Open Cut » Rock Mechanics
Rigorous geo-hazard management practices require accurate spatial data to classify the type of instabilities, monitor their activity and predict their evolution. Terrestrial laser scanners and ground based radar systems, often implemented in mobile vehicles, are widely used within the mining industry to conduct periodic surveys of extended areas and track deformations in almost real-time. Such systems are very accurate and operate in various weather conditions, however, their utilisation is complex and their costs can be in excess. This project aimed to develop a more cost effective and smart solutions that allow capturing all types of instabilities with real-time updates.
This project sought to re-design a prototype of a fixed-base stereophotogrammetry (stereo-vision) monitoring system specifically devoted to surface mining applications, the implementation of an automatic processing pipeline and the development of a user friendly interface. Several issues and limitations found during the proof-of-concept phases were addressed in the current project.
The camera was split from the rest of the electronic hardware. This resulted in two different enclosers but increased the flexibility considerably. The new camera enclosure was designed to accommodate different cameras and lenses allowing a broader applicability of the system at different distances. A mounting bracket was designed to allow easy fine adjustment of the orientation of the camera, crucial for increasing flexibility and facilitating easy and quick installation.
The extensive analysis showed that a test-field calibration or strip calibrations (with or without GCPs) using several images should be preferred compared to self-calibration using two images only. If one of
these two calibrations cannot be performed, a TLS point cloud can be used to improve the calibration obtained from two stereo pair images.
A fully automatic processing pipeline for generating 3D models (including change detection and monitoring) at user-defined frequencies was implemented. The processing is achieved via three components: a processing server, a database and a graphical web interface. The processing pipeline was developed by the University of Parma and is referred to as Slope Monitor.
The web interface was designed to provide the user with the ability to manage the automatic processing pipeline and set parameters for the change detection analysis. One of the challenges in this analysis was the identification of true events. For this project, it was decided to focus on rockfall detachments (i.e. negative changes only). The initial analysis identified too many possible rockfall events, hence, two different threshold filters were implemented. This increased the reliability of detecting true events but resulted in an underestimation of the rockfall volume.
The monitoring system was installed at a mine site for 11 months at an old weathered highwall due to safety and operational constraints. The data was subdivided into 9 periods based on rainfall events since a correlation between rainfall and rockfall events was expected. The analysis allowed identifying the number of rockfall events per period, their approximate volume and shape. It was also possible to identify from which material layer the blocks detached. The recorded data allows mine operators to investigate the correlation between rockfall events (including size, shape and material) and rainfall while providing an accurate estimation of the rockfall frequency, which is crucial for a comprehensive risk assessment.
The algorithm still identifies a considerable number of false positives (around 30%) from items like strong shadows, moving machinery, vegetation and animals. Currently each event needs to be verified by the user. A solution to this issue was proposed in a follow-up project that suggests the use of deep learning models to detect changes and movements more accurately.