Human detection for underground autonomous mine vehicles using thermal imaging

Abstract

Underground mine automation has the potential to increase safety, productivity and allow the mining of lower-grade resources. In a mining environment with both autonomous robots and humans, it is essential that the robots are able to detect and avoid people. Current pedestrian detection systems and the reasons that they are inadequate for mining robots are discussed. A system for human detection in underground mines, using a fusion of three-dimensional (3D) information with thermal imaging, is proposed. The system extracts regions of interest and classifies them as human or background. The scene excluding the pedestrians is assumed to be static and is intended to be used to determine the ego motion of the vehicle. In addition to the thermal camera, a distance sensor will provide depth information and allow the calculation of the vehicle and pedestrian velocities. Various classification methods are compared and it is shown that a neural network provides the best results in terms of speed and accuracy. The results of tests on two 3D sensors indicate that further work is required to determine the effect of the harsh environment on the accuracy of the sensors.