Fusion of 3D Laser Scans and Stereo Images for Disparity Maps of Natural Scenes
Johannes M. Steger · 2010
Extracting three-dimensional structure from multiple views of a scene is a classical problem of machine vision. This task is especially demanding for cluttered natural scenes. We directly measured the 3D structure of natural scenes using a laser range scanner. We developed a technique to use these scans to obtain accurate disparity maps for images taken by a stereoscopic camera rig observing the same scene. The disparity maps thus acquired provide the basis for an algorithmically unbiased ground truth for the evaluation of stereo matching algorithms on natural images. Additionally, they are used to synthesize stimuli for psychophysical and electrophysiological experiments on three-dimensional viewing in humans.