[MMVIII]

Do you mean "Block 80" on Figure 2 (Block 90 on Fig. 3)? That looks like the metering sensor... might have to do with the absolute brightness of the prism since in the Fig.3 arrangement they don't need to account for transmission through one of the internal surfaces of the pentaprism, on top of reducing the space in front of the prism and providing the famous extra 3mm of eyepiece relief?

... and if they have increased the resolution of the exposure sensor (92) to do face recognition à la Canon 90D, the new location might also, like you say, allow for superimposing the AF area on the OVF, showing the detected faces.

Great catch! Exciting times ahead, indeed.

Right! The construction drawing of K-new shows clearly less protrusion in front of the prism. (I know that the illustration is just schematic, but they used at least coherent scale, my overlay is just for fun and not scaled. Especially the second is not of the K-1, probably a generic APS-C).
In this drawing the light meter is not indicated, (92) is described as the "illumination light source" and is for the superimposing optical system (information display device) and is not the metering sensor (which until now actually sits in the upper part of the exit surface of the prism).

I think we can not derive any information on the resolution of an metering sensor and also the kind of information that can be provided in the viewfinder, but a completely new construction might add some new things. I am am very curious.

Attached Images

 

---

Last edited by MMVIII; 09-14-2020 at 10:23 AM.

[MMVIII]

Right! The construction drawing of K-new shows clearly less protrusion in front of the prism. (I know that the illustration is just schematic, but they used at least coherent scale, my overlay is just for fun and not scaled. Especially the second is not of the K-1, probably a generic APS-C).
In this drawing the light meter is not indicated, (92) is described as the "illumination light source" and is for the superimposing optical system (information display device) and is not the metering sensor (which until now actually sits in the upper part of the exit surface of the prism).

I think we can not derive any information on the resolution of an imaging sensor and also the kind of information that can be provided in the viewfinder, but a completely new construction might add some new things. I am am very curious.

Here I tried to produce a pseudo-orthophoto of a cut through a k-5. [I used photogrammetric linear transformation of tiepoints identical on the drawing and the image. The only stable points are the mount and the sensor. There are some uncertainties because the sensor shake reduction might have changed since k-5. This image may be an orthophoto for measurements only within the polygon of the tiepoints, there might be distortions outside of that]

What I get as a result is that the optical system of the viewfinder might be higher up than in the K-5. This 3mm further protrusion of the eyepiece are also recognisable. There is more space in the AF-chamber. The front protrusion of the housing above the mount is less.

Please don't take this to seriously, it is just for fun and I wanted to test if my method would work on that.

[bencoskater]

Hi,

I've read some parts of the patent on the auto-focus system. I might me wrong but it seems that Pentax is not using the SAFOX sensor anymore but the RGB sensor to measure white balance, correction processing, etc, and finally to perform the auto-focus as described in flowchart figure 5 starting by S101 (chromatic wavelength detection using RGB sensor, this is processed also in IR wavelength).
I am not an expert in optical but by reading this and if and only if I am right the joystick will be used for focus points and the number of focus points will depend on the sensor coverage (I assume Pentax will define fixed points). I think we will have a new overlay screen in the OVF, with more points?


I hope this is how the auto-focus is implemented in the new DSLR and giving up the SAFOX. However, how will it perform on lenses having a pronounced chromatic? How fast will it be to focus (depending on CPU/RAM and speed of BUS)?


let's see what Pentax will tell about all of this!

[house]

Hi,

I've read some parts of the patent on the auto-focus system. I might me wrong but it seems that Pentax is not using the SAFOX sensor anymore but the RGB sensor to measure white balance, correction processing, etc, and finally to perform the auto-focus as described in flowchart figure 5 starting by S101 (chromatic wavelength detection using RGB sensor, this is processed also in IR wavelength).
I am not an expert in optical but by reading this and if and only if I am right the joystick will be used for focus points and the number of focus points will depend on the sensor coverage (I assume Pentax will define fixed points). I think we will have a new overlay screen in the OVF, with more points?


I hope this is how the auto-focus is implemented in the new DSLR and giving up the SAFOX. However, how will it perform on lenses having a pronounced chromatic? How fast will it be to focus (depending on CPU/RAM and speed of BUS)?


let's see what Pentax will tell about all of this!

