Selective reconstruction of a 3-D scene with an active stereo vision system

The ability to acquire 3D information about passable free spaces and objects around them is essential for mobile robot navigation in unknown scene. We present a new method of selective reconstruction of a 3D scene around free spaces with an active stereo vision system. Our method is based on the following observation: The larger and deeper passable free spaces may exist in the direction to which the farther 3D objects exist because the set of line of sights connecting the viewpoint with visible object surfaces form free spaces. First, we selectively reconstruct 3D objects far from the viewpoint using parallel stereo. We call them "distant objects". Then, as the direction, to which larger and deeper free spaces may exist, we select the direction of the "distant object" whose projection onto the image plane is larger. Finally, we efficiently reconstruct 3D objects in the selected direction using an active stereo vision system that can fixate any 3D point of the scene. Experimental results of the proposed method applied to the actual scene are also presented.