Quote:
Originally Posted by peter_4059
Ray, I'm happy the curve resembles a V however I'm not too convinced the slope of the LH vs RH side of the V is correct. If I fit regression lines to the data either side of focus I'm getting quite different slopes if I include all the points. If I restrict the data to points close to focus but not in the flat part of the curve I get much closer gradients. I'm not sure what SGP is doing behind the scenes with the data but if it is simply the intersection of the two lines then the different slopes impacts the focus position. I've tried to demonstrate this in the attached image which shows two sets of regression lines - one using all the data points and the other selected ones.
|
Ah OK, understand. Similar questions led me to use small steps and rely more on the fallback statistical estimator rather than the intersecting lines fitted either side of focus. The other issue was that out-of-focus stars can drop below the detection threshold and one way around that in sparse starfields was make sure they did not go too far from focus. Having said all that though, I tried upping the step size as Slawomir suggested and it is working fine with tonight's target region.