Sounds very interesting! Would be nice if Pentax manages to innovate in the autofocus space.

[ffking]

Would be nice if Pentax manages to innovate in the autofocus space.

Doing things differently is almost a trademark - interesting revelations to come

[rawr]

Pentax is not using the SAFOX sensor anymore but the RGB sensor to measure white balance, correction processing, etc, and finally to perform the auto-focus

That would be a very interesting development. I've often thought that Pentax has been under-utilising the capabilities of the current RGB metering sensor. But the 'resolution' of the metering sensor would probably have to be improved in order for that sensor to do a decent job with AF. And it might help to put some more CPU horsepower behind it too, if it's asked to do extra work.

Pentax K-3/645Z/K-1/KP 86k metering sensor


At present the 86k metering sensor only 'see's' a scene via a combined 45x30 matrix of the three RGB colours. This is a big improvement over the old 77 segment metering, and while it may be able to help with scene detection (inc face detection, perhaps) it is not precise enough to take on all the work of auto-focus. The pix below are related and show the limited detail available from the current 86k RGB metering sensor:







But then again, current AF on the K-1, for example, does the job just using 33 (or less, most times) PDAF points, so even a 45x30 matrix could be made to work, as it would be the equivalent of the system using 1350 AF points.

[bencoskater]

That would be a very interesting development. I've often thought that Pentax has been under-utilising the capabilities of the current RGB metering sensor. But the 'resolution' of the metering sensor would probably have to be improved in order for that sensor to do a decent job with AF. And it might help to put some more CPU horsepower behind it too, if it's asked to do extra work.

Pentax K-3/645Z/K-1/KP 86k metering sensor


At present the 86k metering sensor only 'see's' a scene via a combined 45x30 matrix of the three RGB colours. This is a big improvement over the old 77 segment metering, and while it may be able to help with scene detection (inc face detection, perhaps) it is not precise enough to take on all the work of auto-focus. The pix below are related and show the limited detail available from the current 86k RGB metering sensor:







But then again, current AF on the K-1, for example, does the job just using 33 (or less, most times) PDAF points, so even a 45x30 matrix could be made to work, as it would be the equivalent of the system using 1350 AF points.

Wow you did a great job rawr.

In the patent they use a 3x3 matrix for each color:
https://ipforce.jp/Kouhou/img?n=0006740666.tif&id=000005&c=B9&t=111813181
https://ipforce.jp/Kouhou/img?n=0006740666.tif&id=000006&c=B9&t=111813181

Here is an explanation of the auto-focus system:
[0026]
When the release button is pressed halfway, the focus is adjusted. The system control circuit 40 executes AF processing based on the pixel signal read from the image sensor 32. That is, the amount of defocus is calculated to shift the image plane to the in-focus position. In addition, AF processing by multipoint ranging can be executed.
[0039]
In order to form a pair of divided images according to the pupil division phase difference method, the focus detection pixel pair PB is formed with light-shielding films SL and SM that block half of the pixels. The light-shielding films SL and SM are formed at positions symmetrical with each other so that a pair of divided images are formed from each focus detection pixel pair.
[0040]
The focus detection pixel pairs PB, PG, and PR are not limited to the configuration in which they are arranged adjacent to each other in the oblique direction, and may be adjacent to each other in the row direction and the column direction. The light-shielding films SL and SM may also be formed at positions according to the arrangement direction of the focus detection pixel pair so that a pair of pupil division images can be obtained.
[0041]
For AF processing by multipoint distance measurement, a plurality of (nine here) divided distance measurement areas are defined for the entire image pickup area IM of the image sensor 32 ( see FIG. 5 ). When the AF process is performed, the focus detection process is performed based on the focus detection pixel pairs PB, PG, and PR belonging to the divided ranging area in which the target subject is projected.


Multi-point focus will be supported. All this part seems to confirm the end of SAFOX.
might be a 128K sensor (cross our fingers)

PS: Everything translated by google

---

Last edited by bencoskater; 09-15-2020 at 07:52 PM. Reason: images missing

[cfraz]

Hi,

I've read some parts of the patent on the auto-focus system. I might me wrong but it seems that Pentax is not using the SAFOX sensor anymore but the RGB sensor to measure white balance, correction processing, etc, and finally to perform the auto-focus as described in flowchart figure 5 starting by S101 (chromatic wavelength detection using RGB sensor, this is processed also in IR wavelength).
I am not an expert in optical but by reading this and if and only if I am right the joystick will be used for focus points and the number of focus points will depend on the sensor coverage (I assume Pentax will define fixed points). I think we will have a new overlay screen in the OVF, with more points?


