Toe Extractor Point Cloud Task

*This feature is available in the Advanced license level of LP360.

This is a powerful new tool that automatically creates a polygonal toe for volumetric computations of clean stockpiles. Optionally, it can also automatically classify overhead structures that are well separated from the pile such as conveyors.

A "clean" stockpile is one with a clear demarcation of the boundary from a 3D perspective. That is, if you were to run a profile line across the pile, you could clearly distinguish the base at all points along the pile.

Example of a "Clean" Stockpile

Example of an Ill-Defined Stockpile. It is adjacent to a steep slope without a clear delineation between them

How to use the Toe Extractor

1. Use the Add Files command to open the LAS files that contain the data you want to use.  

2. Use the Add Task command to create a new point cloud task. When you are selecting a Task Type in the Add Point Cloud Task dialog, select Toe Extractor. Once you click OK, you should see the unique name that you gave your task selected in the list and the Toe Extractor property page displayed on the Point Cloud Task tab.

3. Specify the settings for the Toe Extractor operation in the Properties page that displays.

4. If you are using the interactive method of digitizing each seed point (i.e. you have the IO Manager input set to Tool Geometry) then the task is executed by clicking the Point tool on the Point Cloud Task toolbar (it will be the only enabled tool) and clicking near the top of the stockpile for which you wish to find the toe.

The Point tool

If you are executing the task from points preset in a shape file or layer (as set by the input method in IO Manager), then the task is executed by pressing the Project button on the Point Cloud Task Toolbar.

The File tool

Overview of the Toe Extractor Point Cloud Task

Input Dataset

• Input Geometry - The Toe Extractor requires a location point to indicate the area of extraction. This point can be input by interactively digitizing a point ("Tool Geometry"), using points (2D or 3D) from a Shape file, or using points (2D or 3D) on a LP360 Map Layer. It can be set within the I/O Manager by clicking the (...) button.

Source Points - This is the standard Point Cloud Task (PCT) Source Points setting dialog.  Use it to filter the point cloud points that should be considered part of the stockpile and surrounding ground when extracting a toe.  If no data have been classified, you can leave this set to the default (filter nothing).  If, on the other hand, you have classified overhead points into some specific class (for example, building or conveyor), exclude these points from consideration.

Excluding the Conveyor class from consideration when determining the toe position

Normally you would only use a source point filter if you had pre-classified some data to clean up the area prior to toe extraction.

Units - The Units setting allows you to optimize parameters in units of Feet or Meters, regardless of the units of the data. You will want to make sure that the Map View units are set to the units of the point cloud data.

Toe Parameters

• Maximum Size - This setting is used simply to prevent a “runaway” algorithm. It is used to specify the maximum area enclosed by a toe. The default is 400 feet. The Maximum Size parameter defines a search square, Maximum Size on edge, centered at the “Seed” point.
  
  
  Example of a large stockpile being "squared off". When you notice this, increase the Maximum size parameter.

• Grid Size - Generally if you set this value by sampling the data with the eyedropper tool, the resulting value will be correct. This is achieved by pressing the eye dropper tool and drawing a polygon in the area where the toe is to be computed. The polygon is drawn by left clicking at each desired vertex and double clicking to end. 
  
  
  Grid size sampling tool
  
  The point cloud data in the neighborhood of the toe computation is partitioned into cells (grids) that are Grid Size on each edge. Some statistics such as minimum Z, maximum Z and so forth are computed for each cell. When the toe extractor looks for slopes, it does so at the horizontal granularity of the grid size. There must be sufficient points within each cell to compute these statistics. Thus the grid size must be two or more times the nominal point spacing of the point cloud data. A grid size that is too small will cause the algorithm to fail. A grid size that is too large will cause either a very coarse toe (jagged) or will result in too much level surface being included in the toe. 
  
  It is recommended that you use the eye dropper tool.

• Minimum Vertical Change - This value should be set to roughly the sum of the absolute vertical accuracy of the point cloud and the size of the aggregate of the pile (for example, you can assume the size of sand is zero). For point clouds extracted from dense image matching, this value should be in the neighborhood of 10 cm. For coarsely spaced LIDAR data, it should be a bit larger (say 25 cm). This and the grid (cell) size are the most critical tunable parameters. The toe algorithms steps “out” from the seed point in all directions and looks at the vertical change in all of the cells over the step distance. If the vertical change exceeds the Minimum Vertical Change, the algorithm assumes that it is still traversing down the stockpile (or up, if it happens to be “hunting” in the uphill direction). If this occurs, the algorithm includes the current cell in the toe “cluster”, steps to a new, adjacent location and continues to search. A minimum vertical change larger than needed will cause the algorithm to step out well beyond the actual toe.
  
  
  Example of the Minimum Vertical Distance set too large
  
  A minimum vertical change smaller than needed will cause the algorithm to hunt around in vertical noise, resulting in a noisy toe.
  
  
  Minimum Vertical Distance set too small
  
  The default value is 0.25 feet.  This works well for a wide variety of data from Dense Imaging Matching (sUAS) data.  You will typically need to increase this value for LIDAR data or vertically noisy data.

• Grow/Shrink Size - This setting simply draws a grown or shrunken toe by the input value amount in the chosen units. The option defaults to 0.00, meaning nothing changes. Setting this to a positive value grows (buffers) the automatically located toe by the specified amount. Setting the value negative shrinks the toe. Note that automatic overhead classification occurs after the buffer operation so overhead with respect to the expanded/contracted toe is targeted for reclassification.

