Search

Robustly and accurately localizing vehicles in underground mines is particularly challenging due to the unavailability of GPS, variable and often poor lighting conditions, visual aliasing in long tunnels, and airborne dust and water. In this paper, we present a novel, infrastructure‐less, multisensor localization method for robust autonomous operation within underground mines. The proposed method integrates with existing mine site commissioning and operation procedures and includes both an offline map‐building process and an online localization algorithm. The approach combines the strengths of visual‐based place recognition, LIDAR‐based localization, and odometry in a particle filter fusion process. We provide an extensive experimental validation using new large data sets acquired in two operational Australian underground hard‐rock mines (including a 600m‐deep multilevel mine with approximately 33km of mapping data and 7km of vehicle localization) by actual mining vehicles during production operations. We demonstrate a significant increase in localization accuracy over prior state‐of‐the‐art SLAM research systems and real‐time operation, with processing times in the order of 10 Hz. We present results showing a mean error of 0.68 m from the Queensland Mine data set and 1.32 m from the New South Wales Mine data set and at least 86% reduction in error compared with prior state of the art. We also analyze the impact of the particle filter parameters with respect to localization accuracy. Together this study represents a new approach to positioning systems for currently deployed autonomous vehicles within underground mine environments.

Many mine-site processes such as vehicle operation require localisation systems that are reliable, robust and work in a range of environmental conditions. In underground operations, GPS is not available: solutions instead rely on static infrastructure or expensive, laser-based solutions with limited operational capability. In this letter we present a new vision-based technique, Tunnel-IMage Texturally-Accorded Mosaic (TIMTAM), for sub-metre, infrastructure-free localisation in underground mining environments using a single camera. Our approach stitches upward-facing camera images to form planar mosaic maps, using locations generated by the coarse mapping engine based on a small number of manually anchored locations. Localisation is achieved by refining coarse location estimations with a best fit pixel location for the query image within a search neighbourhood in the mosaic map. Our direct pixel-based method is more robust to the challenging illumination and surface-texture environments encountered in underground mine operations than feature-based techniques. Localisation refinement is only triggered when a confidence threshold for the estimate is exceeded. The system is evaluated in a real world mine tunnel, with results showing that the confidence threshold approach is predictive of the quality of the location estimate refinement, and achieves a reduction in mean localisation metric error of up to ∼ 66% from simulated coarse results.

Many real-world robotic and autonomous vehicle applications, such as autonomous mining vehicles, require robust localisation under challenging environmental conditions. Laser range sensors have been used traditionally, but of- ten get lost in long tunnels that are the major components of underground mines. Recent re- search and applied systems have been increasingly using cameras, bringing in new challenges with regards to robustness against appearance and viewpoint changes. In this paper we develop a novel visual place recognition algorithm for autonomous underground mining vehicles that can be used to provide sufficiently accurate (sub-metre) metric pose estimation while also having the appearance-invariant and computationally lightweight characteristics of topological appearance-based methods. The challenge of large viewing angle variations typical in confined tunnels is addressed by incorporating multiple reference image candidates. The framework is evaluated with real-world multi- traverse datasets featuring different environments including underground mining tunnels and office building environments. The reprojection error of image registration is ∼ 50% lower than a state-of-the-art deep-learning based method (MR-FLOW) using manually-labelled ground truth on a set of images representing typical scenarios during the underground mining process.

This work presents Semi-Supervised SLAM - a method for developing a map suitable for coarse localization within an underground environment with minimal human intervention, with system characteristics driven by real-world requirements of major mining companies. This work leverages existing information common within a mining environment - namely a surveyed mine map - which is used to sparsely ground map locations within the mine environment, increasing map accuracy and allowing localization within a global frame. Map creation utilizes a low cost camera sensor and minimal user information to produce a map which can be used for single camera localization within a mining environment. We evaluate the localization capabilities of the proposed approach in depth by performing data collection on operational underground mining vehicles within an active underground mine and by simulating occlusions common to the environment such as dust and water. The proposed system is capable of producing maps which have an average localization error 2.5 times smaller than the next best performing method ORB-SLAM2, comparable localization performance to a state-of-the-art deep learning approach (which is not a feasible solution due to both compute and training requirements) and is robust to simulated environmental obscurants.

Autonomous vehicles are increasingly being used in the underground mining industry, but competition and a challenging market is placing pressure for further improvements in autonomous vehicle technology with respect to cost, infrastructure requirements, robustness in varied environments and versatility. In this pa- per we introduce several new vision-based techniques for visual place recognition on under- ground mining vehicles that improve on cur- rent available technologies while only requiring camera input. We present a Shannon Entropy- based salience generation approach that enhances the performance of single image-based place recognition by selectively processing image regions. We then add a learning-based scheme realised by Support Vector Machines (SVMs) that filters problematic images from both the reference map databases and from live processing. We evaluate the approaches on new large real-world underground vehicle mining datasets, demonstrating ∼ 107% improvement in area under the precision-recall curve than a state-of-the-art place recognition algorithm, and demonstrate its generality on a non- mining-based benchmark dataset. Together this research serves as a step forward in developing domain-appropriate improvements to existing state-of-the-art place recognition algorithms that will hopefully lead to improved efficiencies in the mining industries.