I hope this is how the auto-focus is implemented in the new DSLR and giving up the SAFOX. However, how will it perform on lenses having a pronounced chromatic? How fast will it be to focus (depending on CPU/RAM and speed of BUS)?


let's see what Pentax will tell about all of this!

Interesting. If by "not using the SAFOX sensor anymore but the RBG sensor" you are saying the metering sensor completely replaces the PDAF sensor in K-new for autofocus duties, I'm not sure that all the bits of info we have so far add up to that conclusion.

Early images of the K-new with no lens installed apparently showed the reflection of a larger (than in K-3) PDAF sensor under the mirror box, implying a larger coverage area. The cutaway drawing showing the new prism arrangement also shows a large area below the mirror box - I don't see the purpose of that space if not for a PDAF sensor.

I think prior Pentax cameras used the metering sensor to assist autofocus in tracking objects (by color, maybe shape), essentially giving the system cues on which PDAF point to use as object moved across the frame. My take on this may be completely wrong.

The sensor in the patent seems to be a Foveon-like sensor. Maybe it can be used together with a PDAF sensor to enhance the AF system? Or maybe the patent is for a future camera's AF system? Anyway, I'd be surprised if K-new doesn't use PDAF. I have no idea if they'll call it SAFOX or something else. But this is all guesswork by me based on what I think I know - again I may be completely wrong.

[rawr]

Multi-point focus will be supported. All this part seems to confirm the end of SAFOX.
might be a 128K sensor (cross our fingers)

They may be saying that the metering sensor can help the camera do AF even better. But not necessarily that they plan on relying on the metering sensor alone to to all the camera AF work.

The patent seems an evolution of what they are currently doing. At the moment the 86k imaging sensor can feed colour distribition detection and colour motion detection, for example, into the camera CPU to help the PDAF focus system do it's job better.

Now they may be revealing that they have been able able to [finally] evolve their system so that the imaging sensor can help with even more AF tasks - like face/eye detection, for example [pic related]. All without having to switch the camera into LiveView. That type of improvement would be cool.

[repaap]

They may be saying that the metering sensor can help the camera do AF even better. But not necessarily that they plan on relying on the metering sensor alone to to all the camera AF work.

The patent seems an evolution of what they are currently doing. At the moment the 86k imaging sensor can feed colour distribition detection and colour motion detection, for example, into the camera CPU to help the PDAF focus system do it's job better.

Now they may be revealing that they have been able able to [finally] evolve their system so that the imaging sensor can help with even more AF tasks - like face/eye detection, for example [pic related]. All without having to switch the camera into LiveView. That type of improvement would be cool.

They(ricoh) did say that old system (safox) could not be developed any further due restrictions in hardware. So no matter what they developed, it did little if even that. This might be what has been changed. I can easilly believe that good old Safox is gone. There is not 'just more AF points' what is coming, and this might be why they are so confident about it's performance. As theysaid, there is atleast 3 main processors what are completely new.

[Serkevan]

I'm betting on similar AF performance to 90D/D7500 or a bit better. The higher-tier Canikon DSLRs use the RGB sensor for face detection to aid AF, as well as for depth analysis for Z-axis tracking (the infamous "cyclist moving towards camera" test).

[rawr]

As theysaid, there is atleast 3 main processors what are completely new.

Hopefully those new processors are powerful, modern chips that have been given lots of work to do.

While I remain skeptical that they have ditched the whole SAFOX assembly (because it still looks like there remains a place in the mirror box for it), the SAFOX hardware and software certainly needs an upgrade and has hopefully been enhanced and 'hot-rodded' in the K-new.

Depending on how fancy they went with the new AF hardware, maybe they even decided to dedicate a chip just to handle AF, like some sports Canikon's have done.

I guess another of the new chips may be an 'Accelerator' chip v2, doing noise reduction mainly, maybe other image quality or correction stuff.

But then again all these functions could just be performed by the latest iteration of the Pentax PRIME imaging engine (PRIME V?) - ie the main processor in the camera - which the K-new will be the first to use, if they have indeed decided to upgrade PRIME 4.

