Can exposure be bracketed by many sensor readouts during single exposure?

Question

I am getting into some HDR photography and was thinking of a new ways that camera manufactures could help us in capturing HDR pictures.
Say you want to capture three image and you have a constant Aperture and ISO with shutter speeds of 1/2, 1 and 2 seconds (for -1, 0 and +1 exposure compensation).
You set the shutter speed to 2 seconds and press the shutter release button.
1/2 seconds later, the camera takes a readout from its sensor and saves the output to a file.
Another 1/2 second later (for 1 second exposure) it takes another readout and saves the data.
1 second after that (for the 2 second exposure) it takes another readout of the sensor and saves the data.

So for a single shutter count, we get 3 images at different points in time.

I am not sure if any cameras does that but is it technically possible to do it?

Answer

This is not really my area, but I believe that readout is achieved by discharging the charge that builds up at each photosite due to arrival of photons. This would mean sampling pixels during the time the shutter is open would be equivalent to taking a series of separate exposures using an electronic shutter.

Now in this scheme you could then average several images in the shadow areas to reduce noise and hence increase dynamic range. But that wouldn't really be much different to shooting a series of images using a traditional HDR approach, plus you'd have to deal with rolling shutter issues on CMOS chips.

It's a good idea, I'm just not sure it's possible. The best you could hope for would be to have a rough way of detecting when a photosite was going to saturate. You could then sample the photosite at that point, and multiply the value post digitization by the proportion of exposure time it had. That would prevent overexposure, increasing the amount of light the sensor could handle before saturation hence increasing dynamic range. However, CMOS chips read line by line, so I imagine it would be very hard to read individual pixels ahead of time.

You'd also get some weird effects with moving subjects with this scheme, or indeed any time based HDR approach. The best idea in my opinion was Fuji's superCCD R which had an extra set of low sensitivity pixels (giving two readings in each location). This effectively allowed two different exposures to be recorded at the same time.

Check more discussion of this question.