Point cloud cleaning

[Tanguy Nédélec]

Also often the only option if you are dealing with mobile mapping and/or photogrammetry derived point clouds

On the subject of mobile mapping, I would love to be able to display the points on a "time of measurement" basis. Like, for a given trajectory, display only the points between 00:01:30 and 00:02:00, or so. We could even imagine a little graphical time scale with two sliders, for the start and the end. That way we could clean walks more easily, without them being this massive indivisible pointcloud.

An idea for a future feature for SCC, perhaps?

[MikeDailey]

Also often the only option if you are dealing with mobile mapping and/or photogrammetry derived point clouds

On the subject of mobile mapping, I would love to be able to display the points on a "time of measurement" basis. Like, for a given trajectory, display only the points between 00:01:30 and 00:02:00, or so. We could even imagine a little graphical time scale with two sliders, for the start and the end. That way we could clean walks more easily, without them being this massive indivisible pointcloud.

An idea for a future feature for SCC, perhaps?

The hardest thing to me is the software landscape. So much proprietary stuff going on from basic software down to file types. That being said, I bring this up because I would be able to easily do this but your milage may vary.

First off, we use tech from both LidarUSA and MicroDrones, both of witch will give me a .kmz trajectory. We use TopoDOT for extraction software. I can bring the kmz into topodot and use it as a guideline to then cross section from a path at anypoint and width(time range essentially) along that trajectory. This would not be the most efficient way to clean with this particular software suite though.

[kojabov1]

We possess clients who require the point cloud to be purified. One expeditious approach is to apply a range-based clipping for enhanced efficiency. Although Unify is of cleaning is beneficial, certain clients prefer distinct bubble views.

[dhirota]

Hello all

I just wonder how long point cloud cleaning takes for you.
As we all know, this task could be a nightmare especially when the scene contains:
- Mirrors or windows
- Shiny material
- Moving people or objects
- ...

Cleaning is not mandatory but for certains purposes (scan2BIM, scan2mesh, others), there is no way to escape.

Best rgds

Christian

All this might depend on what system you are using to scan and visualize your client's project. We are still testing our Riegl VZ600i versus our Riegl VZ400i, which we have had for many years.

I have had a different approach from many scanning practitioners over the last 15+ years which has resulted in the vendors we have selected to use: Z+F, Riegl and NavVIS. These three vendors provide excellent imaging with their scanning products. We therefore consider it important to remove objects that do not contribute to the overall visual clarity, such as cars and people. Both Riegl and NavVIS will remove most of the moving objects with office processing automatically. The Z+F system may still require additional scanning from the same location to remove moving objects.

A recent test of the Riegl VZ600i was used to scan 220 scan positions, 1,910 meters along a City of Honolulu sidewalk which was finished in approximately 3.5 hours. Office processing included (registration processed in the field by RTK) colorization of scans, identifying moving objects, and removing the moving objects in an octree based 5mm gridded point cloud and generating an exportable database containing the 7,920 12Mpixel images into 220 scan position panoramas.

Total office time approximately 4 hours.

Removing the moving objects and generating the 5mm point cloud, 2 hours with NO MANUAL LABOR (3.583 Bpts to 1.655 Bpts).

UNCLEANED CLOUD.jpg

CLEANED CLOUD OF MOVING OBJECTS.jpg

[dhirota]

This is an interesting thread to try and figure out what the rest of the scanning world is accomplishing for their clients. I am not sure if the group not cleaning their scans are using only the point clouds for an intermediate process to finish into a BIM or multimedia mode where color and trash within the scan may not be as relevant to clients.

If you are scanning detail under tree canopy in tropical urban areas, the street traffic would probably block a significant part of the scan. This might require scanning both sides of the street, where in this project, we scanned only one side of the street from the sidewalk. This saved a significant amount of time removing the street traffic and people walking on both sides of the street.

UNCLEAN-CAPITOL.jpg

CLEAN-CAPITOL.jpg

The amount of time for us, using the Riegl systems, to scan and process large tropical urban areas appears to be less compared to those people that are using other vendor scanners and processing, especially cleaning and coloring their point clouds. The only references come from LSF post as below:

viewtopic.php?p=107213#p107213

This is what we have done so far in testing the Riegl VZ600i without external camera, the total time and visualization result (the RiPANO, check the thread above).

Those are some nice numbers. My faro unit would take double that total time to just scan, and likely triple including registration/colourization and export.

[landmeterbeuckx]

If i deliver data to a client or do volume calcs i filter on moving objects for the exterior. Not really for interior.

If i draw myself in the pointcloud i don't filter.

Also colorizing pointclouds where moving objects are filtered away makes terrible looking pointclouds because roads are a mixture of the whole rainbow and it's difficult to recognize the data you're looking for. I rather deliver in intensity values or one of the colorschemes from Riscan which fits best the materials used on the public domain.

[smacl]

This is an interesting thread to try and figure out what the rest of the scanning world is accomplishing for their clients. I am not sure if the group not cleaning their scans are using only the point clouds for an intermediate process to finish into a BIM or multimedia mode where color and trash within the scan may not be as relevant to clients.

It comes down to client requirement and final deliverables in my experience

Quite a bit of the work I do is in rail, where there simply isn't the same amount of temporary occlusion as you'd see on roads. Typically, static scanning would be taken from both sides of the track as one of the primary features of interest is the track running edge, which is only ever partially visible from one side. Other items of interest are clearance of locomotive and carriage (kinematic envelope) particularly to gantries, and height and stagger distances to overhead lines for electrification. Some visualization is always nice but scanning is also often accompanied by high res drone photography for more in-depth track condition analysis.

Intensity is also often more important than colour on the urban road surveys that I get as these are typically done at night. The reason for this is that capturing top and bottom of kerbs along with roadside pavement condition is very important and this is best done when there are a minimum of parked cars about. Also far fewer people and moving cars from the point of view of cleaning.

Any motorway jobs I've been involved in tend to be scanned using mobile mapping across multiple passes, where automatic moving object elimination works well. Automatic classification also works well in this environment, way better than in urban road scenarios.

For topographic volumes and DTM extraction, automated ground classification works a treat, so once again, manual cleaning isn't typically needed.

[Philip_127]

When using trimble realworks, removing reflective Items are quite easy and fast. I then classify the project to a Ground & Remaining and that also makes it easier to segment vegetation, cars etc.. Usually takes no more than a hour for a big project but you system specs also has an impact on this