So I bet that if there are indeed '3 brand new processor chips' in the K-new, maybe we'll see:
(1) PRIME V
(2) Accelerator v2
(3) RGB Metering Sensor v2.

and maybe a bunch of other new specialized co-processors. Hopefully a LOT of fast RAM will find it's way into the K-new motherboard as well.

[Serkevan]

While I remain skeptical that they have ditched the whole SAFOX assembly (because it still looks like there remains a place in the mirror box for it), the SAFOX hardware and software certainly needs an upgrade and has hopefully been enhanced and 'hot-rodded' in the K-new.

I don't think they will remove the sub-mirror assembly, it's still present in all other DSLRs no matter how "advanced" their AF is. Take the D5, still showing a fair bit of space dedicated to the AF assembly at the bottom of the mirror box:


I assume the RGB sensor will do scene recognition to feed the tracking algorithms while the new and improved phase detection module handles the actual measurements. Of course, I guess we can count on wider coverage, more points with smaller spacing for finer tracking, and better sensitivity on non-center points as "hardware upgrades". But software is the key point to good AF.C from what I understand, and it needs to run on faster hardware - dedicated AF processor like you mentioned, hopefully?

[rawr]

I assume the RGB sensor will do scene recognition to feed the tracking algorithms while the new and improved phase detection module handles the actual measurements

My guess is that Pentax may emulate [fellow Milbeaut image engine user] Nikon in terms of their AF evolution, and also maybe a few other camera functions.

You can see some of the similarities between Pentax and Nikon's processing logic, from an overview of the D6 flagship's 'Advanced Scene Recognition System'.


The D6 brochure [3MB PDF] also may give an idea of where Pentax might be going with their AF:

"POWERFUL, BRAND NEW AF SYSTEM

Advanced Scene Recognition System improved to further enhance AF performance

AF performance is enhanced even further by improvements in the Advanced Scene Recognition System, which works together with the D6’s dedicated AF and EXPEED 6 image-processing engines. Face and upper body information, defocus information and motion detection information improve the camera’s ability to maintain focus on laterally moving subjects in 3D-tracking and auto-area AF modes.

In a first for optical viewfinder shooting*, the D6 also prioritizes focus point selection based on the positions of the subject’s eyes. This makes it easier to keep the subject’s eye in focus for portrait opportunities.

* When autofocusing using AF sensor module."

The D6 does all this with a relatively old 180k pixel RGB metering sensor (introduced with the D5 in 2015, I think), working alongside the AF system, so it wouldn't be a big step for Pentax to also upgrade their 86k imaging sensor to something that could work the way the D6 works.

[repaap]

Hopefully those new processors are powerful, modern chips that have been given lots of work to do.

While I remain skeptical that they have ditched the whole SAFOX assembly (because it still looks like there remains a place in the mirror box for it), the SAFOX hardware and software certainly needs an upgrade and has hopefully been enhanced and 'hot-rodded' in the K-new.

Depending on how fancy they went with the new AF hardware, maybe they even decided to dedicate a chip just to handle AF, like some sports Canikon's have done.

I guess another of the new chips may be an 'Accelerator' chip v2, doing noise reduction mainly, maybe other image quality or correction stuff.

But then again all these functions could just be performed by the latest iteration of the Pentax PRIME imaging engine (PRIME V?) - ie the main processor in the camera - which the K-new will be the first to use, if they have indeed decided to upgrade PRIME 4.

So I bet that if there are indeed '3 brand new processor chips' in the K-new, maybe we'll see:
(1) PRIME V
(2) Accelerator v2
(3) RGB Metering Sensor v2.

and maybe a bunch of other new specialized co-processors. Hopefully a LOT of fast RAM will find it's way into the K-new motherboard as well.

They did with DFA85 something like that they basically put on the table what they have and what others have. Then they did what they needed to get the best out of it, no matter what. I believe that this is what they did with K-New, and so far it looks like it is the case. All has been improved to K-3II. No reason to believe they would stop there(bigger brighter OVF).

This remains to be seen still

Not saying that Safox is gone, but it should be improved and a lot must have been changed inside.. they added one more Pin to battery, for better power and faster recharge, perhaps there is more demand for that improvement?