Estimation of motion parameters for a roaming robot using panoramic images

Abstract

SUMMARYThis paper relates to studying a roaming robot rotating about a vertical axis while moving horizontally on a flat floor surface, and to a method for estimation of its motion parameters. For estimation of motion parameters this method uses correspondences of five or more direction angles in panoramic images captured from three viewpoints by a roaming robot. In this study, first, motion parameters of a roaming robot were limited to horizontal components of translational and rotational parameters, and then the authors concentrated on direction angles as horizontal components of features. Thus, the difference between the measured direction angles and the directional angles obtained by observation was formulated as the error evaluation function, and parameters were estimated by nonlinear minimization. In addition, numerical and computer graphics-based simulation techniques were used for evaluating accuracy, and the effectiveness of this method was confirmed by tests using real images. This method enables capturing panoramic images by a roaming robot in multiple viewpoints with compensation of the capturing location errors.

Location and pose estimation for active vision using panoramic edge histograms

“…These techniques are divided into view-based matching techniques [10][11][12][13][14][15][16] and matching techniques based on geometric features (or landmarks) [3,5,[17][18][19]. A view-based matching technique stores features extracted in advance from omnidirectional images at various locations along a learning route, and when estimating the sensor location, compares the features obtained at the current location with the learning image features that were stored in advance to obtain the closest location on the learning route.…”

“…Li and Tsuji [19] used stereo and color information from panoramic views to extract landmarks and used landmarks to perform robot navigation. In addition, techniques for estimating sensor motion parameters (relative location or pose) from corner tracking [13] or vertical edge matching [3,18] or associations of feature points obtained by using floor edge tracking [5] between frames of a sequence of omnidirectional images have been proposed. Although these techniques can estimate the sensor location by using few feature points or landmarks, the feature points or landmarks must be stably extracted and matched.…”