Support Website Contact Support Blog

GCP challenges with MicaSense RedEdge Rig-captured images

Hi,

I am trying to process MicaSense RedEdge data.  Step one goes smoothly with Pix4D recognizing the RedEdge data, pulling EXIF metadata, and aligning the images fairly well. The issue is that I can’t get accurate alignment with my GCPs.

When I use rayCloud to manually identify my GCPs, I get some bands predicted well and others poorly. Generally band 2 is predicted well and tagged GCPs are close enough to the prediction to be used (yellow circle) but the other bands tagged GCPs are generally farther away and hence considered outliers (pink circle).  It doesn’t take into consideration any images from bands other than 2 in the realignment (because they’re outliers)… so I can manually identify one GCP in over 60 images and still have the GCP green + meters away from the clearly visible true GCP. 

My questions: 

  1. Why does it do this? - It knows these are rigs and that all 5 bands should be almost exactly the same capture right? I’ve looked into changing how it treats the rig but haven’t had any luck there yet.  How does it group images into a rig group? Is it by image time? GPS? name? If it is by name, what is the standard? I can’t use the RedEdge’s default naming because it repeats the file names fairly often (and starts a new folder) so I’ve had to add the folder name to the file names to use the repeating-name images in one project. The five bands from one capture still have matching file names except for the band number - but maybe I need that band number digit somewhere specific?  

  2. If Pix4D can’t align the 5 bands from each capture event automatically, presumably the the GCPs would force them to align.  But the GCPs in 4 of the 5 bands end up falling outside the “outlier cutoff distance” so they aren’t used in the alignment. Here is a typical example - I’ve manually tagged ~60 images that should have my GCP #6.  Only band 2 images have the GCP within the “yellow” allowable distance.  The rest are pink and aren’t used. I end up with a really accurate fix for band 2 - not that different that it was (see image) but it still has meters of offset for the other bands.  How do I get around this?  

Thanks!

Kerri

GCPs are pretty good to start with: 

Band 2 GCPs align well (a subset of GCP tag examples for band 2): 

Other Bands - not so much: 

Some examples from band 1:

Some examples from band 3:

Some examples from band 4: 

Some examples from band 5: 

GCP is adjusted slightly per only the yellow circles (almost all of which are band 2 images, with a few band 4 images as well).  GCPs in other images are still meters away from the predicted location. 

Any help would be greatly appreciated. 

 

Hi Johnatan,

 

The rig assignment in Pix4D is based on the unique capture id of the images. When one capture is shot then all the 5 images assigned to it will share the same identifier. This algorithm is available only for Rededge and Sequoia. 

Following during step 1 depending on the calibration method, the master band will be treated fully while for the others the rig relatives and subset of images will contribute to alignment. (see Processing-camera-rigs )

Giving the details above since the point cloud is generated based on the master band, green one, and few images from secondaries, then it is expected behavior that the software will not be able to identify the checks for the GCP’s in the other bands.

However, to assure that everything is correct could you share with me the quality report, p4d file, and log file so I could take a look?

 

Cheers,

Ina

 

 

 

Hi Ina,

Thanks for the insight into how Pix4D identifies sets of bands.  

I still don’t see how it can be treating them as a group and yet predicting the GCPs in different locations in each band. Here’s an example of one capture. You can see the images are all of the same geographical extent (as expected) and you can see GCP #7 is marked by hand in the same place on each.  You can also see that the blue circles of expected location of GCP #7 based on the imported lat/long are different for each band. How can this be if Pix4D knows these images all have the same spatial extent? It seems to me they are aligned differently if that given lat/long is in different places on each image. How can that be if they are tagged as the same capture and mainly the green band is used in the alignment?

Here are the 5 bands from that one capture event as an example:

Thanks,

Kerri

Hi Ina,

How do I send you the quality report, p4d file, and log file? It doesn’t seem like I can attach files other than photos here. 

Thanks

You could share it with Google Drive or WeTransfer. Make sure you have no rights on the folder. 

Thanks,

Ina