DJI Phantom 4 PRO and GCPs

PIX4Dmapper PIX4Dmapper Questions/Troubleshooting
1

Hello,

I have a problem after processing RGB images obtained from a DJI Phantom 4 PRO and improving the accuracy of the map (our target is under 1 meter accuracy) with GCPs.

First, I would like to know where in the Quality Report should I check the accuracy of the map with no GCPs? And just to be sure, can anyone confirm that in order to read the resulting accuracy I have to look at the first page of the Quality Report under Quality Check > Georeferencing > yes, 5 GCPs (5 3D), mean RMS error = 0.088 m? If it’s not that, then where should I look to find out if I obtained an accuracy of under 1 meter?

While the Quality Report for the images processed with no GCPs is looking fine, I have a problem with the one where we’ve integrated the GCPs:

Can anyone tell me why on the first page of the Quality Report the georeferencing line has an red triangle? I ask this because the GCPs we’re marked as good as I could.

In the “Calibration Details” and"2D Keypoint Matches" the green ellipses are of the charts, shouldn’t they be as small as possible?

I’ve attached the 3 quality reports that I generated from the start trying to improve the accuracy of the map: https://drive.google.com/open?id=1kPck1t1qmR1u5Tk3Y8KKFEekhM3HqfAh 

Can anyone check them out and give me feedback?

Thank you.

2

Hi Alex,

I had a close look at the quality reports and notice that there were some issues with the camera optimization in the project. This can be seen in the Camera Optimization and Internal Camera Parameters sections of the quality report:

The camera has a rolling shutter which can be an issue for the reconstruction depending on the flight conditions and the area that is mapped. To mitigate the effect, we suggest applying a specific algorithm especially for flights with a regular pattern (lines). 

If you see any symptoms in your project that are commonly due to excessive amounts of linear rolling shutter distortion in your images, e.g. your results bending or an extreme difference between your camera’s initial and computed internal camera  parameters, etc., we recommend the following:

 

To ensure that none of your questions fall through the cracks, please allow me to address each one individually. 

1. Where in the Quality Report should I check the accuracy of the map with no GCPs?

Typically, without the GCPs, only the relative accuracy can be evaluated since there are no absolute values, e.g. GCPs, introduced in the project to check the reconstruction.

To get an idea if the reconstruction is suitable or not I would recommend checking if the:

  • Camera Optimization. The percentage of the difference between the initial and optimized focal length should be less than 5%.
  • Absolute Geolocation Variance. The values displayed can be used to check if the estimated accuracy of the image geolocation (e.g. 5/10 m) is realistic or not. In case that you notice high RMS Error, then maybe the image geolocation is not as precise as expected. 

2. To read the resulting accuracy I have to look at the first page of the Quality Report under Quality Check > Georeferencing > yes, 5 GCPs (5 3D), mean RMS error = 0.088 m? 

Yes, when using GCPs, the final accuracy is visible on the first page (Georeferencing). A more detailed report can be found in the Ground Control Points section.

3. Why on the first page of the Quality Report the georeferencing line has a red triangle?

This indicates that the GCP error is bigger than 2 times the Ground Sampling Distance. This may indicate a severe issue with the dataset or more likely an error when marking or specifying the GCPs. 

In your specific case, the GSD is 1.55cm and the expected error would be 2 x 1.55cm = 3.1cm. Since the actual RMS value is 8.8cm, the warning is displayed. 

4. In the “Calibration Details” and"2D Keypoint Matches" the green ellipses are of the charts, shouldn’t they be as small as possible?,

Keep in mind that the ellipses in the quality report are magnified, 50x times in Figure 3 and 1000x times in Figure 5.

One thing to keep in mind is that the uncertainty ellipses are highly influenced by the accuracy of the image geolocation. If you, for an example, have a DJI drone with a GPS receiver that has the expected accuracy of coordinates e.g. 5m for the horizontal and 10m for the vertical than the ellipses will be bigger   than from the one that has the image geolocation acquired with RTK receiver. Since you are using a DJI drone we think that the size of the ellipses is as expected.

Best,