I don't how many advantange can be obtained if the algorithm can provide the direction.
If you adopt mount climbing method then
1. First time you get the focus value and keep the motor at the original position.
2. Second time you get the focus value then compare the value to preview one. And you
can know the position.
It only need two times.
My poor method is as the following ..
1. Calculate the focus depth by the lens specification and the motor specification.
It usually needs at most 9 points. (9 focus depth can cover for most cases).
If you need near shooting, you may need another table for focus depth.
2. Use mount climbing method to seach the 9 points and find the best focus point.
This method can find the focus position at average 4 times. Even you use the global
search method, you can find the focus position at 10 steps. If this method can meet yourrequirement??
+5 RD币
I add some thought about the algorithm on calculating the focus value.
My concern on the algorithm are:
1. If the algorithm is simply ? I don't want a lot of comuptation power.
2. If the algorithm can generate the peak from 0~infinite. If you want to evaluate the
algorithm, you need to check a lot of scene (one object, two object ,and ...etc). To
check if the algorithm can generate the peak from 0~infinite. If the algorithm can't
generate the obviously peak value, then this algorithm is not good enough. There
are many algorithm is focus on this field.
3. How the algorithm handle the focus depth. If there are two objects, one is in the front,
how the algorithm gernerate this two peak ?.
4. How the algorithm handle the noise ? The noise is high frequence signal. It will affect
the algorithm if the algorithm don't hand the noise very well.
Above are my poor idea for your reference.
[此贴子已经被作者于2007-2-6 15:06:42编辑过] |