Pointcloud data appears to be warped

PIX4Dmapper
1

After processing a 140 acre mall and parking lot, I compared my results to a 2019 federal LiDAR dataset. When running cross sections, shallow Xs are formed where my data crosses above then below the LiDAR. I could live with a uniform offset, but above and below does not make sense. I have 14 control points, 4 GCPs to process and 11 checks. RMS error is right at the 2x GSD, and the check points are all labeled as accurate, but the errors in check points are in the 1 to 3 foot range, much higher than I’m used to seeing. GCP errors in XY are between 1 and 6 inches, Z is very tight.

P4 Pro with Map Pilot Pro, I know about the camera selection issue and I follow the workaround.

2

Hi there

Could you share the quality report?

3

Site is close and small enough that I’ve been able to fly it multiple times, all with the same results. Quality reports for the flights.

23-013 BoulevardMall 20230402_report.pdf (1.9 MB)
23-013 Boulevard Mall 20230409_report.pdf (1.9 MB)
23-013 Boulevard Mall 20230412_report.pdf (1.6 MB)

4

There’s something funky going on with your checkpoints and gcps. How were they obtained?
I’m not an expert on imperial systems but I would say that 1,54 feet in z-error is quite a lot…

image
image1331×1083 158 KB

5

Hello @Jaakko_Laihola, I believe the issue might be due to the number and the distribution of GCPs. First of all, make sure that the GCPs number and the distribution is suitable to properly georeference the project, and then add an extra check point to verify the accuracy. The GCPs distribution of the last quality report looks better. I would have GCPs as shown belo for the project.

image
image698×682 83.2 KB

I can also see that the GCPs marking for the below point are not good. I recommend to remark them. While marking GCPs, make sure that you are only marking the one that are clearly visible, zoom to the maximum extend and mark them, mark 8-12 good images.

image
image753×641 84.7 KB

6

Unfortunately, there is a shopping mall in the center of the site, so GCPs can only be established on the periphery.

The GCPs show up nicely on the images, so picking the PK nail really isn’t an issue.

I’m wondering find the large structure in the middle of the pointcloud is creating an issue.

I’m also seeing automatic tie points in the distance, with nearly every photo hitting that point. Not sure if sun glare was the cause or not, just seems odd these 3 points are there. I’ve attached a screenshot showing these points.

Looking at the pointcloud, the photos on the edge appear warped, but maybe that’s the poor solution.

distant tie points
distant tie points1904×1021 182 KB

7

If you can share the whole dataset i could take a crack at it.