Random Points: sUAS Mine Site Mapping

A 1.233Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE

We have been developing tools for surface mine site (quarry, stock pile yards) mapping using small unmanned aerial systems (sUAS) for the past several years. Mine site volumetrics are ideal candidates for sUAS technology for a variety of reasons. First of all, the stockpile areas tend to be compact, often less than 50 acres. This makes these sites amenable to flying with a rotary wing sUAS. Secondly, they tend to be free of vegetation. Overhead interference such as vegetation is a real issue for dense image matching algorithms. Evan so, these sites are not without their challenges.

As part of our development, we have surveyed a fair number of mine sites (yes, we do have an FAA 333 exemption!). This have put us in the positon of "eating our own cooking" in terms of our hardware and software platforms. We use structure from motion (SfM) which is often termed dense image matching (DIM) or SemiGlobal Matching (SGM) to construct surface models. These algorithms (implemented in popular commercial programs such as Pix4D Mapper and Agisoft PhotoScan Pro) construct high resolution 3D point clouds from highly redundant imagery (meaning many overlapping photographs). The real advantage of SfM is that it allows light weight, low cost cameras to be used as the sensor. The disadvantage is that the resultant point cloud is a surface model with none of the vegetation penetrating multiple returns that are characteristic of LIDAR sensors (I anxiously await the arrival of reasonably priced focal plane array LIDAR systems!).

Before I go in to some detail, rest assured that the SfM technique of extracting point clouds from images is the single best approach for collecting low cost 3D maps of mine sites. In Figure 1 is illustrated an almost perfect stockpile yard of a mine site in West Virginia. Here we have superimposed the point cloud 3D models (classified to ground and colored orange) over an orthographic image of the site. This entire data set was constructed using SfM software on a set of images collected with a consumer grade Sony NEX 5 camera. Notice that most of the piles are cleanly separated from one another and from surrounding vegetation.

In low activity areas where stockpiles sit undisturbed for long periods of time, vegetation can encroach or even grow on the piles. An example of a stockpile with vegetation encroachment is illustrated in Figure 2. This is an example from a stockpile yard in south Alabama (where vegetation quickly takes over anything not in motion!). Note that this would not be a problem for high density multi-return LIDAR but SfM simply walks over the surface. It is difficult to determine from the orthophoto the impact this will have on computed volumes.

Mine site owners often want contours as a delivered product (along with digital orthophotos). One issue encountered in mine sites is overhanging walls. As shelves are blasted out of the mine, the wall face is often undercut, meaning that the wall curves inward. This results in some very interesting phenomenon when constructing contours. A limestone mine site we surveyed in northern Kentucky has several faces with an extent of over 400 vertical feet. Figure 3 illustrates a section of the wall where an undercut occurs. This results in what appears to be an odd ordering of the contour lines.

A second issue that often occurs in this type of contour rendering is the common use of "2 D" triangulated irregular networks (TIN) in mapping software. 2 D TINs have only a single entry for a unique set of X, Y values. This caused issues on purely vertical walls and especially undercut walls. The most logical solution to these sorts of issues is to directly use the point cloud data rather than flowing contours to downstream engineering software applications.

As the industry continues to adopt dense image matching from sUAS borne cameras, we will need to provide education on the various phenomenon that can occur in the data. While some issues will be ameliorated with the advent of focal plane array LIDAR (e.g. vegetation), understanding the modeling aspects of these sites will remain important.

Lewis Graham is the President and CTO of GeoCue Corporation. GeoCue is North America’s largest supplier of LIDAR production and workflow tools and consulting services for airborne and mobile laser scanning.

A 1.233Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE