Epipole-based visual servoing for mobile robots

This paper proposes a visual servoing algorithm for mobile robot navigation based on the epipolar geometry retrieved by object profiles. The main motivation for this approach is that for unstructured scenes the task of solving correspondences is definitely a harder task than contour detection. An unstructured three-dimensional (3-D) scene consists mainly of objects whose most important 2-D features are their apparent contours in the image plane. Apparent contours are used to estimate the positions of the epipoles and some special symmetry conditions whereby the visual servoing is able to steer the mobile robot to the desired position.