Hello,
I have flown a concrete street in town and processed it. My drafting software is creating some erratic cross sections on what should be a smooth road with curb and gutter. I zoom in close to look at the densified point cloud and it appears that there are piles of pixels in the middle of the road? Obviously this is relating to the erratic cross sections. What can I do to improve this? Also the face of the curb and gutter tends to be smoothed out and not abrupt as I would like it. I have tried several settings in Pix4D including medium and sharp setting in DSM/Mosaic. The point cloud settings are Image Scale = 1/2; Point Density = Optimal; Min # of matches = 5; using both filters on. Any help would be greatly appreciated.
Thanks,
Hanns.
It sounds like you need to make MTP’s, surfaces, GCP’s, etc. They will help the images “smooth out” and get rid of some of the noise (piles of points). In order to have an abrubt surface on the curb you will have to have some oblique images in your project as well as nadir, just like a building facad. That would be where I would start. Hope it helps.
I am getting this too would be good to know the best ways and software to filter out the noise
I agree with Whitnie, you can always add GCPs, manual tie points and if needed remove points.
You can as well clean the point cloud with external software and then import the point cloud and process step 3 with the imported, but in general is not needed.
And if you want a nice DSM/mosaic, then you can add surfaces in the raycloud, that will help for dsm and orthomosaic, and additionally, edit the mosaic with the mosaic editor.
Any luck Hanns?
I am currently doing a large-ish linear urban project (approx 9km long with atleast 3 times that in roads) flown by a client and getting the same result on my roads (mix of chip seal and bitumen). I am using similar settings to you as well (i am using min# =6). It is worse in areas of shadow but not all shadows. It is difficult to determine a reason other than false / poor feature matches on the road.
MTPs is not an option for me as it would require such a large number of them and the time to implement them would be longer than practical. Will lowering the scale help with the point cloud by generalising features at the expense of accuracy?
Thank You all for your advice! I flew an area at a lower altitude and increased my photo overlap to 85% front and 75% side overlap. This really seamed to help. I also added some obliques as Whitnie suggested. Thank you. I generally think that it was the way I initially collected the data. Josh what is the external software that you use? Josh when I add a surface in the raycloud, are you suggesting that I surface say the concrete pads and curbs? To be clear on this “Surface” idea, first I need the initial point cloud densification correct? Then add these surfaces in the raycloud and then reprocess the point cloud and Mosaic? Gareth, I’m not sure that will help. I would like to get the DSM in better shape though so I can use it to create contours in Microstation.
Thanks, Hanns.
Hanns, Did you ever find a solution? I am having the same problems with piles of points on a flat road and I am unable to contour my projects in AutoCad. Thanks
Â
Tom
@Hanns: Regarding the surfaces, I believe that Josh meant something similar to what is described in this article:Â
https://support.pix4d.com/hc/en-us/articles/202560449 . The idea is to add a surface on the road, or the sidewalk once the densified point cloud is created (after step 2) and then check " Use for DSM and Triangle Mesh" in the surface options in the right hand panel. Then, you can run step 3 and the DSM should be flat (i.e. without noise) in the areas with surfaces.
@Josh: I’d also be curious to know about the external software you use to clean the point cloud.Â
@Tom: Please check the previous comments on the post as well as Hanns’ last comment. These should give you hints as to how you can proceed to remove the noise in your project.
Some good suggestions in here. I have struck this issue many times in the past myself, and have not been able to resolve it fully. While adding a surface works well for car parks, roads are not flat. They have a rounded peak to assist water run-off. Hence putting a surface will flatten the road, would it not?
Â
Roads generally lack features to match using photogrammetry.
Â
LiDAR is the most viable option, once the price of airborne (UAV) sensors drops - which it soon will, thanks to the race for Autonomous Vehicles. In the meantime, we minimise point cloud noise and manually edit when needed.
