3-D Scene Data Recovery using Omnidirectional Multibaseline Stereo

A traditional approach to extracting geometric information from a large scene is to compute multiple 3-D depth maps from stereo pairs or direct range finders, and then to merge the 3-D data. However, the resulting merged depth maps may be subject to merging errors if the relative poses be-tween depth maps are not known exactly. In addition, the 3-D data may also have to be resampled before merging, which adds additional complexity and potential sources of errors. This paper provides a means of directly extracting 3-D data covering a very widefield ofview, thus by-passing the needfor numerous depth map merging. In our work, cylin-drical images are first composited from sequences of im-ages taken while the camera is rotated360 0 about a vertical axis. By taking such image panoramas at different camera locations, we can recover 3-D data ofthe scene using a set ofsimple techniques: feature tracking, an 8-point structure from motion algorithm, and multibaseline stereo. We also investigate the effect ofmedianfiltering on the recovered 3-D point distributions, and show the results ofour approach applied to both synthetic and real scenes.