Edit Point Cloud in Matic

Hello All!

I’m somewhat experienced when it comes to using Pix4d Mapper, but recently have become frustrated with the lack of Geoid and Vertical Datum support. I also have what I assume is an edge case use, where I need to export ~50 surfaces on every project (mapper requires each surface to be exported separately if you want to retain each file without combining them all into one file)

All of this has led me to evaluate Pix4d Matic and Survey. While I am so far impressed with the speed improvements, there doesn’t seem to be any way to edit the point cloud before processing the DSM. In Mapper you assign what you don’t want on the disabled layer, then process the DSM. According to the Mapper support article: https://support.pix4d.com/hc/en-us/articles/115001671066-How-to-improve-the-Outputs-of-Pix4Dmapper#label6

"How to improve the DSM

The DSM receives as input the point cloud. If the point cloud is noisy, the DSM may also be noisy. Make sure that the results of step 1 and the point cloud are optimal before generating the DSM. However, if the DSM is still noisy, there are some actions to be taken:"

Anyone else experiencing the same circumstances? What have you found as a work around? Or am I missing something completely in front of my face? Any help is appreciated.

Hi DroneLink,
The workflow using PIX4Dmatic and PIX4Dsurvey is a bit different than PIX4Dmapper. The most notable difference is that Mapper derives a DSM from the point cloud. To contrast, Survey will ultimately derive a TIN, or a Triangular Irregular Network from the point cloud. Survey will first perform an outlier filter and terrain filter on the point cloud before generating a grid of points. Once you are happy with the classification then you can generate the TIN. In the end, both programs will generate a surface but it does so in different ways.

Hey @Mike_K,

The problem with a TIN is the resolution isn’t the same. A DSM uses the GSD (usually around 3.2cm for me) whereas a TIN uses the Grid of Points (which I can only get down to .7m grid spacing) resulting in a lower resolution surface.

Here’s a rough screenshot of the difference I can see.


For anyone facing the same issue, here is the workflow I have come up with for the time being.

  1. Calibrate, Densify, Generate DSM and Ortho in Pix4D Matic
  2. Terrain Filter in Pix4D Survey
  3. Export Filtered Point Cloud from Pix4D Survey
  4. Convert las 1.4 file to 1.2 with las2las.exe
  5. Import Photos into Pix4D Mapper and run step 1
  6. Open las 1.2 file in Pix4D Mapper
  7. Generate DSM from imported file
  8. Create XYZ tiles in QGIS from Orthomosaic
  9. Draw Polygons in Pix4D Survey
  10. Export Polygons as .shp file
  11. Use Split Vector Layer (QGIS) for Polygons: Vector > Data Management Tools > Split Vector Layer…
  12. Import DSM and .shp files into QGIS
  13. Extract Rasters from mask layer

Hi DroneLink,
I brought this issue up with our developers and it has generated some interest. Unfortunately I do not have an immediate solution. I would highly suggest posting in the feature request category for PIX4Dsurvey. This is closely monitored for interest amongst topics.

Hi @DroneLink ,

Thanks for that feedback, and the workflow that you use! As Mike said, its an interesting question, and we’re looking at it on the development team. We can certainly shorten the number of steps by quite a lot if we enable a raster output of the current TIN and a min resolution equal to the GSD. You would lose sharpness in the output from breaklines that you have - would that tradeoff be OK for you?

thanks,
Nate

Hey @Nate_Moore,

Thanks for the reply. I think anything to shorten the workflow will be useful. As far as the loss in sharpness from breaklines; are you referring to the setting in the TIN processing options, or the physical breaklines in the elevation of the collected surfaces? If the former, that does not affect my use case.

As to the latter, that would definitely be a downside to me. I deal with slopes under 2%, therefore the precision is key. That’s essentially the reason I found this issue in the first place. The TIN in its current state is not precise enough for my projects.

Thanks for the input, and I look forward to your response.

Screen Shot 2022-03-07 at 5.35.11 PM