I would be grateful if you could help me understand an issue I’m encountering.
I have seven Ground Control Points with (RMS) errors ranging from 10 to 20 mm. I decided to use two of these as ground test points to assess the accuracy of my model.
However, the deviations at the ground test points are significantly higher compared to those at the GCPs. Here are some examples:
GCP 14: Total Error 0.001 m (X: 0.000 m; Y: 0.001 m; Z: -0.000 m)
GCP 15: Total Error 0.001 m (X: 0.001 m; Y: -0.000 m; Z: -0.000 m)
Ground Test 12: Total Error 0.039 m (X: -0.018 m; Y: -0.004 m; Z: -0.035 m)
Ground Test 13: Total Error 0.182 m (X: -0.098 m; Y: -0.089 m; Z: 0.125 m)
Workflow:
I processed aerial images captured with RTK positioning and some handheld images. After aligning the images, I disabled the positional data from the images (set to ‘prior to unknown’), imported the GCP file, and ran a new alignment.
Question:
Why are the ground test points showing significantly higher deviations compared to the GCPs? If the GCPs have an accuracy within 2 cm, why aren’t the ground test points reflecting similar accuracy?
Hello Rango,
such results could depends on more things.
How precisely were the GCPs measured outside and also over the images.
What is the distribution of the GCPs and which of them were used as ground control.
It also depends how your alignment was provided. What is the result of RTK alignment and what settings have you used?
Also, the resulting precision depends on your GSD, as it is 2-3 x of this value.
Each GCP was measured four times, and the average value was taken, achieving an accuracy within 2 centimeters. The building is round, with a distribution of five GCPs placed all around it and two on top. Two bottom and one top GCP were chosen as Ground Test
The Real-Time Kinematic (RTK) alignment was mostly fine. I had to apply some Control Points (CPs), but all of them are acceptable - though a few were marked in yellow (deviations greater than 2 pixels), and one was in red. After adjusting the red CP, I obtained a better result on Ground test number 12, which had a deviation of 0.039 meters, is now close to zero. However, Ground Test Point Number 13, still has a deviation of 0.18 meters. This point is visible mostly in handheld camera from close distance images and does not show deviations in pixels.
Settings are mostly on default. I set detector sensitivity to high and changed distortion parameter.
I noticed one more thing. In the report I get Geroreferenced as Yes, but metric as No.