I seem to recall Canon working on a new autofocus system which got me wondering earlier... could DPAF technology be implemented in the AF module. As opposed to using it in the image sensor, couldn't it possibly be used in the AF modeule to allow for full-scene AF coverage with RBG metering and tracking? Is DPAF responsive and speedy enough to replace the 61pt AF system in the 5D-III/1Dx or is this idea perhaps flawed in some other way?
New Implementation for DPAF Technology?
- Thread starter StudentOfLight
- Start date
The name "Dual Pixel AF" refers specifically to the focus system found on the Canon 70D, where a pixel captures light only on the left side, and the other captures the right side only. This system was designed to work with millions of double pixel. I'm no expert on optics, but I believe the angle of incidence of the light reaching the AF module is too perpendicular, making the difference between left and right very thin to allow comparison and determine the distance accurately.
Upvote
0
Wouldn't that defeat the purpose of DPAF? I believe that the goal is to merge contrast and phase detection on the sensor so that you won't need a separate auto-focus module, thus removing one of the needs to have a mirror in your camera. The last piece would be to make an electronic view finder that is acceptable to use, then you would have a mirror-less camera that would focus as fast as an SLR. Not to mention, with processor technology getting faster and faster, frame rates could get ridiculously high as the mechanical component, and it's speed limitations, of the camera would be removed.
This is in the future of course, but I believe it's closer than we think.
This is in the future of course, but I believe it's closer than we think.
Upvote
0
StudentOfLight said:
This is a bit counterintuitive. DPAF is a very specific image sensor design that utilizes a split photodiode for each pixel, allowing the light from the left and right halves of the lens to be detected independently. This enabled phase detection capabilities in the image sensor, which compliments the image sensing capabilities. The concept has no meaning or relevance in a standard DSLR AF unit...it simply does not apply.
Standard DSLR AF units are not image sensors. They are not neatly arrayed grids of tiny pixels, they are actually strips of HUGE pixels (relatively speaking) that are ULTRA sensitive to light. They are also designed in such a way that phase detection is their intrinsic behavior, and they are exceptionally good at it. DPAF is new and intriguing, but it doesn't perform at the same level as a dedicated PDAF unit, and probably won't for some time. A dedicated PDAF unit allows extremely fast, precise AF, usually with a single focus group movement. With Canon's 61pt AF system, focus can usually be nailed, with extremely high accuracy and precision, on a repeatable basis, with a single focus group movement...it's so accurate, it rivals or surpasses contrast detection AF and manual AF (both of which have high precision.)
There is no reason, or for that matter no way, to apply DPAF technology to standard PDAF units.
Upvote
0
StudentOfLight said:
You do realize that the mirrors in current DSLRs have, for a very long time, been half-transparent in the center? That's how they work. The AF unit is housed in the bottom of the mirror box. The secondary mirror receives light through that semi-transparent mirror, deflects it down into the AF unit, which houses a dedicated PDAF sensor.
They are called half-silvered mirrors, and it isn't new. There are also pellicle mirrors, used in some DSLRs. The entire mirror is half-silvered, and the mirror does not move, so some light is always redirected up to the viewfinder and some is passed to the sensor. Using a pellicle is like always having a 1-stop ND filter on your camera...your sensor has to work twice as hard, meaning it has twice the noise. That's why pellicle mirrors aren't used in more than less than a handful of cameras...no one likes noise.
DPAF is a technology with a very specific design, for a very specific purpose, and it won't be bringing new capabilities to other technologies any time soon. It's not really all that amazing...all it does is change from a single large, square photodiode per pixel, to a split photodiode per pixel, with independent readout of each half. All that does is allow the left and right halves of the light cone from the lens, the left and right "phases", to be processed independently. The design of a DPAF sensor is actually not as good for AF as a dedicated PDAF sensor. It puts more of a burden on firmware, instead of having advanced AF capabilities built right into the AF sensor itself (like diagonal AF lines, double AF and double diagonal AF lines, ultra high sensitivity to very low light, extremely high precision, etc.)
Upvote
0
I've heard of a pellicle (pelican) mirror but haven't read up about it, so thanks for the feedback.
My thoughts were that if an image sensor was used then it could do RGB metering as well (instead of having a separate metering sensor on top of the pentaprism) thereby freeing up space for a Wifi or Radio module. However, as you say since the sensitivity/precision is not on par it with traditional PDAF there is no justification. No one would want to compromise AF performance unnecessarily.
My thoughts were that if an image sensor was used then it could do RGB metering as well (instead of having a separate metering sensor on top of the pentaprism) thereby freeing up space for a Wifi or Radio module. However, as you say since the sensitivity/precision is not on par it with traditional PDAF there is no justification. No one would want to compromise AF performance unnecessarily.
Upvote
0
StudentOfLight said:
I think it will be a good while before we see focal-plane AF systems comparing to dedicated AF systems. Right now, in the case of Canon's DPAF, they basically turned the whole sensor into one giant line sensor. That's about as advanced as Canon's first AF sensor back in the 80's, which was also a single strip line sensor. Were going to need to see DPAF evolve into QPAF, where the sensor can be read twice, once for a horizontal line sensor and once again for a vertical line sensor (now were up to the late 80's/early 90's); beyond that were going to have to see advancements that bring us multi-orientation phase detection at the focal plane (horizontal, vertical, and both diagonals in directions 90° perpendicular); were going to need to see significant improvements in the speed and accuracy of focal plane AF (at this point, I believe the low-light edge will disappear, and fall behind that of dedicated AF sensors, since right now the sensor can be exposed for longer to support very low light AF...but when speed becomes critical, image sensors are WORSE than the huge pixels of dedicated AF line sensors at sensing in the dark), etc.
Focal-plane AF is in it's infancy. It's trendy and cool right now, and it brings new capabilities to the table for a new class of digital camera. But that new class of digital camera is demonstrably at least a couple generations behind DSLRs in terms of critical functionality. It will mature, and at some point mirrorless cameras will ultimately become superior to DSLRs in every way, including AF, but that is a very long way of into the future (especially with the bulk of western economies, the US and Europe, effectively rejecting mirrorless at a mass scale at this stage.)
DPAF, as much as everyone here loves to talk about it (probably because it's really the only key sensor innovation from Canon in quite some time), is really a very simple, minor EVOLUTION of prior focal-plane AF technology. Canon took the next logical step, but overall the technology is in it's infancy. DPAF, if one were to rank sensor innovations from the last three years, ranks pretty low overall. It isn't the kind of magical technology everyone hopes it is...it's just a basic evolution of technology that already existed in a simpler form.
Canon really needs to be radical with their sensor innovation to really make anyone start thinking about them as an innovative leader in sensor technology again. They need to move to a fully on-die image processing pipeline (move ADC onto the sensor die), preferably go hyperparallel, also preferably move to either a dual-stage CDS (analog and digital) or at least a digital CDS, etc. All of this would really require a die shrink...which means Canon needs to move to a 180nm process. Even if Canon does that...it still isn't going to make DPAF a particularly intriguing or compelling technology in the world of CMOS image sensors. Even if Canon moves to QPAF (I haven't seen any such patents yet), that would still be a minor evolution of DPAF...it still wouldn't rank high among all of the radical innovations that are constantly occurring in the image sensors world.
Upvote
0
Similar threads
