On Sensor Phase Detection - Depth Map possible?

[Lichtgestalt]

with the new 70D sensor i wonder if it is possible to extract usefull depth information for a depth map from the sensor?

if i get it right EVERY photosite on the 70D sensor is recording phase detection?

im coming from 3D graphics and it was my dream for years to get depth information from a camera sensor. so i would have some kind of Z-BUFFER information for postprocessing.

that z-buffer could be used to do much better DOF in post processing (with PS for example).

im not asking if the 70D is recording this info... just if it is possible with this technology.

 

[Mt Spokane Photography]

While every photosite is measuring the distance, that does not mean it is recorded anywhere. Its used to instruct the lens to move to the focus point selected.

 

[Lichtgestalt]

im not asking if the 70D is recording this info... just if it is possible with this technology.

the question is could it be possible to readout the needed information.

could ML for example hack the firmware so that the info is stored in a separate file.

or could this technology be used in future cameras to record depth information.

i think olympus already tries something like that.
to use multiple focus information for postprocessing DOF.

i know im stretching things here but this on sensor phase detection could offer some neat possibilities.
for the first time every pixel could have it´s own depth information.

 

[Mt Spokane Photography]

Anything is possible, this is new technology, a separate internal processor is used for the AF system because it is so data heavy. Usually, that means it is difficult to gain access to the firmware from a external program.

I could only guess as to what form the data is in.

Canon likely has a roadmap for future development.

The camera with 3d capability is so processor intensive that it is limited to a very few mp, it might tale a supercomputer to handle 20mp.

http://www.amazon.com/Light-Field-Camera-16GB-Red/dp/B0099QUUBU

 

[Lichtgestalt]

The camera with 3d capability is so processor intensive that it is limited to a very few mp, it might tale a supercomputer to handle 20mp.

but it´s not a 3D camera.

the z-buffer would only need to store a 8 or better 16 bit number for each pixel.
that´s what the camera had to store in addition to the normal RAW or RGB pixel information.

i don´t want to get sharp focus after the fact. like a lightfield camera.

it may slows down the camera but in camera HDR does this too.
you could make it an extra shooting mode.

this is used to add DOF, to blur the image in postprocessing.

imagine photographing with a 17-40mm f4 and later give it the DOF characteristics of a 17-40mm f1.2.
or better example shot with a 18-55mm and then later give it the DOF characteristics of a 85mm @f1.2.

i do that all the time with 3D rendered images and videos.
it´s no problem to create an additional depth map with 3D programs.
and later i can choose whatever DOF i like for a video scene or image.

the postprocessing software is there already.
it would only need an additional depth map.

so could this technology tell you that pixel X is 3m away and pixel Y 5m?
or that pixel X is x-times further away then pixel Y.

or does it work completely different and such infos can not be extracted from on sensor phase detection?

 

[Mt Spokane Photography]

I mis-understood what you were asking.

Canon is not telling much about how the new AF works in practice. It uses many different dual pixels to determine the focus distance.

There is a patent, but it does not really tell you what you want to know either.