Forming a three dimensional environment model using multiple observations

An autonomous navigating agent must form a three-dimensional model of its environment using passive sensors. Typical stereo algorithms produce sparse depth maps and cannot be used to distinguish between holes and solid objects in the environment. The authors present a novel methodology for creating a three-dimensional model of the environment. They divide the environment into a set of disjoint cells. Using multiple images obtained from different view points, they estimate the mean and variance of intensity observed for each cell. The computed variance can be used to distinguish between empty and full cells in the environment. The technique, unlike the typical stereo methodology, does not rely on solving the correspondence problem. The resulting model of the environment is dense and can be used directly for navigation. Experimental results are presented