Accession Number : ADA193090

Title :   3-D Road Structure from Motion Stereo.

Descriptive Note : Technical rept.,

Corporate Author : MARYLAND UNIV COLLEGE PARK CENTER FOR AUTOMATION RESEARCH

Personal Author(s) : Asada, Minoru

PDF Url : ADA193090

Report Date : Apr 1987

Pagination or Media Count : 30

Abstract : This paper presents a new method for reconstructing the 3-D structure of road boundaries from consecutive images. First, we present a method for estimating depth information by applying a motion stereo method to consecutive images, given an estimate of the interframe motion. The relation between depth, motion and disparity is investigated, since the accuracy of the depth depends on the disparity range. Next, the error of the estimated road structure due to quantization errors and motion estimation errors is examined. Finally, a representation for road boundaries is proposed that makes explicit the error of the road edge location in 3-D space. Experimental results are shown for an input image sequence taken by the ALV(Autonomous Land Vehicle) simulator robot in the Center for Automation Research at the Univ. of Maryland. Keywords: Three-dimensional shape recovery; Computer vision.

Descriptors :   *ROADS, *OPTICAL DETECTION, ACCURACY, AUTOMATION, BOUNDARIES, COMPUTERS, DEPTH, EDGES, ERRORS, ESTIMATES, IMAGES, INPUT, MOTION, QUANTIZATION, RECOVERY, ROBOTS, SEQUENCES, SHAPE, SIMULATORS, THREE DIMENSIONAL, GROUND VEHICLES, CLOSED CIRCUIT TELEVISION, COMPUTER APPLICATIONS

Subject Categories : Optical Detection and Detectors

Distribution Statement : APPROVED FOR PUBLIC RELEASE