Hello,
I have a weird problem with Pix4DMapper. I have a series of about 200 oblique aerial photos of an airport area. I want to create a point cloud from that (eventually, a mesh).
However, when I create the point cloud, I have lots of artifacts especially on plane surfaces. You can see in this image:
The white surface (some concrete) is not flat, but instead has many points along the red arrow. It seems that the plane is too homogeneous to be clearly distinct. Then in the epipolar planes the points are assumed to be closer (and further away) since they do not match correctly.
When you look at the point cloud from the angle of the cameras, it looks very nice. But when you look at it in from the side, it looks like this.
I already tried reducing the number of images up to every tenth and increasing the number of required matches to six, but to no avail.
Does anyone know what I could do with this data to get a clean point cloud? This mainly affects homogeneous areas like the concrete or untiled building roofs. Note that this problem occurs not only with this one series, but I have seven of those. All shot with the same camera, but at different angles and heights, still the same.
Best regards,
Julian Fagir
Hi Julian,
Welcome to the Pix4D’s community!
And thanks for sharing the screenshot of your dense point cloud.
As you mention, you captured images of an area that includes “homogeneous areas like the concrete or untiled building roofs”.
How you can try to improve current projects
1. I suggest you change some processing options for step 1 then process step 1 again:
-
1. Initial processing > General > set the Keypoints Image Scale to 1/2.
-
1. Initial processing > Matching > check the box for Use Geometrically Verified Matching
-
1. Initial processing > Calibration > set the Calibration method to Alternative if the dataset follows the characteristics mentioned in this article section. It is hard to tell in your screenshot if there are objects in the scene but you do mention building roofs so only change this option if the scene is relatively flat. If there are buildings, keep using the Standard calibration method.
2. Check the quality report and visual aspect of the automatic tie point in the rayCloud.
-
If there are no errors anymore in the automatic tie points, try processing step 2 with the default options. If the result is not great, you can try 2. Point Cloud and Mesh > Point Cloud > setting the Image Scale to 1/4 or 1/8 then process step 2.
-
If there are still errors in the automatic tie points, please attach the quality report and screenshots of the automatic tie points in the rayCloud.
For future projects
In order to get the best representation possible with image content that is not ideal for photogrammetry applications, such as homogenous, complex, or reflective image content, we recommend capturing a ground sampling distance (GSD) of no less than 10 cm and overlap of at least 85% frontally and 70% laterally. If you need to capture a smaller GSD than 10 cm, we recommend you increase the front and side overlap.
I hope this helps and let me know how it goes!
Dear Rhea,
thank you for the extensive reply. I tried the first two points you mentioned, but it did not help. The third point does not apply, since it is an oblique flight track.
I now realize that the problem already happens at finding the tie points, not the densification step.
Attached there is one screenshot (I cannot put more) and the report PDF.
report.pdf (3.2 MB)
Best regards,
Julian
Hi Julian,
Thanks for providing additional screenshots.
I now realize that the problem already happens at finding the tie points, not the densification step.
This is what I suspected and why I mentioned some changes that you can make followed by processing step 1 again. Sorry if I was not clear!
The points 1 and 2 I wrote above should be followed in that sequence. If the results are not good after processing step 1, do not process step 2 before solving the issue.
The main issue in your project seems to be a lack of sufficient overlap to compensate for image content that includes “homogeneous areas like the concrete or untiled building roofs”. The options I mentioned can help improve the results but if they do not work, you need to capture more images to increase overlap in your project.
See Selecting the Image Acquisition Plan Type.
As mentioned above, I am happy to look at the quality report of the project. It’s unlikely but maybe I can think of something else that could help.
The third point does not apply, since it is an oblique flight track.
You mentioned that it is an airport area. What type of flight mission did you execute? A double grid?
What is/are the output/2 that you are most interested in generating with Pix4Dmapper?