Here, the value is set to shrink the toe by -5.000 feet.

Here, the value is set to buffers the toe by 2.000 feet.

Overhead Points

• Exclude Overhead Points - To enable the exclusion of overhead points, check the “Exclude Overhead Points” option.

• • Overhead Gap Size - The algorithm for overhead point investigation analyzes each cell for the vertical range of the elevation within the cell.  If that range exceeds the setting of Overhead Gap Size, the cell is considered to contain overhead points that should be excluded from the toe extractor computation.  Thus this value should be set larger than the anticipated vertical dimension of the surface components of the cell (due to both slope and noise) but not so large as to consider overhead structures to be part of the stockpile.  The default setting is 1 foot.  You will need to increase this for steeply sloped stockpiles with large grid sizes.  For example, a stockpile with a slope of 45 degrees and a grid cell of 5 (this would be an example of very large nominal point spacing LIDAR data) would have a natural vertical displacement of 5 feet for the surface material over the span of the cell.  Thus an Overhead Gap Size less than 5 would potentially exclude some of the surface points. 

  • Std Dev From Min - Once a cell is determined to have overhead points (as established by the Overhead Gap Size), a computation is performed on each point to see if it should be included as part of the stockpile surface for toe computation or considered an overhead point.
    
    The standard deviation in the vertical direction (σ_z) is computed for the points in the cell.  Any points that are within Std Dev from Min are considered to be part of the stockpile surface and will be included within the surface computations.  Any points beyond this value are considered overhead points. A value that is too large will result in overhead points being included in the toe computation.  
    
    A value that is too small will exclude some surface points in the toe computation.  The default value is 0.25 feet. 

• Classify Overhead Points - Checking this option and setting a destination class (via the Destination Class… button) will move all points considered to be in the Overhead category to the specified destination class.

Output Dataset

• Output File - The output file determines the location of the file containing the extracted toe. It can be set within the I/O Manager by clicking the (...) button. A good practice is to use the Project Path as the root location of the output file. This allows you to use the same task output file setting, regardless of project location.

Forcing a Toe Location

The current version of LP360 Standalone (and LP360 for sUAS) does not include vector editing tools. This means that if the toe does not place correctly using the automated tool, you have to delete it and hand digitize the toe using the LP360 Conflate tool (you have to use Conflate to ensure that the Z is properly set for each vertex).

This can be aggravating when the toe is almost correct for a complex stockpile. An example is shown below.

Toe wandering over a slope edge

Here the rather complex toe is correctly digitized on the southeast side of the stockpile but wanders down the slope on the northwest side. The inability of the algorithm to trace the pile is being caused by the stockpile “falling” over the steep embankment toward the northwest side (the upper left of the toe). Note that we are limiting the distance that the toe can wander by setting the Maximum Size parameter. This can significantly reduce the time you have to wait to see if the extraction will be correct.

One way to address this issue is to classify the points where the wandering occurs and then exclude this class in the Source Points filter. 

Note in the figure below that we have classified points along the offending edge to the Low Vegetation class.  These points are not actually low vegetation – we are simply using this as a convenient choice.

Limiting the toe extent via classification

The Source Points filter is then set to exclude the Low Vegetation class by unchecking this class.

Excluding Low Vegetation from the toe computation

Running the toe extractor (with all default parameters except Source Points) now results in the toe below.

Tuning Procedure

The Toe Extractor default parameters are optimized for Dense Image Matching (DIM) data sets collected using cameras on a small Unmanned Aerial System (sUAS) followed by point generation using a tool such as PhotoScan or Pix4D. These point clouds typically have ground sample distances (GSD) of 5 cm or so and vertical noise at a comparable level.  If the GSD varies greatly from this or the noise is larger, it will be necessary to tune the default parameters.

When trying to change the performance of the toe extractor by adjusting parameters, it is important to adjust only one parameter at a time. 

When adjusting parameters, we recommend that you set the Maximum Size parameter to some reasonably small value (say start with 20) to limit the length of time the algorithm will run when you have ill adjusted parameters.  

First set the Grid Size parameter using the eye dropper tool.  In some rare cases, you may have to manually set the Grid Size parameter.  This should be at least twice the nominal point spacing of the point cloud data. 

If the default for the Minimum Vertical Distance is not giving good results, have a look at the extent of the noise and aggregates in the profile view.  Try to start this parameter so that it reflects this range.  Again, set the Maximum Size parameter to a small value while tuning to limit your wait time.  Do not vary the Minimum Vertical Distance parameter by large amounts – try 0.25 or so increments both above and below the default value.  Note that values below about 10 cm can usually create bad polygons and they take a very long time to run!

Note that if there is no separation between areas of the overhead points and the stockpile, the overhead separation/classification will result in erroneous classification.  This can often occur if the conveyor was in operation during the imaging operation.  Rocks falling from the conveyor to the stockpile are imaged and included in the surface model.   All you can do in these cases is manually clean up the overhead/pile points with the manual classification tools.  You should note, however, that if an adequate separation does exist everywhere along the toe line (which is usually the case), a proper toe line can still be automatically extracted.  This was the case in the example below.  In these cases, use the exclude overhead points option but do not use automatic classification (unless you find it more expedient to clean up the misclassification rather than manually classify the overhead).

An area with no vertical separation between stockpile and overhead

It is sometimes the case that the automatic toe extractor simply cannot find a proper toe no matter how long you play around with the parameter settings.  In this case, simply hand digitize the toe using one of the conflate tools.