A baseline dataset for performance evaluation of visual detection and classification techniques in mining environments

Project dates: 03/01/2018 - Ongoing

Technology Aim

Scenarios where there is a need to safely negotiate a dynamic environment are ubiquitous in many industrial settings such as open pit mining where haul trucks, light vehicles, personnel and various excavators travel in and out of the pit with frequent interaction. As the industry progresses from on-board human operators to tele-operated and semi-autonomous operations the demand for intelligent perceptual sensing technology, capable to understand the environment that surrounds them, increases.

The short-term aim of this seed project is to build a realistic baseline dataset to evaluate the performance of vision-based state-of-the-art scene understanding algorithms, in particular for the tasks of object detection and classification, in mining environments. The data will be collected by a common sensor suite in a few different mine sites and in a variety of environmental conditions. It will then be labelled to provide ground truth.

A follow-up project would then conduct evaluations of the performance of object detection and classification algorithms against the needs of the industry. The data could also be used to evaluate other aspects of scene understanding such as the identification of hazardous elements in the environment.

Industry Benefit

Short-term benefits include:

  • A realistic baseline dataset, with ground truth, available to evaluate the performance of state-of-the-art scene understanding algorithms, in particular object detection and classification (e.g. for the detection of
    people, light vehicles and other assets), in realistic mining environments
  • A straightforward potential extension is the evaluation of methods to track dynamic objects in mining sites, e.g. to prevent collisions or better understand and optimise operation

Longer-term benefits include:

  • The development and deployment of reliable vision-based object detection and classification methods, capable of detecting and recognising personnel, light vehicles and heavy vehicles from significant distances, night and day, including in difficult environmental conditions
  • The development and implementation of advanced collision avoidance techniques benefitting from the aforementioned object detection and classification technique
  • Substantial improvements in safety of mine sites thanks to the aforementioned method
  • Passive monitoring of assets and personnel on mine sites without the need to fit all of them with a GPS receiver.
  • Future improvement in productivity and general operation of mine sites thanks to the aforementioned monitoring technology


Funding / Grants

  • Mining3 (2018)


Other Team Members

Dirk Lessner