Incorrect Geolocation on Orthomosaic

PIX4Dmapper PIX4Dmapper Questions/Troubleshooting
1

I have geolocated a project using 10 GCP’s acquired from a total station ( I work in engineering and surveying so this is not a new process). 

On this project I have found the orthomosaic to be about 0.75’ offset from the GCP’s. The point cloud is dead on with the gelocation report (<0.02’ RMS) See screen captures below. I have verified the world file is creating the translation for the image.

As I regularly work with coordinate systems and ground control points I am perplexed. Has anyone else experienced this?

orthomosiac compared to GCP in Autocad

Point cloud compared to GCP in Autocad

 

2

Dustin, unless the camera is dead center over an object, that object has to be heighted correctly in a surface model to have an accurate location. The farther off from vertical (nadir) the worse the effect. (You can really see this in tall objects like poles and towers with their apparent leaning over. The top of the object is displaced.)

The fact that the point cloud fits like a glove is testament to what is probably a pretty good field survey, and proof too that all this crazy stuff can work really well.

But it looks like your control points are on the tops of elevated manholes, themselves on an elevated vault, on the side of a slope?

Orthophotos are ‘draped’ over a surface model created from the point cloud after being cleaned of objects to approximate bare earth. If the statement above characterizing the field conditions is accurate, the surface model would have been smoothed of irregularities and objects.

You might be able to go into the orthomosaic seam editor and pick a more nadir photo to use. Check out the planar and ortho toggle.

You might draw (at their correct elevation) a little surface around objects so they make it through the process at their correct elevation.

Steve

3

Totally agree with what Steven has replied with, I think that being able to get so close with NADIR and Pix4D is fantastic and shows to thew world what we have to offer and that using RPAS,NADIR and Pix4D who knows what the future is going to be like in only a few years.

 

Keep up the great work Pix4D and we look forward working with you in the future.

4

I understand what you are saying however it is a specific offset (about 8 inches) that is consistent across all GCP’s regardless of their elevation difference. Many of the images (a total of 300, 138 contained the GCP’s) used to mark the GCP’s were vertical or nearly vertical. This seems like it would be the difference between Us Survey Foot/International Foot but accounting for this change does not correct the issue.

5

Dustin, 

This is interesting and I’d love to hear what you find out.

I agree with all the comments and suggestions above, especially drawing a surface on each flat top elevated MH.  

Does the error appear to be systematic or random?

Is the Ortho off 0.75’ in the same direction for all GCP’s ?

Can you verify the exact same projection is being used in Pix4D and your AutoCAD project?

Did you check/adjust the Marks for the GCP’s in all 300 images ?

6

Dustin, just to clarify… It doesn’t matter how many images were near vertical when exposed. It just matter which individual photo was used when creating the mosaic. All vertical images will exhibit this effect across the lens to its extremities.

Also, looking at the pixel size in the screencaps, it could be pushing the orthophoto to expect the kind of accuracy you’re looking for on individual features. You might be better off dragging the point cloud into your cad and working with that.

But all that aside, what’s your pixel size? It sounds like you need the worldfile for import? If its a uniform shift, check that your software uses the center of the upper left pixel and not it’s corner. If all else fails and it is a uniform horizontal shift, making a tweak to the coordinates in the tfw would be fair.

Local or projected system? Grid or ground?

Steve

7

Dustin, 

 

Are you generating an ortho at 1x GSD, or are you downsampling it? 

 

I’ve had similiar results when generating an ortho at 5x GSD, but the problem is not present at 1x GSD. 

8

I just noticed this as well… 1xGSD vs 5xGSD results in observable differences on the ortho (differences in feature locations, not just resolution). This does not seem to be a result of vertical relief correction. Is there any resolution to this?