Autonomous Robot Navigation-A study using Optical Flow and log-polar image representation

This paper describes a methodology for autonomous robot navigation, based on log-polar transform of images and optical flow. The navigation task for robots involves the detection of obstacles in the traversable path. This is considered as basic capability for mobility that includes the measurement of the height of objects to classify them as to be avoided or to be ignored. The vanishing point in the image corresponds to the Focus of Expansion (FOE) since we are assuming that the mobile robot moves with a translational velocity parallel to the ground plane. The FOE is determined from the optical flow field using a phase-based approach. From the FOE in the images and assuming the robot moves in the levelled ground, the planar homography H, is recovered and any object on the floor can be detected. In this paper we prove that it is not necessary to recover the homography H explicitly but sufficient to evaluate the displacement of tracked points along epipolar lines in the image. This article describes how these epipolar lines are computed and their relation with the FOE, when the robot moves with translational velocity.