# accuracy

I was established 8 control stations around the building. I have been conducted building survey with total station for relative Geo-referencing.

Each control station having 4mm error and total station- Pentex  laser light accuracy 3mm

Now I got GCP accuaracy RMSE xy = 1.97 cm RMSE z = 0.7 cm.

For check points RMSE xy = 1.64cm RMSEz = 0.9 cm

According to article Horizontal accuracy should come 2* GSD = 6mm for my model. but I am having high error.

can you please suggest  regarding this information

2* GSD it’s due to the pixel size. If 1 pixel=1cm, your planimatric accuracy it’s 2 cm and height accuracy 3 cm.

Yes sir, My model having GSD= 0.3 cm

The accuracy link to GSD is a complicated function and not simply one number and not even the same across the entire point cloud.  I can get true independently validated relative accuracy below 2 GSD and very close to 1 GSD (horizontal and vertical) because I don’t use “industry standards” for flying and processing.  The setup for the camera, drone, and Pix4D are all factors for accuracy, which also means having a 3 mm GSD could mean your true accuracy could be 3 GSD or 5+ GSD.

The Quality Report is more a measure of how Pix4D is functioning internally in each Step and contains very little about absolute or relative accuracy of the end result.  You do want to keep the error listed in the Quality Report to a minimum so you know the software is performing at its best but this does not correlate into the accuracy of the project data.

Hello Chandrasekhar,

What are you trying to reconstruct? Could you please post some pictures of the rayCloud?
What drone and camera are you using to capture the dataset? Is this a nadir or oblique dataset?

Please note that the GSD displayed in the quality report is just an approximation. More specifically, an approximate GSD value is computed for each image. To obtain the average GSD of each image, some ATPs are randomly sampled and the distance(3D) / distance(image) is computed. Then, the median is used so that the GSD is less sensitive to extreme values.
The average GSD in the QR is weighted by the number of images at each GSD. The way that the GSD is calculated indicates that it is possible that the GCPs errors exceed the 2 times the GSD limit, especially when the GSD is that low as in your project.

Please post here the PDF quality report of the project to better understand your project and whether something has gone wrong.

Regards,