[bwDraco]

I've posted a diagram of autofocus systems from various manufacturers. Though it's not very neat, it states how the points in autofocus systems are typically arranged, to the best of my knowledge. I may not be correct on all points, though, so let me know if there are any errors.




Points indicated by a vertical line are sensitive to horizontal contrast, while points indicated by a horizontal line are sensitive to vertical contrast. Cross-type points are indicated by an intersection of both horizontal and vertical lines. Points indicated by horizontal, vertical, and diagonal lines are high-precision dual cross-type that are sensitive to diagonal contrast with lenses of at least f/2.8. For more information, see this Cambridge in Colour tutorial.

As you can see, the autofocus points are often arranged in groups. However, I know from experience that Canon and Nikon AF systems can indicate that one, many, or all points are in focus at the time the AF system is activated, regardless of whether these points are directly connected to each other as in the diagram. (I'm not sure about Sony, though.)

Compared to Canon and Nikon, there are several vital limitations with the Pentax AF system that hamstring its ability to cope with fast action. The most critical limitation is that no more than two adjacent points will activate at any given time, and the linear horizontal-sensitive points at the sides will not work together with any of the cross-type points. This has several important consequences that hinder the implementation of a reliable predictive focus tracking system:

• It cannot use a large group of points simultaneously to track a moving subject, greatly reducing its ability to maintain a lock on a fast-moving subject or predict its position.
• The camera will not use the outer points of the AF array unless there is nothing to focus on in the center of the frame, even when the camera is set to select from all 11 points.
• Because it isn't considering the objects around the active AF point(s), the AF system is not capable of accurately distinguishing between the intended subject and any obstructions, making it prone to inappropriately refocusing when such an obstruction appears in front of the subject.
• Two points is often not enough to cover the entire subject. This can make it impossible to keep track of the entirety of a moving subject, making it vulnerable to losing track of a fast-moving subject moving across the frame or directly toward the camera.

Furthermore, Pentax AF points are much larger than that of competing cameras, since the nine cross-type points cover a fairly large area of the frame and are directly connected to each other. (For comparison, Olympus, though not shown in the diagram, uses a similar layout in the E-3 and E-5, but with all points dual cross-type in a shape similar to that of a number sign [#], and the points are not connected together.) Points this large simply can't be arranged with sufficient density to form an area AF system such as that of the Nikon D7000. As such, there are too few points in the AF system, making it significantly more difficult to keep track of a fast-moving subject.

My analysis is intended to explain why Pentax AF is widely believed to be inferior to C&N systems. Keep in mind, though, that I'm not trying to bash Pentax AF, but to help Pentax make it better. The current SAFOX IX system is quite fast for static subjects in both outdoor and indoor environments. Given the limitations listed above, here's what Pentax can do to improve AF performance with moving subjects:

• Allow any number of AF points to work together, even if they are not connected to each other. This is the single most important thing Pentax can do to improve AF performance. It requires treating each point as a separate entity, yet allowing them to be used simultaneously in an arbitrary arrangement. Doing so eliminates many of the issues and limitations listed above and paves the way for much more advanced predictive tracking algorithms to be implemented. It may be possible to implement this for current cameras via a firmware update, but hardware limitations may prevent this from happening.
• Shrink the AF points and arrange more of them in a dense grid. Smaller AF points can more precisely and predictably focus on smaller subjects or parts or subjects. More importantly, though, they can be arranged together in large, dense groups to form an area AF system as mentioned above. The result is a array of points, each with subject-to-camera distance information, which can be used to predict the position of the subject more accurately. This can also improve AF performance with static subjects. Even if an area AF system is not implemented, relatively few points covering a large segment of the frame should not be connected together as they are in the current design, as this will require that the points be larger than they should be.
• Develop better predictive AF tracking algorithms. Improvements in AF hardware are not sufficient. Without better software algorithms, the AF tracking performance will not be much better than the existing system.

I hope Pentax takes these suggestions seriously so that we can see a much better AF system in the next generation of cameras, so that it can meet the rigors of sports photography as well as other fast-action photography. This will make Pentax much more attractive to professional photographers than it currently is.

Moderators: Consider making this thread sticky so it can get more attention from the community as well as Pentax itself. These are important suggestions to Pentax that can have a substantial impact on AF performance in future cameras.

--DragonLord

---

Last edited by bwDraco; 04-09-2012 at 07:44 AM.

[JinDesu]

What a cool read - your diagram of the Pentax AF grid appears to jive with a few other reports (including AF sensor size and focusing tendencies).

[aurele]

interesting ready. I have to say that i could'nt agree more about the fact that Pentax' AF points are too big, and that can be a bit disturbing.

[Douglas_of_Sweden]

I would be surprised if Pentax r&d staff couldn't figure this out by themselves. From that follows that there must be a reason why they have not already implemented this sort of improvements.
What would your guess be for the reason?
I can't help wonder if they simply discovered that their processor was not up to the task to do the calculations using more sensors at the same time. Both 1 and 3 on your list may be limited by the processor.

