StereoDROID

Stereo and structure-from-motion processing are combined in an attempt to overcome the disadvantages of each. The absolute base-line provided by the separation of the stereo cameras enables the speed-scale ambiguity to be resolved, and the long base-line resulting from processing a long image sequence enables good long-range accuracy to be obtained. StereoDROID is an approach to stereo structure-from-motion. It was developed from M.J. Stephens DROID vision system (1989) which uses the monocular structure-from-motion approach. The techniques employed in matching, locating the 3D features positions and calculating the ego-motion are first presented for the monocular case and then it is shown how stereo data may be coherently integrated into the solution. Although the results are presented for a stereo system any number of cameras may be used. An analysis is developed for a moving stereo camera head in an otherwise static world. Corner-point data extracted from the images are used to form a scene representation consisting of 3D feature-points. These are specified by their positions and positional uncertainties expressed as normal probability distribution functions.<
>