TR-M-0025 :1997.8.22

スダンシュ クマール セムワル, 大谷淳

Geometric-Imprints: A Significant Points Extraction Method for the Scan & Track Virtual Environment

Abstract:We have implemented a new algorithm, called the geometric-imprints method for extracting key-features of a given contour. In our implementation, the user can draw an arbitrary 2D-contour, and specify a set of contour-points in a clockwise order. This series of points on the curve are used to find a geometric-imprint set of the contour. The algorithm is based upon the assumption that an articulated human and other figures have mostly cylindrical body-parts. The algorithm recursively partitions the contour based upon the cylindrical nature of human-silhouettes, and a starting point. The geometric-imprint set consists of a description of cylindrical portions of the contour. This property is important for VEs where multiple cameras are used to estimate the motion of participants as similar topological contours are expected to produce similar geometric-imprints, thus the correspondence problem is somewhat simplified. Results of our implementation are presented.

Key words: 2D contours, Geometric Features, Unencumbering, Virtual Environments.