Hazards of all types on mine sites are often contextual or industry-specific. In mining, a static vehicle is typically not a hazard, but if placed on a dangerously steep slope or next to delicate equipment, it becomes a possible roll or crushing hazard.
Dangers such as trip hazards also depend on context – a ladder placed vertically against a wall is generally not a trip hazard, but if lying flat on the ground it may become one. Rubble or dirt is typical on mining sites – but if in the wrong place, it can be a trip hazard or risk for air intakes.
Most camera-based hazard-detection is based on recognition of objects or things in the scene, without an understanding of the context within which that object is placed. Automatically recognizing hazards based on context would provide the mining industry with the ability to automatically monitor the safety of their working environment using person- and vehicle-mounted cameras, static surveillance cameras and robot-borne cameras (such as drones). Such a capability would have two primary benefits: 1) the enhancement and augmentation of existing, human-driven safety assessment procedures e.g. safety inspectors and 2) the automation of safety monitoring processes.
In this project we undertook a review and preliminary evaluation of existing technical solutions using vision sensors only. This included:
- a review of existing research knowledge and current state-of-the-art in hazard detection techniques,
- a review of existing commercial solutions and technology, and current manual (e.g. done by a human) safety monitoring processes
- initiatory experimental setup and data collection, and
- preliminary testing and basic evaluation of selected approaches.
Funding / Grants