Robot Pose Estimation: A Vertical Stereo Pair Versus a Horizontal One

In this paper, we study the effect of the layout of multiple cameras placed on top of an autonomous mobile robot. The idea is to study the effect of camera layout on the accuracy of estimated pose parameters. Particularly, we compare the performance of a vertical-stereo-pair put on the robot at the axis of rotation to that of a horizontal-stereo-pair. The motivation behind this comparison is that the robot rotation causes only a change of orientation to the cameras on the axis of rotation. On the other hand, off-axis cameras encounter additional translation beside the change of orientation. In this work, we show that for a stereo pair encountering sequences of large rotations, at least a reference camera should be put on the axis of rotation. Otherwise, the obtained translations have to be corrected based on the location of the rotation axis. This finding will help robot designers to develop vision systems that are capable of obtaining accurate pose for navigation control. An extensive set of simulations and real experiments have been carried out to investigate the performance of the studied camera layouts encountering different motion patterns. As the problem at hand is a real-time application, the extended Kalman filter (EKF) is used as a recursive estimator.