Hi friends. How are you?
At the company I work for, we have a process where an AI analyzes images of oil tankers and makes markings in areas with oxidation.
We are using Reality Capture software to generate the mesh.
We would like to create a textured model with these images. The problem is that not all images that pass through the AI have the same demarcation, that is, they have no repeatability.
I would like to keep the demarcation done by the AI, we change the colors of the demarcation to purple, however since in some cases we don't have the same demarcation (some areas are not demarcated in similar / close pictures). When I generate the texture, in some cases I lose the purple information.
It would be interesting if the demarcation was still present.
I posted a photo of the model with the new color and also inserted three nearby images demarcated by the AI exemplifying that there is no constancy in the demarcation. I marked only one point where the problem occurs, but there are many.
We are open to testing other software for success.
Steps to reproduce the problem:
- Alignment
- Mesh generation
- Texturing
Thanks.
Photogrametry help
-
- I have made <0 posts
- Posts: 1
- Joined: Tue Nov 08, 2022 5:09 pm
- 1
- Full Name: Carlos Queiroz
- Company Details: Orbitall 3D
- Company Position Title: owner
- Country: Brasil
- Linkedin Profile: Yes
Photogrametry help
You do not have the required permissions to view the files attached to this post.
-
- V.I.P Member
- Posts: 544
- Joined: Fri Feb 24, 2017 10:47 am
- 7
- Full Name: Martin Graner
- Company Details: PointCab GmbH
- Company Position Title: Research and Development
- Country: Germany
- Linkedin Profile: No
- Has thanked: 160 times
- Been thanked: 175 times
- Contact:
Re: Photogrametry help
Dear Carlos,
I think the issue is very intriguing, but I think the subforum you have choosen is not ideal.
Laserscanning Europe is a company providing service, sell scanning accessories and do trainings.
Perhaps you have luck with anybody having ideas though
Cheers
Martin
I think the issue is very intriguing, but I think the subforum you have choosen is not ideal.
Laserscanning Europe is a company providing service, sell scanning accessories and do trainings.
Perhaps you have luck with anybody having ideas though
Cheers
Martin
- smacl
- Global Moderator
- Posts: 1409
- Joined: Tue Jan 25, 2011 5:12 pm
- 13
- Full Name: Shane MacLaughlin
- Company Details: Atlas Computers Ltd
- Company Position Title: Managing Director
- Country: Ireland
- Linkedin Profile: Yes
- Location: Ireland
- Has thanked: 627 times
- Been thanked: 657 times
- Contact:
Re: Photogrametry help
Well spotted Martin, moved to general forum.VXGrid wrote: ↑Fri Nov 18, 2022 2:45 pm Dear Carlos,
I think the issue is very intriguing, but I think the subforum you have choosen is not ideal.
Laserscanning Europe is a company providing service, sell scanning accessories and do trainings.
Perhaps you have luck with anybody having ideas though
Cheers
Martin
-
- I have made 30-40 posts
- Posts: 30
- Joined: Wed Dec 12, 2018 5:56 pm
- 5
- Full Name: Mike Dailey
- Company Details: Survey work
- Company Position Title: Office Tech
- Country: United States
- Been thanked: 6 times
Re: Photogrametry help
Very cool stuff. This is outside of my area of expertise so take what I say with a grain or two of salt.
I'm assuming you are running the drone pics through your AI tool and then rendering the modified images through the cloud/mesh generation suite. This cloud/mesh generation is choosing what picture to use for colorizing a given point and so you would have two options.
1) Determine how it decides this and try to override that choice.
2) Try to get your AI to make more uniform decisions on its oxidation determinations.
I think #2 is your best bet. From looking at the sample you shared it would appear to me that your algorithm has a hard time with lighter areas. As you fly around and take pictures from different angles, those lighter colored surfaces get more or less washed out by the sun. You highlight one example in the picture and you can see a similar example on the covered walkway opposite the crane. Maybe try adjusting the contrast and or saturation levels of the project as a whole, individual photos or maybe even specific areas in the photos if possible. Doing this, I believe, would allow your AI tool to generate a more consistent identification of the oxidized areas. I would assume that once this is done, the mesh generation tool would be more likely to hold these highlighted areas into the mesh.
I'm assuming you are running the drone pics through your AI tool and then rendering the modified images through the cloud/mesh generation suite. This cloud/mesh generation is choosing what picture to use for colorizing a given point and so you would have two options.
1) Determine how it decides this and try to override that choice.
2) Try to get your AI to make more uniform decisions on its oxidation determinations.
I think #2 is your best bet. From looking at the sample you shared it would appear to me that your algorithm has a hard time with lighter areas. As you fly around and take pictures from different angles, those lighter colored surfaces get more or less washed out by the sun. You highlight one example in the picture and you can see a similar example on the covered walkway opposite the crane. Maybe try adjusting the contrast and or saturation levels of the project as a whole, individual photos or maybe even specific areas in the photos if possible. Doing this, I believe, would allow your AI tool to generate a more consistent identification of the oxidized areas. I would assume that once this is done, the mesh generation tool would be more likely to hold these highlighted areas into the mesh.
- stevenramsey
- Global Moderator
- Posts: 1937
- Joined: Sun Aug 12, 2007 9:22 pm
- 16
- Full Name: Steven Ramsey
- Company Details: 4DMax
- Company Position Title: Technical Specialist Scanning
- Country: UK
- Skype Name: steven.ramsey
- Linkedin Profile: Yes
- Location: London
- Has thanked: 30 times
- Been thanked: 72 times
- Contact:
Re: Photogrametry help
first thought would be to weight the images that don't have the correct color to something like 0.1 so that the corect colors have a higher weighting
-
- V.I.P Member
- Posts: 1237
- Joined: Mon Jan 04, 2010 7:51 pm
- 14
- Full Name: Jed Frechette
- Company Details: Lidar Guys
- Company Position Title: CEO and Lidar Supervisor
- Country: USA
- Linkedin Profile: Yes
- Location: Albuquerque, NM
- Has thanked: 62 times
- Been thanked: 220 times
- Contact:
Re: Photogrametry help
I would treat the color textures and oxidatized regions as seperate texture maps and do your compositing after the photogrammetry rather than before:
- Generate a base color texture map using the original source photos and standard photogrammetry processes.
- Run the source photos through your oxidation detection algorithm and generate an oxidation mask image for each source photo where oxidized pixels have a value of 1 and all other pixels have a value of 0.
- Use the camera pose and distortion information from step 1 to generate an oxidation texture map for your model using the oxidation mask images. Depending on what types of errors you can tolerate most easily there are a few different ways you could combine the oxidation masks.
- If Type I errors (false positives) are better than false negatives set the pixel value of the oxidation texture map to the maximum pixel value of all images that project on to that point. That will give you a texture map with a value of 1 anywhere oxidation was detected on at least one source image.
- Alternatively, if Type II errors are better use a minimum function to only mark pixels where oxidation was detected on all source images that project to that point.
- An even better option might be to add up all the pixel values of the source images that project on to a point and use that as a reliability metric based on the assumption that the more overlapping source images oxidation is detected on the more likely it is to be real.
- Finally, composite the oxidation texture map over the base color map for visualization
Jed