RMSe in Agisoft StereoScan

[sarhat]

Hello All

I have searched through topics but couldn't find any help regarding the RMSe in Agisoft StereoScan.
You might all aware of that Agisoft can produce a summary output report. In that report you can find rmse error as can be seen in the sample shown below
Matching images...
Detecting points...
photo 1: 23709 points
photo 2: 28310 points
Matching points...
Calibrating stereo pair...
radial distortion calibration: enabled
using 40 of 52 matches
focal length: 3698 pix (unknown, trying 50mm)
38 points reconstructed, rms error 0.294266.
   
My question here, is there any one knows how agisoft generate that rmse error.
As far as we know the rmse error is a good indication of the geometry  but in my tests show that no relevant relation can be drawn between the that rmse error and the geometry and also proved that rmse error is decreased with decreasing the intersection angle (see attached image) but have shown also decreasing the accuracy of geometry.


Regards