[bwDraco]

A-DEP mode, which uses many or all of the AF points of the camera simultaneously to produce depth of field sufficient to cover all of these points, was seen on film Canon EOS cameras from as early as the 1990s. This requires the ability to use an arbitrary combination of AF points. If this technology existed for this long, there's no reason any modern image processor can't do it. Either there is a limitation in the AF hardware itself preventing this from being implemented, or Pentax hasn't fully understood how to use an arbitrary number or arrangement of AF points to focus on a subject.

It's likely Pentax would have to redesign the AF system from scratch, rather than improve upon a existing system as they did with SAFOX IX. The AF array has remained essentially the same from SAFOX VIII--11 points, 9 cross-type--only that the optics and sensitivity have improved so that performance is increased. The design and layout of the AF points and the algorithms controlling the AF system, however, still have the inherent limitations discussed above.

--DragonLord

[Docrwm]

Very good read and consistent with my recent research into the D700 and the Canon 7D.

[benhai]

What a comparision! Nikon Multi-CAM 1000 system. Nikon D80 and D200 are using them. Do you indicate that K-5's AF is still struggling to keep up with Nikon's 5-6 years old system ^_^. Actually that's one of the reason why I have to grab an old Nikon D200 recently. You kind of proving my decision.

[bwDraco]

The current AF system hardware and algorithms are deeply entrenched in the Pentax K system and it is not easy to totally redesign the AF system. This would explain the incremental nature of the improvements in the SAFOX IX system relative to SAFOX VIII. Hopefully, given that Ricoh has much deeper pockets than Hoya, it should be possible for R&D to develop an all-new autofocus system addressing the aforementioned problems and limitations.

--DragonLord

[RonHendriks1966]

I wouldn't mind having more of this in my next camera.

[Reportage]

they would have to find their own solution as i am guessing canon and nikon have patent how their af works and construction. Ricoh not exactly well known for ultra high speed af.

looks like pentax is stuck in only the body design and ways to make the picture nicer.

I am a sports shooter using a K20D while the rest are along the lines of 7D, D700 and even their flagship models. I love the Pentax lens but for my work, accurate and lightning fast af tracking is what i am looking for and doesnt seem to be on Pentax present roadmap.


To capture this on a 3fps K20D, either have to luck out or really know what you are doing.

[Rondec]

The current AF system hardware and algorithms are deeply entrenched in the Pentax K system and it is not easy to totally redesign the AF system. This would explain the incremental nature of the improvements in the SAFOX IX system relative to SAFOX VIII. Hopefully, given that Ricoh has much deeper pockets than Hoya, it should be possible for R&D to develop an all-new autofocus system addressing the aforementioned problems and limitations.

--DragonLord

Is it possible to license an existing auto focus system from some other company (Sony, Samsung)? Or do you have to develop your own in house?

It just seems as though a number of components of the Pentax system (PRIME) were developed by other companies and then cobbled in.

The other issue with AF system is SDM lenses, which tend to be pretty slow and poorly responsive. Seems like HSM lenses on Pentax bodies do a better job of getting where they need to get and following focus.

[bwDraco]

Is it possible to license an existing auto focus system from some other company (Sony, Samsung)? Or do you have to develop your own in house?

It just seems as though a number of components of the Pentax system (PRIME) were developed by other companies and then cobbled in.

The other issue with AF system is SDM lenses, which tend to be pretty slow and poorly responsive. Seems like HSM lenses on Pentax bodies do a better job of getting where they need to get and following focus.

PRIME refers to the image processor. The autofocus system is called SAFOX.

As I stated before, for best performance, Pentax R&D needs to start from scratch and not attempt to improve upon an existing system that is subpar compared to C&N, however expensive it may be.

I cannot emphasize this enough: AF points that cannot function independently of each other and work in arbitrary combinations don't make for good subject tracking performance. At a minimum, AF algorithms need to be reworked.

I also need to state that independent AF points in arbitrary combinations can be used to simultaneously cover a larger subject and optimize depth of field. With AF points that are tied to each other as in the current SAFOX IX system, this isn't possible and the entire subject may not be uniformly in focus.

--DragonLord

[bwDraco]

I know I'm reviving an old thread, but can anyone say what's changed with the K-30's AF algorithms? I know there's AF point expansion, and the linear horizontal-sensitive points can work with the cross-type points in this mode, but in Auto 5 or Auto 11, can the points work in arbitrary combinations?

--DragonLord

[slip]

I would really like pentax to come out with a high end camera that drops the movie mode and spends R&D on cutting edge focus, but not as expensive as a 645D
A quicker way to switch from different metering modes on the fly would be helpful in mixed light as well

randy

[bwDraco]

I would really like pentax to come out with a high end camera that drops the movie mode and spends R&D on cutting edge focus, but not as expensive as a 645D
A quicker way to switch from different metering modes on the fly would be helpful in mixed light as well

randy

What's the advantage with removing movie mode? The technology is already well-refined in the K-30, so I don't see how movie mode would require a substantial amount of effort to implement.

With regard to metering, we already get physical switches on the higher-end models, so why would this be necessary?

--DragonLord