Depth estimation for autonomous robot navigation: A comparative approach

Abstract

Depth estimation has long been a fundamental problem both in robotics science and in computer vision. Various methods have been developed and implemented in a large number of applications. Despite the rapid progress in the field the last few years, computation remains a significant issue of the methods employed. In this work, we have implemented two different strategies for inferring depth, both of which are computationally efficient. The first one is inspired by biology, that is optical flow, while the second one is based on a least squares method. In the first strategy, we observe the length variation of the optic flow vectors of a landmark at varying distances and velocities. In the second strategy, we take snapshots of a landmark from different positions and use a least squares approach to estimate the distance between the robot and a landmark. An evaluation of the two different strategies for various depth estimations has been deployed and the results are presented in this paper.