Human Motion Representation Using Eigenspaces

Recognition of human motions using their 2D appearance images has various applications. An eigenspace method is employed in this paper for representing and recognizing human activities/motions. An eigenspace is created from video images taken by multiple cameras that surround a human in motion. The image streams of the motion compose several curved lines in the eigenspace that are closely situated with each other. The motion is described by this set of lines, which ultimately compose a curved surface. It is used in this study for recognizing a human motion observed from an arbitrary orientation. Experimental results of applying our approach on multiple cameras are given to demonstrate the effectiveness of the proposed technique.