Presents a new method based on 3D facial section analysis for human face identification. The range data on different curves of intersection, such as, vertical, horizontal, and circular, with the 3D face are used as the distinguishing surface features. These are robust against changes in makeup and lighting as compared to the 2D approach, and more efficient for computation and data storage than other 3D approaches. The performance of this method for human face identification is evaluated through several matching experiments. It has been observed that the curves of intersection crossing the facial central area vertically and contain the features, like, nose and mouth, have the major distinctiveness. Those crossing near the inner corners of the eyes and a part of the nose are also effective for human face identification. But it seems that such an effectiveness requires the accurate extraction of the sections against their locational variations.>

A non-metric path planning algorithm, Curv1, is developed for moving a point automaton between two given points along a guide-track, amidst unknown obstacles. No physical mark is made on the guide-track and no position on distance information is used. The nonheuristic algorithm is shown to converge and an upper bound on the path length is derived. This nonmetric formulation is shown to be related to, and in some sense, a generalization of the maze-searching problem. For the curve-following task, the Curv1 algorithm is shown to be a generalization of the maze-searching Pledge algorithm (H. Abelson and E. DiSessa, 1980). The robustness of the algorithm makes it suitable for an industrial application of autonomous robot guide-track following.>

We present a new guidance system using optical reflectors as landmarks. The advantages of using the reflectors are; (1) easy to identify, (2) inexpensive, (3) easy to set up, and (4) inconspicuous to human. We also present a method to generate and follow smooth paths between the discrete landmarks, and to control a mobile robot in real-time for a smooth and non-stop motion.

The problem of making a map of an unknown scene containing objects of arbitrary shapes is considered. A specific formulation of the terrain acquisition problem due to Lumelsky et al. is investigated. The motivation is to develop efflcient new algorithms and understand the basics of the problem. A new, generalized algorithm GenTer is developed. Varying a parameter (Y in GenTer produces a family of algorithms. A particular version of the generalized algorithm, called Terl, performs better on the average than the existing algorithm, the Sightseer Strategy. Terl offers a new feature called the Hierarchical Map Making, through which a good approximate map can be efflciently created.

A region-based reconstruction technique is proposed which simplifies the process of scene reconstruction and increases reliability. Integrating grid coding and intensity image analysis methods, stable regions which make up the skeleton structure of a room (i.e. wall, ceiling, floor, etc.) are extracted. The indoor scene is then reconstructed based on 2-D and 3-D topological relationships. The experimental results show that local errors do not radically alter the topological relationships of detected regions, so robust reconstruction can be achieved.>