Mining: Why In-Pit Rock Fragmentation Measurement Systems based on Photographic Imaging cannot be Reliable
Commercial systems for in-pit fragmentation measurement are varyingly based on photographic imaging and 3D from stereo photographs. These include systems from a variety of established vendors that provide images of the muckpile, truck tray or excavator bucket.
The goal of a fragmentation measurement system is to provide consistent reliable measurement during mining excavation in order to provide reliable feedback to blasting and feed forward to the mineral processing plant.
The key work here is “consistent”, in that the system must be unaffected by changes in the outdoor environment such that measurements at any given time can be reliably compared to measurements over the last week, last month, or last year. Only then, can reliable fragment size distribution information be provided to blasting and processing, in order to optimize these processes.
All fragmentation measurement systems based on imaging employ image processing techniques to delineate the individual rocks and these may be described by the vendor in terms of the different algorithms used, or simply as “Artificial Intelligence”. Discussing the merits of one image segmentation method over another is however not particularly relevant, as the key question is whether the measurement technology is suitable for the application such that there is a robust expectation that the particle delineation could be reliable and realistic given a sufficient particle segmentation algorithm.
In order for particle delineation/segmentation to work consistently the measured data must be unaffected by the outdoor conditions regarding lighting (day/night/direct sun/cloudy), obscurants (dust/rain/snow), shadows and other factors common to an open-pit mining environment.
For photographic systems and 3D from stereo photography this is simply not possible because light intensity variations, sun angles and rock shadowing, equipment shadows, and differences between natural and artificial lighting all contribute to significant and extreme differences in the photographic data and therefore in the particle delineation/segmentation.
No amount of making “particle delineation” algorithms better, can account for the fact that a saturated image does not contain any information in the saturated portion, or that a darkly shadowed part of the image no longer contains sufficiently detailed quantisation in the grey level intensities to allow consistent edge detection and particle delineation, or that dust/rain/snow obscures the view of the rocks.
Even without these fundamental limitations, sometimes the grey level intensities in a photographic image are too similar to notice boundaries between particles and create problems that even a human struggles to delineate correctly. Anyone who has spent considerable time manually delineating photographic images has experienced these problems; grey levels may be too uniform to separate individual particles, strong colour variation in the material looks like a particle edge, or areas-of-fines look like a large rock or a cluster of larger rocks.
Stereo photogrammetry can be used to get 3D data, however this is still based on photographic imaging and therefore still very sensitive to lighting conditions and obscurants. Furthermore, uniform areas of greyscale in the images (also referred to as low texture) results in areas of missing 3D data, which appear as lots of holes in the 3D data. In addition, methods to fill these holes such as interpolation smooth out the 3D data degrading edges of rocks and degrading particle delineation algorithms. 3D from stereo data quality is also highly dependent on the stereo algorithms used and all these factors negatively affect the capacity to measure consistent data and therefore identify particles in a reliable way.
The solution is to use active measurement technology such as laser scanners to collect consistent 3D profile data of rock piles that is robust to both lighting variations including the day/night cycle and airborne obscurants including dust, rain and snow.
Innovative Machine Vision’s Principal Scientist Dr Matthew Thurley has developed world leading research and analysis algorithms over the last 20 years that allows processing of 3D data from laser sources to provide this consistent and robust fragmentation size distribution data. This includes evidence-based peer-reviewed publications on measurement of muckpiles, drawpoints, underground excavator buckets and on conveyor belt. Using robust 3D data with advanced algorithms allows not only consistent particle delineation but also the capacity to detect areas-of-visible-fines instead of mistaking them as large boulders, and detect “ice-berg” like rocks instead of mis-sizing them based on their small visible portion resulting in consistent reliable size distribution measurement.
If you are interested to learn more about the limitations of existing fragmentation measurement systems and current/future possibilities using laser technologies, contact Principal Scientist Matthew Thurley