Panoramic stereo vision sensor

Classical computer vision, like human vision, only allows one to consider a narrow angle of view of an observed scene. Panoramic vision gives a solution to this problem by permitting a full 360° vision. This paper presents a general overview of all the panoramic sensors used up to now. We also introduce our new panoramic sensor based on the use of two linear CCD arrays of resolution 1024 pixels. The aim is to determine the geometric shape of indoor scenes using matching algorithms simplified by the architecture of our system allowing a real time constraint. An interpolation algorithm of 3D data based on free form deformation is presented. It allows one to find the position of walls in the 3D data even in a case of poor reconstructions