure in the database in order to speed up its retrieval. Once an unknown human motion is taken the images by a video camera, the motion which is the most similar to the captured motion is searched in a high speed in the motion database. The found motion gives the 3-D model of the unknown human motion. Experimental results show satisfactory performance of the proposed system.By the proposed motion capture system, one only has to take video images of a person in motion at a site. Then the motion recovers in a 3-D way only by the video stream. This implies that the proposed motion capture system can be applied to a variety of human motions compared to the existent systems. One can then expect wider use of the motion capture system in various communities where human motion recovery/analysis is the primary concern.