Â
Â
Â
Â
I agree with Peter. These methods do not help much for curved roads, as they are not planar surfaces.
I flew a site with the highest overlap settings in Pix4dCapture, doing a set of 3 aerial grids, and collecting 897 photos. My drone is a DJI Mavic 2 Pro. I flew 4 hours before sunset to avoid shadows. My goal was to collect data as conservatively as possible to see if Pix4D could eliminate the piles of pixels hovering above roadways.
My processing settings were as follows:
My pictures were taken at constant altitude (150 feet) with I believe a 70 degree camera angle as follows:
My ground sampling distance was reported to be 1.17 cm/0.46 in.
I found that for concrete portions of roadway, the program did pretty well:
But for asphalt it did not:
I have some oblique shots that I took from a manual flight at a separate time that has more shadows that I am considering merging with this project to see if things improve. Then I would be at 1000+ photos and processing would likely take 10+ hours.
I use this point cloud to perform reverse projection photogrammetry to place cameras that captured video at this site. Therefore, I need reasonably accurate points on the road for my calculations. If nothing else, this point cloud makes for a somewhat unprofessional display as an exhibit. I also sometimes need to make smooth meshes from cross-sections of the roadway starting with a NURBS surface. The current point cloud seems difficult to extract the actual cross-sections.
I will try cleaning up the point cloud using CloudCompare, but I would really like it if someone found a way to improve this automatically in Pix4D, either through intervention in the software, or during data capture.
Hi Evan,
When creating the point cloud in stage 2, I would recommend unticking the Multiscale option, and increasing the number of matches to 4 or 5 due to your high overlap during data acquisition.
This is likely to create some areas which do not have points, but will also significantly reduce noise.
In Cloud Compare I can recommend using the virtual broom tool. While it is really designed for removing traffic from LiDAR based scans it can be used to reduce noise both above and below the road surface.
If you would like any information, please reach out to me at jason@geodronesurvey.com
www.geodronesurvey.com
Thank you Jason. I am also a fan of the virtual broom tool in CloudCompare, it really speeds things up.
I will try your recommendations and give a later update for future readers. I had actually checked Multiscale on after comparing a prior project with a colleague. We found that with Multiscale unchecked, there were significant holes in the roadway, whereas his point cloud was filled in with Multiscale checked.
can anyone confirm or deny if these blots of points are caused by wet spots/ puddels of water on the pavement?
I am not sure but I think we can remove point cloud noise using “Edit Densified Point Cloud” on Pix4Dmapper and then we can run step 3.
Hi,
As already suggested by other members, you can deactivate Multi Scale.
- If the point cloud is noisy near the corners (buildings) and thin objects (towers and lines) we recommend deactivating the Multiscale option in the processing options of step 2. Point Cloud and Mesh > Point Cloud.
- If the point cloud is noisy and there are some points to be removed, it is possible to edit the point cloud and assign the points to the Disabled point group. These points will not be exported when the point cloud is exported from the rayCloud. They will also not be used for all the output generated after the editing of the point cloud (e.g. DSM, Orthomosaic, 3D Textured Mesh, etc).
How to edit the point cloud: How to edit the point cloud in the rayCloud.
How to export the point cloud from the rayCloud: How to export the Point Cloud.
I hope this information helps you to improve the results.
Cheers
I always have that issue on the smooth pavement , especially new pavement . I think it relates to all the pixels in the area being the same color. The algorithm is having a hard time to match same color pixels, because they are all the same color, no spots, stains rocks, cracks etc. The result is cloudy noisy point cloud, a similar one to the moving pixels, like water and tree foliage. I create break lines manually in Pix4d, then export them out to Cad, that way I do not use the noise points, just the correct ones. It is fairly easy to pick the right points, when manually creating the break lines.
Sure, new pavement is really challenging for (any) algorithm because of the reasons you mentioned (in particular the absence of a texture).
I’m glad you have found a workflow suitable for you.
Cheers