On 3-10-2018 16:12, Bald Eagle wrote:
> I think if you look at:
> 
> http://www.povray.org/documentation/view/3.7.1/246/
> 
> you'll see several things.
> 
> 1. The +/- 0.5 range for x and y

Yes, of course. I new it was fundamental (but always push at the 
fundamentals ;-) ). Had to be reminded.

> 
> 2. camera angle is determined by changing the length of the direction vector.

Indeed. It is the way in which Gilles Tran's visibility test macro is 
made operational in a given scene.

> 
> 3. I think this relates to the error you see, and it's decrease with smaller
> camera angles.

probably yes.

> 
> Here's my crappy reasoning-by-analogy in the absence of the code in front of me,
> and on short notice:
> 
> When you swivel a line through pivot point, a small change at the end results in
> a large change in the other end when the pivot point is close ( small direction
> vector / wide angle)
> 
> When you pull the camera way back, you have to make very large changes in the
> position to have the same effect - or put the other way - any changes at the
> other end are very small (your errors) for the same amount of change.
> 
> Think about that in terms of any trace() results and calculations based on those
> - they're like angles, and like levers they can magnify (by integers or
> fractions) the results.

That seems reasonable indeed. I agree.

> 
> 
> 
> I don't think you're/we're going to get very far fiddling with code or working
> backwards from the existing macro.
> 
> Write some code to model a default camera scene, and then play with point
> locations and their projection onto the z=1 plane.
> 
> Then you can work from there.
> 
> What you do (and understand) in a forward, stepwise manner, you can follow
> stepwise, backwards to see what calculations you need to to go from screen
> coordinates to 3D vector coordinates.
> 
> 
> 
> 
> 


-- 
Thomas

Post a reply to this message