Robot motion estimation with a multi-camera system

In this paper, we propose efficient solutions to estimate the robot motion with a multi-camera system. Specifically, we focus on the case that the camera is rigidly mounted on the robot, which roams in indoor environments. In this case, the roll and pitch angles with respect to the ground plane can be initially estimated, and we only need to care about a 3-DOF pose estimation problem. We derive closed-form and approximate minimal 3-point algorithms based on the hidden variable technique. We show that they lead to a 6-degree univariate polynomial with up to 6 real solutions and a 3-degree univariate polynomial with up to 3 real solutions, respectively. Experimental results show that the proposed algorithms are robust enough for real applications and significantly faster than the existing algorithms.