Hello Pix4D:
I am 3D mapping my car for a university project however I am having difficulty with getting no holes/gaps in the model. Specifically, I notice problems near the hood and the top part of the vehicle. There were reflections which may have contributed to the missing imagery. I have tried adding more photos (256 photos which is more than enough). When looking at the point cloud, it shows no information in those affected areas of the car.
I have used an ND 8 filter which helped but it didn’t completely resolve the issue. I have also tried adding MTPs to see if this would help, but it did not. I did not use any GCPs for this project as well, as the vehicle is small.
Please note this project was done using DJI Point of Interest intelligent flight mode then the photos were imported to Pix4D. This allowed me to fly lower than 32 feet and closer to the vehicle. I have additionally tried rendering the model in higher resolutions to see if it helped but I didn’t notice any improvement with the missing imagery. The drone used was a DJI Phantom 4 Pro V2.
Thank you for your assistance in advance. Lastly, there were I think 5 uncalibrated photos but I don’t think this would have impacted my rendering.
Specs:
Intel i7 10700KF at 4.8 GHZ
32GB 3200 MHZ RAM
RTX 2080 Super Overclocked
256GB NVME SSD
Pix4D Quality Report:
Pix4D Cobalt 200_report.pdf (592.4 KB)
Here you can see the complete front of the hood is missing in the Point Cloud. Although there was reflections, there was not a significant amount in my opinion.
Can I manually edit in the missing information in the Point Cloud? Or is there something else I can do to fix this?
Hi Beckerd
You could use DSLR or a mobile phone to take photos of the area where the information is missing in the point cloud and then combine the datasets.
DSLR would be ideal for this kind of job. You could use tripod and a wide-angle lens with a small aperture. You could even use circular polarizer to minimize reflections. Just my two cents…
If you can share the dataset I could take a look at it.
1 Like
Thank you for your reply. I would really like to use a DSLR to capture the missing detail; however, the project limits us to only using a UAS. (Its just our assignment policy)
How would I exactly share the dataset? You mean upload all 268 photos with the p4D file? Or would I just do a file upload to Pix4D Cloud and give you a link?
Upload just the photos to cloud (Onedrive/GDrive/Dropbox) and send me the link.
https://1drv.ms/u/s!AnN5qAdg77WxnQMBbhd319O_gQ81?e=K4MZ5H
Thank you, this OneDrive link should work. Please note the file size is rather large (a couple GB’s). You will notice I got some close-up photographs too to help with shadows.
Hi Beckerd
Here My results:
with these settings:
You should try those settings in your projects specially since you’ve already added MTP’s so the point cloud should be much cleaner.
EDIT: Btw in this kind of project you can also use the drone as a gimbal. Just handheld the drone and take those near ground pictures that way. You could also use some kind of masking tape or spray in those reflective areas like the hood and the sides of the car.
Thank you so much for the reply. I will try those settings right away and let you know how it works.
Unfortunately the results are not that good. There is just way to much noise/artifacts around the model. The 3 overlapping images resulted in a cleaner point cloud.
It really does make me wonder, would placing a sticky note on the hood and the top of the vehicle help? I noticed the point cloud placed a lot of automatic keypoints near the small antenna on the roof of the vehicle. I think because the hood is a consistent color that Pix4D struggles with placing keypoints. This can be seen in the point cloud where there are practically none on the hood.
I also tried adding MTPs to try to clean up the point cloud with 4 overlapping images but it didn’t help.
I think a sticky note in the center on the hood would be useful because the program would recognize it if it was a different color.
Also, are you aware of a method to actually edit the point cloud to fill in the missing imagery?
Update: I have used a planar constraint surface to the top of the vehicle. I reprocessed the 3D mesh and that fixed the issue.
Hi
If you want more points in the area where there is only few or no none at all it will almost certainly add more noise elsewhere in the model. But it’s pretty easy to remove the noise manually or use some programs that have automatic noise removal tools. CloudCompare has pretty good noise filters and best of all it’s open source.
There is a method to add the missing information to a point cloud. You’ll need to take another set of pictures of the same subject or a part of the subject. Process the projects separately and after that you can merge the projects together.
