CloudCompare forum

I wanted to post my methods for creating a digital elevation model in cloud compare. I am referring to creating a new point cloud from raw point cloud data which only contains points/hits associated with the bare earth (ground hits). FYI my point cloud is collected by using an aerial UAV system through photogrammetry so noise, trees, and man-made objects are what need to be removed. My main purpose for this is to create contours.

Not only I am hoping that this will help people but I am sure some of you have your own methods and may be able to give advice on how this process can be best performed/refined.


Digital Elevation Modelling using Cloud Compare

1.) Open Point cloud

2.) Remove Outliers Using Statistical Approach (on main interface, SOR)
a. > Number of points to use for mean distance estimation: 10 (Default)
b. > Standard Deviation Multiplier Threshold: 1.00 (Default)
(10 and 0.1 seem to work well for me though I am not entirely sure about the math behind this, more of a guess and check approach)

3.) Manually remove buildings and known infrastructures.
a. > Select the cloud in the DB (Database) tree
b. > Select the ‘segment’ tool on the interface (scissors)
c. > Manually click around the object you want to segment
d. > Right click to close the polygon
e. > Select ‘segment out’ in the segment toolbar
f. > Continue selecting points which you want to segment*
g. > Select ‘Confirm Segmentation’ in the segment toolbar
h. > Viewing only the segment which you just created, tilt it such that the ground can clearly be distinguished from the structure and use segment again to remove it.

4.) Segregate bare earth surfaces (pits, roads, barren areas) from vegetated or noisy areas (forests, water). Use segment tool again

5.) Resample bare earth point cloud to 1.00m (or whatever step size is necessary) using the average point.
a. > Tool
b. > Projection
c. > Rasterize
d. > Step: 1.00
e. > direction: Z
f. > cell height: average height
g. > Fill with: leave empty
h. > Export: click ‘Cloud’

6.) Resample vegetation point cloud to 10.00m (or whatever step size is necessary) using the minimum point.
a. > Tool
b. > Projection
c. > Rasterize
d. > Step: 10.00
e. > direction: Z
f. > cell height: minimum height
g. > Resample Input Load ‘Checked’
h. > Fill with: leave empty
i. > Export: click ‘Cloud’

7.) Manually remove noticeable noise/vegetation that is remaining using the segment tool.

8.) Export Contours
a. > Tool
b. > Projection
c. > Rasterize
d. > Step: 1.00
e. > direction: Z
f. > cell height: minimum height
g. > Fill with: interpolate
h. > Export: click the Contour Tab
i. > Just minimum vertex as needed to avoid producing tiny contours*
j. > Select Generate
k. > Select Export

I generally finish the contours in ArcGIS by removing small ones (<20m in length) as well as smoothing the contours as they can be very rigid/messy from small areas of interpolation as well as area of very dense point coverage.

*EDIT 1(21/01/2016):

Here is a before and after example of this process. Both figures show point clouds overlaid on the associated aerial photography. As you can see in Figure 1, the point cloud is very dense.


Figure 1: Raw point cloud with associated imagery.

As seen in figure 2, this process does a good job of only selecting points which are associated with the gaps between trees. The spacing can be adjusted based on the density of the vegetation cover and how dense you need your resulting point cloud to be.


Figure 2: Raw point cloud with associated imagery.

Thanks for the contribution!

2) indeed the default parameters are good to use in most cases. You could increase the number of points if you had a very high density with a lot of noise. And the the std. dev. let's you control how strong the filter is.

3) Once you click on the inside or outside button, you can rotate the camera without leaving the tool (it's 'paused'). Then you can proceed with the segmentation by unpausing it - use the dedicated button or the space bar). This way you can avoid leaving and restarting the tool too many times.

4) Have you looked at the Canupo plugin? It can performs automatic classification. And once tamed it's very powerful.

8) You can increase the 'Min. vertex count' to avoid creating too small contours

daniel wrote: 3) Once you click on the inside or outside button, you can rotate the camera without leaving the tool (it's 'paused'). Then you can proceed with the segmentation by unpausing it - use the dedicated button or the space bar). This way you can avoid leaving and restarting the tool too many times.

This is a great piece of advice! It should speed up my processing time and prevent me from having file names in the DB tree like .segmented.segmented.segmented... to infinity Haha!

daniel wrote: 4) Have you looked at the Canupo plugin? It can performs automatic classification. And once tamed it's very powerful.

I have glanced at this plugin but I'll will now look into it. I'm curious as to how well it will separate trees from the small areas of bare earth/low lying ground vegetation between them.

Yes. And if necessary don't hesitate to post your cloud (or parts of it at least) so that we can help you with the classifier creation.

Hi Andrew, could you please contact me? thanks

Andrew and Joak-

I have been working on this same workflow. Also incorporating in a CSF depending on the type of landscape, etc. Feel free to message me if you'd like to chat about it.

Thanks