Understanding AF Accuracy?

[Playdo]

I've done a lot of reading up on AF accuracy and I have a question. I know that there are a lot of technical people here, so put me straight if I'm going wrong.

Accuracy: This is the distance of achieved focus from perfect focus.
Precision: For multiple shots, this is the distance (or range) between the shots. Precision can not be measured from a single shot.

- Canon has cross-type sensors sensitive for f2.8. They state that these sensors are accurate to 1/3 x DOF. Therefore using an f2.8 lens, at any aperture, the lens will focus at the f2.8 aperture. So regardless of what aperture it is shot at it will always be as accurate as 1/3 x DOF at f2.8?

- Canon has cross-type sensors sensitive for f5.6. They state that these sensors are accurate to 1 x DOF. How does this relate to an f2.8 lens? Would it have the accuracy of 1 x DOF at f2.8 or would it have the accuracy of 1 x DOF at f5.6?

Please add in-depth links if you know of any.

[AJSJones]

I think you are referring to Canon's specifications for the AF - the good ones will place the plane of focus somewhere (anywhere) within 1/3 of the DoF while the others will be within 1 DoF. I've not seen any data from Canon or anyone else on how accurate they are in a more useful sense.  Since they are open loop systems, when the AF system tells the stepper motor to go to location XYZ it will get close but will have some variability.  The fineness of the stepper motor will ultimately limit this but not all steps will have the same effect on "proportion of DOF".  Therefore bell-curves of accuracy vs frequency of attempts will not necessarily be symmetrical and we don't know whether they change shape as the focal distance changes.  Even then, we don't know what the success frequency is at the point where the curves cross the spec limit (5%? 2%?).  It's a pretty complicated thing to put a specification on, so that's probably why they don't provide much info on the spec.  In any case, it is the DoF obtained by using the CoC they used for film way back when (~0.035mm) and given the small sizes of pixels nowadays it is likely that one could do a series of shots that were all technically "in spec" but yet, when viewed at 100% on screen some would look sharper than other or the plane of focus could be seen to vary - I think this is what happens when folks complain about the variability of the 7D's AF system!
I know I didn't answer your question, but it's not a simple question either.

[bradleygibson]

Wikipedia has a good article on Accuracy and Precision.  I think it addresses your question nicely.

Excerpts:

"...the accuracy of a measurement system is the degree of closeness of measurements of a quantity to its actual (true) value. The precision of a measurement system, also called reproducibility or repeatability, is the degree to which repeated measurements under unchanged conditions show the same results..."

"Although the two words can be synonymous in colloquial use, they are deliberately contrasted in the context of the scientific method."

"...A measurement system can be accurate but not precise, precise but not accurate, neither, or both..."

HTH,

[AJSJones]

From RDKirk's great summary:

As a result of this tolerance (within the depth of focus or within 1/3 of the depth of focus), the camera can place the actual plane of focus at random anywhere within the tolerance range, and not necessarily at the same place each time

the accuracy of 1 x DOF at f2.8

   We don't know what the accuracy actually is (it's usually measured in % )  Canon says they can hit the DoF target each time, but we don't know how close to the centre they achieve.

However, your question was related to the performance of a 2.8 lens using an AF point that's based on a 5.6 AF-type sensor.  The performance of the AF sensor is determined by the geometry of the light rays coming in at the edges of a lens aperture set at 5.6 (not by their intensity).  These sensors are closer together (have a shorter "baseline") than those using a 2.8 type sensor and so will have limited "accuracy" no matter how much light or what other rays are entering the lens.  SO a 5.6 sensor will only have the 1 DOF specification/tolerance, no matter how big the lens aperture is or the amount of light available. From the wiki on AF systems

Two optical prisms capture the light rays coming from the opposite sides of the lens and divert it to the AF sensor, creating a simple rangefinder with a base identical to the lens's diameter. The two images are then analysed for similar light intensity patterns (peaks and valleys) and the phase difference is calculated in order to find if the object is in front focus or back focus position. This instantly gives the exact direction of focusing and amount of focus ring's movement.

  That link shows the layout of the system from a patent drawing and illustrates just one AF point..

[wolfnowl]

3.14159265 is a very precise number.  However, if the correct answer is 5, it's not very accurate.  That's the best way I know to explain the difference.

Mike.

[AJSJones]

Wikipedia has a good article on Accuracy and Precision.  I think it addresses your question nicely.

3.14159265 is a very precise number.  However, if the correct answer is 5, it's not very accurate.  That's the best way I know to explain the difference.

Mike.

I don't think the OP was asking "What's the difference between accuracy and precision?"
I think the Q was if "I use a 2.8 lens and a 5.6 AF point will it be more accurate than using a 5.6 lens and a 5.6 AF point?"  My answer was NO, it's not the amount of light  but the geometry of the AF point that matters.

[bradleygibson]

Re-reading the OP's question, I think you are right, thanks.  Sorry about that, Playdo!

[Rob C]

Remember the joys of split-image screens? Wonderfully accurate - in my experience, at least. But that's partly the price to pay with these autofocus lenses - rob you of rapidly changed alternatives like that.

;-(

Rob C