Location and pose estimation for active vision using edge histograms of omni-directional images

This paper proposes a new method for location and pose estimation of active vision using the Stereo Omni-directional System (SOS), which can acquire both color and real time stereo images in all directions of a scene. With SOS, we propose a robust method to estimate the location and pose of the sensor by matching the edge histograms of the omnidirectional images obtained at the present location and that obtained beforehand at reference points. The proposed method first estimates the location and relative horizontal rotation angle by matching the vertical edge histograms of the omni-directional images. Next, for each horizontal direction, we generate horizontal edge histograms of the omni-directional images obtained by the SOS, using the perspective images with a FOV of 90° centered at the concerned direction. We then match the horizontal edge histograms in each direction and use the matching results to estimate the pose of the sensor. Experimental results show the effectiveness and robustness of the proposed method.