Terrain Scene Analysis and Obstacle Reconstruction for Navigation of Mobile Robots
Final technical rept.
ARMY ARMAMENT RESEARCH AND DEVELOPMENT CENTER WATERVLIET NY LARGE CALIBER WEAPON SYSTEMS LAB
Pagination or Media Count:
For a robot to be able to identify the objects in the scene, a form of sight must be provided. The first stage in achieving sight is the collection of visual input data in the form of depth information, such as that received from a laser range finder. The laser range finder is mounted on a mast from which a noisy measurement matrix is generated. With these noisy measurements, a rapid estimation scheme is used to detect the presence of horizontal and vertical edges on a terrain by processing the range data successively along each column and row of the range matrix. The result of the estimation is a collection of data points to form a curve in space that belongs to some edge of an obstacle from the vantage point where the laser range finder is located. The orthogonal surface slopes of a obstacle can be determined from the range slopes which are estimated from the range matrix. The segmentation of range data on the basis of surface slopes provides groups of connected data points that belong to one particular face of some observed obstacles. The problem of grouping range data points of different planar surfaces on the basic of their surfaces slopes becomes an application of clustering analysis. The first approach of a heuristic scheme for object reconstruction and formation is presented here based on input data containing depth information. The set of heuristic rules is based on geometric considerations and insight into the characteristics of objects, which are found to be the convexity and colinearity of edges. Methods for determining if an edge is convex or concave and if two edges are colinear are developed.
- Human Factors Engineering and Man Machine Systems