3D reconstruction by a mobile robot using multi-baseline omni-directional motion stereo based on GPS/DR compound navigation system

In this paper, a unique dense 3D shape reconstruction method featuring GPS/DR coupled omni-direction multi-baseline motion stereo system is presented. The epipolar plane equation between two omni-directional vision images is computed from the GPS/DR's position and posture information. Then, the precise epipolar lines can be obtained robustly by projecting the intersection line between the epipolar plane and the ODV's image plane. The robust matching method featuring hybrid use of the future based matching and the area based matching, as well as the bi-directional matching which chose common mutual matching points to improve robustness is also presented. Voting process using multi-baseline is used to reduce distance error of motion stereo. Hundreds of dense range images are unified based on the precise position/posture information to generate the successive dense 3D outdoor model. The range estimation accuracy within 10 [m] area is 140 [mm], which is equal to a laser radar. It can be said that the proposed omni-directional stereo vision has robustness toward environmental complication and accurate distance estimation performance for rich textured object