Evaluating the effect of illumination on the performance of visual odometry in underground mining environments

Project dates: 01/02/2017 - Ongoing

Project Overview

Visual Odometry (VO) is a vision-based localisation technology which has many potential uses and benefits for the mining industry, especially in GPS-denied environments such as underground mines. However, the performance of VO is known to degrade in these low-light conditions. This study evaluates the effect of illumination on the performance of VO in underground mining environments to identify suitable illumination configurations that should be used to obtain the best performance of VO in these environments.

This project is the Masters of Philosophy conducted by Ian Greyvensteyn with the support of Mining3, supervised by A/Prof. Thierry Peynot (QUT & Mining3), Prof. Michael Milford (QUT) and Prof. Paul Lever (Mining3).

Project Summary and outcomes so far

The effect of illumination is quantified by acquiring and evaluating data from an operational underground mine in Australia. The data consists of kilometres of high-resolution stereo images illuminated with five different light configurations and includes ground truth measurements acquired from an automated total station. The data are evaluated using state-of-the-art direct, feature-based and semi-direct VO algorithms and current evaluation metrics. Results show that light does affect the performance of VO in underground mining environments and that the most considerable improvement in accuracy is achieved using light techniques that enhance contrast in the images. For example, in our experiments significant decreases in Relative Pose Error (RPE) were observed when the light source was moved away from the camera and when direct light was used compared to diffuse light. Brightness and contrast-enhancement improved the accuracy of the VO algorithms, but also reduced the impact that the different light configurations had on the performance of VO. Direct methods and semi-direct methods were shown to be more sensitive to different configurations of light compared to feature-based methods. The research quantifies the effect of three key attributes of light on the performance of VO and identifies lighting strategies that are likely to improve the performance of VO in underground mining environments.


Funding / Grants

  • Mining3 (2017 - 2020)

Team

Other Team Members

Prof. Paul Lever (Mining3)