Research in Acquisition of Three Dimensional Information

点击翻译按钮获取中文摘要

The main goal of this research in low- and middle-level computer vision is to develop new algorithms for the acquisition of 3-D sensory data for robotic-vision systems. The work will investigate: -- new passive vision techniques for obtaining surface orientation and curvature information about surfaces in the 3-D world; candidate techniques include: polarization-based stereo, color stereo, direct photometric curvature calculation, and shape-from-specularity -- improvements in both the accuracy and efficiency of stereo-matching and surface-reconstruction techniques from depth and orientation information. These improvements will result from a theoretically unified treatment of matching, surface reconstruction, and surface segmentation. The work will include theoretical analysis and experimental verification. It differs from previous research in that the new algorithms are to be derived from the general theories of stereo processes, which are to be finalized as part of this research. The new algorithms will either use different types of information (e.g., images under varied source polarization) or treat standard information under more-realistic assumptions (e.g., that the disparity surfaces should be smooth, but not necessarily constant or planar).