Volume 26 Issue 6 - Publication Date: 6 June 2007
3D Reconstruction Using Multibaseline Omnidirectional Motion Stereo Based on GPS/Dead-reckoning Compound Navigation System
J. Meguro Waseda University, 17 Kikuicho, Shinjyuku-ku, Tokyo, Japan, J. Takiguchi Mitsubishi Electric Corp, 325 Kamimachiya, Kamakura, Kanagawa,
247–850 Japan
, Y. Amano, and T. Hashizume Waseda University, 17 Kikuicho, Shinjyuku-ku, Tokyo, Japan
This paper presents a motion-stereo device using the GPS/DR (dead reckoning) combination along with omnidirectional cameras to reconstruct 3-D environments by means of sensors small enough to be installed on a mobile robot. The proposed technique is based on the fact that it is possible to determine epipolar lines by using the precise positions and heading angles measured by a combination of GPS and various sensors; high-precision stereo matching may then be performed by means of geometrical restrictions. Furthermore, in comparison to the other stereo techniques, it is also possible to establish longer baselines for peripheral objects and simultaneously provide a higher precision in calculating the distances to far objects by applying voting processing to a series of distant motion-stereo images, which can be attained by using more than one baseline length, in 3-D spaces based on the precise positions and heading angles. In short, the proposed technique provides a substantial increase in measurement precision as compared to that with the well-known SFM (structure from motion) technique for reconstructing 3-D environments with monocular cameras; this is because the proposed technique utilizes a greater number of mobile parameters, which are determined using precise cameras. In this experiment, the measurement precision of the proposed technique has been evaluated in reconstructing the shapes of vehicles parked alongside a road; the measurements were found to have a standard deviation of 140 mm within a range of 10 m. It can be stated that the proposed omnidirectional motion-stereo technique is robust to environmental perturbations and can accurately estimate distances in the case of highly textured objects.
Return to Contents