Support Website Contact Support Blog

Rolling shutter correction model results in erroneous DSM

We tried to generate the orthomosaics and DSMs with Pix4Dmapper Pro 2.1 using UVA images, which was processed with and without the rolling shutter model. We found that the rolling shutter model significantly improves the accuracy of the check points in x and y direction, but it deteriorates the accuracy in z direction (table 1 and table 2).

Fig. 1 is the generated DSMs of the two processing methods. It seems the DSM without rolling shutter correction is visually smooth and consistant while errornous horizontal strips are clearly present in DSM with rolling shutter correction.

Fig. 2 is the enlarged subset of the red box in fig.1. The abnormal horizontal strips can be clearly identified (b). The difference of DSM subsets © indicated elevations are greatly changed after rolling shutter correction.

Our experiment was performed on a newly plowed agricultural field, covering an area of 4.4336 ha. The flight was conducted with DJi PHANTOM 4 UAV equipped with 12 megapixel consumer grade cameras. The mission was planed with Map Polit app, with an altitude of 30 m above ground, achieving a GSD of 1.33 cm. A total of 368 images were acquired using an overlap of 80%. 25 ground control points (GCPs) were measured with a surveyor grade dual frequency RTK GNSS receiver. 12 of the 25 GCPs were used in the processing to georeference the reconstruction, while the remaining 13 were used as check points. (Fig 3).

Any comments on this will be appreciate!!

Hello,

Please send us your dataset (images and .p4d file) through Pix4D Cloud or through another file transfer platform and send it to: support@pix4d.com?

 

Best Regards,

 

 

Thanks for your response!

we will send you the data address once we finish uploading the dataset.

Dear Support:

         We sent our dataset address. Please find it in the email.

Cheers~

 

 

Dear Hui Li,

Thank you for bringing this to our attention! 

Indeed , there was a problem in the Dense Processing in this version. By refactoring the Dense Processing with rolling shutter, these effects on the point cloud and the DSM have been removed.

That is now fixed in the new version 2.1.57 that is already online :-)! 

Best Regards, 

 

Thanks for the efforts u made to improve the software. The erroneous strips in DSMs were removed in the new version. however, new problems seem to arise from the new version.

We conducted two projects with and without the new rolling shutter model correction. it was found that for the check points the model can significantly reduce the RMS error in x and y direction, however, the model seems to increase the RMS error in the z direction as well.

 

 

  

 

 

 

 

Dear Hui Li, 

Thank you for the feedback :slight_smile: it help us to continuously improve.

Please send us the dataset (images and .p4d files) for the two projects with and without Linear Rolling Shutter model to: support@pix4d.com.

Best Regards,

Dear support:

the dataset have been sent. plz check the email.

 

Dear Hui Li,

We have not received an email from the address you are using to post on the forum.
We also checked our server and did not find any uploaded projects associated to this account.

Could you please upload the files here: https://www.dropbox.com/request/GgJPfZVsS4cWjF30AgiR
Only the Support Team will have access to it.

Thank you for your help.

Regards,