Visual computation of egomotion using an image interpolation technique

Abstract

A novel technique is presented for the computation of the parameters of egomotion of a mobile device, such as a robot or a mechanical arm, equipped with two visual sensors. Each sensor captures a panoramic view of the environment. We show the parameters of ego-motion can be computed by interpolating the position of the image captured by one of the sensors at the robot's present location, with respect to the images captured by the two sensors at the robot's previous location. The algorithm delivers the distance travelled and angle rotated, without the explicit measurement or integration of velocity fields. The result is obtained in a single step, without any iteration or successive approximation. Tests of the algorithm on real and synthetic images reveal an accuracy to within 5% of the actual motion. Implementation of the algorithm on a mobile robot reveals that stepwise rotation and translation can be measured to within 10% accuracy in a three-dimensional world of unknown structure. The position and orientation of the robot at the end of a 30-step trajectory can be estimated with accuracies of 5% and 5 degrees, respectively.

Catchment areas of panoramic snapshots in outdoor scenes

“…Our comparison between difference functions in scenes with different depth structure suggests that the shape and depth of these functions depend on the spatial frequency content of images, on the degree of occlusion and on the depth structure of the scenes they are recorded in (see also Refs. 7,8,[16][17][18][19]. Our evidence for this conjecture is at this stage still qualitative, but the following considerations do give it some weight.…”

“…The dependence of image differences on the depth structure of a scene, allows us to predict that the pixel differences contributing to a given difference function are not distributed equally across the 'visual field'. Much like the image velocity vectors in the optic flow field experienced by a moving optical system, the image differences generated by a displacement, depend on the direction of translation, with largest differences occurring in directions of view perpendicular to the heading direction 8,9 . We would also predict that, if the depth, shape and smoothness of difference functions depend on the spatial frequency content of a scene, low-pass filtering the images should make the functions shallower and smoother.…”

“…In most cases in which panoramic images have been used for view-based robot navigation [6][7][8][9][10][11] , image information has been greatly reduced, mainly because this generates a noise-free 4 pattern of signals in the indoor worlds of experimental robots where man-made structures and artificial lighting produce clearly identifiable, and mostly vertical edges.…”

“…Or to state the question more accurately, how different is the visual world when viewed from neighbouring vantage points, is this difference correlated with, and does it vary smoothly with physical distance? Our approach is inspired by the image interpolation technique developed by Srinivasan 7 and Chahl and Srinivasan 8 , who have shown that over small distances and for small angles of rotation, the position and orientation of an image can be determined by interpolation from reference images taken at two different locations. The technique has been successfully used to measure egomotion, range and surface orientation 7,8,[16][17][18][19] in simulated and in indoor environments.…”

“…Our approach is inspired by the image interpolation technique developed by Srinivasan 7 and Chahl and Srinivasan 8 , who have shown that over small distances and for small angles of rotation, the position and orientation of an image can be determined by interpolation from reference images taken at two different locations. The technique has been successfully used to measure egomotion, range and surface orientation 7,8,[16][17][18][19] in simulated and in indoor environments. We apply a simple variant of the technique here to analyse the principal constraints of view-based homing in outdoor scenes by studying how global image differences develop with distance from a reference position.…”