Co-host: The Institute of Image Electronics Engineers of Japan, The Institute of Image Information and Television Engineers
In this paper, we propose a new approach for 3-D omnidirectional reconstruction using a rotational camera whose field of view is separated by two planar mirrors. We use two mirrors for obtaining a larger baseline stereo than that in the case of a previous method that uses a single rotational camera without separated field of view. For reducing matching errors caused by image noises and repeated patterns, we extract trajectories of 3-D points from spatio-temporal image using SSSD (Sum of SSD). Moreover, it is possible to reduce influence of occlusion by evaluating SSSD values about the trajectories in spatio-temporal volumes of the left and right mirrors separately. The validity of the proposal method is shown by the experiment results using synthetic and real images.