Recovering Human Actions Using Eigenspace

This paper presents a new technique for recovering a 3D human action from its video images by a computer vision system. For the recovery purpose, there are already some motion capture systems commercially available. They are, however, still not very convenient for common use because of camera calibration and the use of artificial markers attached on a human whose 3D positions are tracked. Unlike these existent systems, the proposed technique employs neither camera calibration nor markers. Instead it employs a database that describes human actions and, once an unknown action is captured, the technique performs 3D recovery by retrieving a similar action in the database. The database is created employing an eigenspace method, since the method reduces image information effectively as well as it assumes less amount of image preprocessing. The proposed technique was examined its performance by the experiment for recovering some basic human actions and the result was satisfactory