AF Sensor Behaviour - More complex than it might appear

[CaptivatedByBeauty]

After a shot in which the focus was not where I expected, and the AF point was spot on where I wanted it to be, I started looking for the cause. I couldn't find anything.

Then I started searching for AF sensor accuracy/precision/reliability, and found a number of very informative posts. Those who tested their AF sensors either used depth of field charts, or stickers on walls. I tried a few things, and found that I could repeat the focus problem.

But, it seemed to me that the behaviour of the AF sensors must be quite complex. The posts I'd read mainly concerned the location of the AF sensors compared to where they appeared to be in the viewfinder. But, each sensor has a complex design, and that makes the way it detects patterns complex, but it also means that near objects could attract more 'attention' than the object under the centre of the sensor.

I've just been taking some shots of a table full of Lego. Because there's lots of depth and pattern, it turned out to be a very good target. I chose a day when it's not too bright (cloudy), indoors, but with lots of window light.

I used a Canon 60D with 70-200mm f/2.8 II, at f/2.8 and ISO 400, with AI servo, and using the AF-on button. On a tripod, using the shutter button. I selected on AF point manually.

I started pointing at spots and pressing the AF-on button, releasing and then pressing the shutter, but that doesn't record the AF point used for DPP. I found it better to press the AF-on button and very slowly move the camera around and look at the behaviour of the focus. After some experimenting, I added a 'bridge', so I had a hole with things on all sides.

What I found is that each sensor square doesn't match where the actual sensor is, but more importantly, high contrast objects near to the sensor caused that object to be focussed on and not the object I was trying to focus on.

Having read about this recently, I suggest this applies to all cameras. I suggest that it's quite instructive for people to try this themselves, to get a feel for how the AF sensors behave. That way, when trying to focus, you will have a better chance of anticipating problems, and also you may slightly offset your AF point knowing it will 'grab' the object you want.

Anyway, here are some pictures. This is just a small subset of what I took.
The 'failures' are in the main more interesting than the 'successes'.

These are using the top sensor.
This is what you'd expect:


Moving the sensor down just a little, and instead of focusing on the pot behind, it focused on the near-by closer objects:

And again on the other side:


Again, what you'd expect:


more...

[CaptivatedByBeauty]

Not focusing on what's at the sensor, but what's below it:


This one is maybe most surprising, instead of focusing on the object at the sensor, it's selected the lower object:


The centre sensor is obviously bigger than the square:


The top-right sensor was surprisingly good:


more...

[CaptivatedByBeauty]

and

and

and


It got this right:


more....

[CaptivatedByBeauty]

But got this wrong:


As I hope you can see, it's not just the sensor location compared to the squares in the viewfinder, it's also how big the sensor area is, and how it reacts to complex patterns of the scene. I think I've only scratched the surface!

These are some useful links for more background:

Previous tests:
7D - Problem found when mapping AF points - Canon Digital Photography Forums

Great diagrams of AF sensor design:
7D af point size and alignment: Canon EOS 7D / 60D - 10D Forum: Digital Photography Review

5D MKII outer focus points
5D MKII outer focus points [Page 3]: Canon EOS-1D / 1Ds / 5D Forum: Digital Photography Review

Autofocus - Wikipedia, the free encyclopedia

Nikon D70 Focus test chart
Nikon D70 Focus Chart

[ravncat]

I know when it comes to Nikon, the mode that you are in is more important than it might seem, and there's lots of "magic" (aka interactions between settings) going on behind the scenes that can be very hard to figure out. (There are alot of special interactions between AF modes, AF settings, AF points and Exposure metering modes in the Nikon bodies)

I don't know enough about the canon AF sensors and settings to talk about them, but I would expect them to be quite complex too. I would advice those who want to try and test the AF with a chart, to be very methodical and make sure that your chart is parallel to your sensor plane, and do some research before you try.

In general, if you can't get very accurate AF - switch to manual focus, and/or if you have the time, use CDAF (contrast detect autofocus through live view on a tripod mount.

The images you posted may be helpful and interesting to some - the results you have obtained, may vary at larger or shorter distances though

[OsmosisStudios]

Were you setting the focus point manually? Ie Were you choosing which point the camera should use, or was the camera choosing which point? How fast were you moving the camera after focus selection? Were you using focus lock?

[mrteacherdude]

"Bad things" happen to photos when IS is turned on and a tripod-mounted camera is expected to record an image of something that isn't moving. I wonder if the setting at AI servo is somehow messing with you in a similar way.

[autofocus]

Originally Posted by mrteacherdude

"Bad things" happen to photos when IS is turned on and a tripod-mounted camera is expected to record an image of something that isn't moving. I wonder if the setting at AI servo is somehow messing with you in a similar way.

+1...agree, try doing it with camera set to one shot. Both AIservo and AIfocus will allow the camera to take the shot even without locking in your focus. In one shot, the shutter won't even work until your focus is actually locked in

[CaptivatedByBeauty]

"Were you setting the focus point manually?" - Yes, I selected one AF point.

"How fast were you moving the camera after focus selection?" - For all the pictures I've posted the camera was stationary. See below for how I used AI-servo for getting a feel for how the AF sensors are reacting.

"Were you using focus lock?" - See below.

"Bad things" happen to photos when IS is turned on and a tripod-mounted camera is expected to record an image of something that isn't moving." Yes, I forgot to mention IS was off.

"Both AIservo and AIfocus will allow the camera to take the shot even without locking in your focus. In one shot, the shutter won't even work until your focus is actually locked in." - I'm specifically wanting to take the picture when the focus is not right to demonstrate the issues. Note below on hunting and AI-servo.

I always use AI-Servo and the AF-on button. That way I can press and release the AF-on button, or I can press and hold the AF-on button. In this case I needed to press and hold in order to have the AF sensor used recorded for DPP. It did also enable me to get a feel for the sensor behaviour by very slowly moving the camera and watching how the focus changed.

A couple of things I didn't mention.

1. In some situations the lens hunted backwards and forwards slightly and didn't stop. It "couldn't decide" which object to lock on to. Not those posted.

2. I did release the AF-on button and then press and hold again to get the camera to start the focus operation on a new fixed part of the subject. I did this because I think the AI-servo operation is different to the initial capture of focus. I think all the pictures I've posted were done this way, so I think in effect, it would be the same without AI-servo. After the initial focus, I didn't see any focus change if I kept the camera still (apart from the hunting).

3. IS was off. This is relevant not just for image quality (which is worse on a tripod with IS on), but also because the IS can cause the subject to move, which will affect the location of the AF sensor. Yet another complication.

I'm mainly concerned about situations in which the camera focuses only slightly differently to what I expect, so that I can't tell the focus is wrong in the viewfinder. I'd prefer not to review and magnify every picture just to check whether the camera focused on the point I put the AF point on!

Notice in one of my links the highly regarded 7D is not immune from this problem:

7D af point size and alignment
7D af point size and alignment: Canon EOS 7D / 60D - 10D Forum: Digital Photography Review

[CaptivatedByBeauty]

This is an excellent short description of the 2 issues, sensor size and location:

Andre's Blog • Cross type AF points in EOS 7D

His 1st picture demonstrates exactly what I've found in practice.

His 2nd diagram shows the issue of AF sensor size. The centre sensor seems to always be in the centre!

His 3rd diagram shows 2 issues, the size and the location of the AF sensor!
This is what my "real world" tests were done to demonstrate, but I wanted to investigate the effects of different subjects close to each other, and how the sensor/software would decide where to focus.

In order to focus with confidence, you have to memorize the direction of each asymmetric AF point and make sure that the cross inside does not extend beyond the subject

Yes quite.