In December of 2008, after a series of back of the envelope type discussions on how to calibrate a Sentinel camera, Ken formalized the discussions in a pdf. To see Ken’s method Click here for Ken’s original paper on how we can calibrate the Sentinel camera.
As a footnote I (Jeff) did a work through with my camera system. After the spreadsheet was made and a small Python program run Ken took my results and plotted it to see how his model stood up. Here are those results.
I could not resist having a quick look. Your data is lovely. I simply plotted
R = sqrt((x-x0)2+(y-y0)2)
against zenith angle.
In the first plot I assumed the camera zenith was (371,240), and in the second I optimized it, and got (370,231). This of course would be the camera zenith not the centre of the frame.
HOWEVER: Look at the nice clean plot and good correlation. I think you can use the camera zenith method and ignore trying to find the centre of the frame; they are obviously very close because the plot, including the very slight nonlinearity due to the fisheye effect, integrates so cleanly.
The very slight fisheye effect can be approximated more than adequately by a 2nd-order polynomial, as Martin Connors said it would.
I would say, for your system, you should be able to simply calculate R and use the equation to get the zenith angle, and then do a rotation to find the azimuth error…. job done.