Location and pose estimation for active vision using panoramic edge histograms

Abstract

AbstractIn this paper, the authors propose a new technique for robust location and pose estimation for an active vision sensor using images obtained by a camera system (omnidirectional sensor) that can acquire panoramic information of the environment. The proposed technique first calculates a panoramic edge histogram in advance from images obtained by the omnidirectional sensor at several reference points in the environment. To estimate the sensor location and pose, dynamic programming (DP) is extended to two‐pass DP, and the current location is matched with the panoramic edge histogram at the reference points to obtain the edge shift magnitude at each direction angle. These shift magnitudes have the same periodicity as a sine curve, the sensor rotation corresponds to the offset of the sine curve, and its movement direction corresponds to the 0‐phase of the sine curve. In the current research, the sensor's movement direction and rotation angle are efficiently estimated by independently fitting the offset and the 0‐phase of its period from these edge shift magnitudes. Finally, the exact location is calculated from the movement directions for the two closest reference points. The effectiveness of the proposed technique is shown by experiments in an actual environment. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(14): 32–43, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10699