Solving the Monocular Visual Odometry Scale Problem with the Efficient Second-Order Minimization Method

Abstract

Estimating the localization of an autonomous robot is a key problem in robotics. Monocular cameras have been successfully used to solve this problem, but they suffer with the absence of depth information. It limits measurements made with only monocular cameras to be recoverable only up to a scale. This condition is known as scale ambiguity. In this paper we propose an approach to solve the scale ambiguity problem using a single monocular camera and with knowledge of its mounting point. Our solution uses the Efficient Second-Order Minimization method to track a patch of the road through the frame sequence and estimate the motion scale between frames. The scale estimated is used to update the motion estimated through a standard monocular visual odometry method.