Depending upon the settings used in TerraScan’s ground classification routine one may see issues with certain high frequency features, such as berms, dikes and mounds, not being properly classified to the ground class. These features can be more complicated to get into the ground class as their small footprint and frequent occurrence can lead to unwanted low vegetation being added to the ground surface if one makes the ground routine parameters too aggressive. The following are some approaches one may take to remedy these situations.
Probable Resolution #1:
Where possible, decrease the Max building size parameter to a size smaller than the desired feature to be classified. This forces the routine to place one ground point on the feature and increases the chances that the Classification Maximums will not be exceeded when comparing candidate points to the initial surface. The minimum size one may use will be limited by the nominal ground point spacing beneath thick vegetation, otherwise one will end up with ground points in the trees.
Probable Resolution #2:
Use a low fidelity pass with not very aggressive parameters to determine a general idea of the ground. Use that surface to then remove the definite non-ground points from the candidate class(es). Run a second high fidelity ground routine with aggressive settings, but starting with the current ground, to identify additional potential ground points. Selectively, or wholly add these potential ground points to your existing ground as desired.
Probable Resolution #3:
Outline the areas of high frequency features needing to have better classification. Can be done with a fence, but usually best done with a shape element so as to be able to re-run if necessary. Apply a negative Z linear transformation to the dataset, to flip it upside down (Figure 1). This makes the classified valleys now the tops and the unclassified tops now the valleys. Re-run the classification routine in the fence(s) using the same classification maximums. As the unclassified valleys are now below the tops they will become classified. Apply another negative Z transformation to flip the dataset back upright. The area within the defined fence should now have the correct classification. One may accomplish this as a semi-automated tool for ground cleanup by creating a macro that performs the three steps on selected shape elements or an interactively drawn fence.