Estimation of Robot Position and Orientation Using a Stationary Fisheye Camera

A core problem in robotics is the determination of the location and pose of a mobile robot in its environment. The localization is a basic operation, which must be successfully carried out in complex environments using imprecise and/or contaminated data and is essential for a broad range of mobile robot tasks, since the robot behavior depends on its position. In this work, we propose the use of a stationary fisheye camera for real time robot localization in indoor environments. We employ a model for the formation of the image by the fisheye camera, which can be used for accelerating the segmentation of the robot's top surface, as well as for calculating the robot's true position in the real world frame of reference. The proposed algorithm for robot localization exploits the calibrated fisheye camera model and the known dimensions of the robot, whereas it does not depend on any information from the robot's sensors and does not require visual landmarks in the indoor environment. Furthermore, the pose (orientation) of the robot is determined using a triangular shape placed on top of the robot's flat top surface, using Hu's moment invariants, appropriately modified using the calibrated fisheye camera model. Initial results are presented from video sequences and are compared to the ground truth position, obtained by the robot's sensors. The dependence of the average positional error with the distance from the camera is also measured.