Integration of Multiple Cues for Robust 3D Object Description: A computational and Psychophysical Study with Applications

摘要

The project involves a comprehensive study of object description using multi-sensors. The study examines two basic scenarios for surface reconstruction. The first scenario provides a 2D - to - 3D mapping from images to surfaces and will indude stereo, focus, zoom, vergence, shape from shading, and shape from texture. The second scenario uses active range finders to provide direct depth information about the object, i.e., provides a 3D - to - 3D mapping. The research focuses on the representation and fusion of information form differing image sources and the use of machine learning techniques to perform the fusion. Psychophysical studies conducted include investigating the applicability of the recently introduced 'quasi 2D coding hypothesis for 3D surface representation' in machine vision; and the behavioral evaluation of human performance with 3D fused imagery. The investigations of 3D surface reconstruction in human, computer and robot vision have an important applications in military, manufacturing and medical areas. As described in technical report this research has been conducted in the Computer Vision and Image Processing Laboratory (CVIP Lab) at the University of Louisville. The 'vision environment' created in the CVIP Lab enabled integration of multiple cues to sense, explore and reconstruct the environment layout. As a result, an active vision system called the CardEye was created.