| 202A1 |
|---|
| On Global Eigenspace for Human Motion Recognition |
| AUTHORS |
| Md. Masudur Rahman, Seiji Ishikawa (Kyushu Institute of Technology) |
| ABSTRACT |
| The paper describes a technique for representing and recognizing human motions employing an eigen-space (ES) method. Automatic recognition of human motions or activities has drawn much attention by many researchers because of its importance in various application fields. Unlike 3-D techniques for human motion recognition proposed to date, this paper presents a 2-D technique based on sampled images. Human motion is captured a video image by a camera from his front position. The image stream is sampled to produce a set of images representing the motion and the image set is employed for making an ES after having been transformed into blurred edge images. The ES can contain as much motions as possible and is called a global ES (GES). Employing the GES, recognition of an unknown motion is performed. Six motions of a cricket umpire taken from a subject are employed for creating a GES in a performed experiment. Four subjects take the same motions and their images are processed for the recognition. The motions were recognized 100% exactly, whereas the sampled postures in each motion were recognized 92% correctly. |
Back to Program
Time Table
Top of This Site
Annual Conference 2002