Manual/Automatic classification and automatic segmentation for small photogrammetric datasets.
Goal: extracting rocks from the ambiant background (ground) and segment them so that you can export individual point clouds for further processing.
Methodology:
Sofwtare = CloudCompare Beta 2.8 (http://www.danielgm.net/cc/release/ it has to be this one because the CSF filter is not included in previous version)
1. Manual classification with heightmap.
The easiest way to classify your data if you have a highly contrasted and flat dataset (which is almost never)
- Clone you PCL (pointcloud) to keep the original RGB information somewhere (if relevant)
- Compute the heightmap as RGB
- Convert the RGB values as Scalar Fields
- Pick the relevant classification values with the Scalar Field histogram
- Proceed with "select by values" to extract the relevant part of your data
- Start again if you need multiple classification parameters
- Clear the rgb colors from each extracted PCL and transfer the RGB values from the cloned PCL (if relevant again)
2. Automatic classification with CSF plugin (see CloudCompare documentation for more informations http://www.cloudcompare.org/doc/wiki/...) )
A more robust alternative
- It does not work with very small datasets (here around 4m²) so we have to scale up the PCL to trick the plugin into thinking it's a relatively big area
- Still, I recommend using the finest settings to get good results with this very example
- In the end, you get two PCL with extracted features
3. Automatic segmentation
- If your extracted features which are somehow isolated one from another, you can run the segmentation tool (Tools - Segmentation - Label Conncted Comp)
- You get in return a list of each feature as a separate PCL ranked in descending order of volume
- The point here was very specific because we need to export each feature separatly to run surface and volume analysis in another software.