"Light from an external object is directed by a lens (L1) onto the agile pixels plane of a programmable digital micromirror device (DMD). To initiate the imaging operation, the DMD micromirrors are set to spatially route the incident light to the CMOS sensor to create an initial target-scene irradiance map. Based on this initial image intelligence, the DMD is programmed in its CAOS mode to create specifically located agile pixels that sample image zones of interest.
This agile-pixel programming capability via the DMD allows the agile pixels to operate with different time-frequency coding methods such as frequency/code/time division multiple access (FDMA/CDMA/TDMA) schemes common in cell-phone radio-frequency (RF) communications.
Experiments demonstrate a CAOS-CMOS camera dynamic range of 82.06 dB, which can be improved upon by further optimization of the camera hardware and image processing.
“The CAOS camera platform, when used in unison with current multipixel sensor camera technology, is envisioned to enable users to make a smart extreme-dynamic-range camera, opening up a world of the yet unseen,” says Nabeel Riza."
Nabeel Riza publishes a Youtube video explaining the CAOS-CMOS camera principles:
Update: A new paper "Demonstration of 136 dB dynamic range capability for a simultaneous dual optical band CAOS camera" by Nabeel A. Riza and Pablo La Torre has been published in Optics Express. "...the experimental camera demonstrates an agile pixel extreme dynamic range of 136 dB, which is a 56 dB improvement over the previous CAOS-imaging demonstrations."
I think everyone who has worked on image sensors achieving greater than 82dB DR ought to email Prof. Riza with a copy of their paper so he is better informed. I am not saying this is not valid research, but he seems uninformed about the state of the art. Also, 1000x = 60dB, right? So 60dB higher than 22dB...or 60dB higher than his reported 82dB? Plus there is the whole optics issue of stray light when you couple the sensor to a lens which is non-trivial above 120dB DR.
ReplyDeleteThey've just published a new paper today in Optics Express claiming 136dB DR, a 50+dB improvement over the previously published version:
Deletehttps://www.osapublishing.org/DirectPDFAccess/8D0BB4AF-E0EF-C0FE-C592DD44EBA37721_356010/oe-24-26-29427.pdf?da=1&id=356010&seq=0&mobile=no
does this link work for anyone?
DeleteYou can try this one instead:
Deletehttps://www.osapublishing.org/viewmedia.cfm?uri=oe-24-26-29427&seq=0
When Shree Nayar published his DMD camera many years ago I used similar idea and from practical point of view it's probably impossible to get good imaging quality because of the tilted DMD (+ you want to have low F#).
ReplyDeleteIt is definitely possible to get good imaging quality with this scheme, and decent frame rate too. You can read the numerous already existing papers on the implementation of Compressive Imaging cameras.
DeleteI remember commenting a while back that 100 wasn't much, good for a few anonymous replies.
ReplyDeleteA logarithmic ADC with slopes adjusted by filling speed is what will likely 'win' with conventional methods. For extreme dynamic range the CUDOS lithographically fabricated waveguide can separate bright from dark and also increase the seeing (correct distortion caused by the atmosphere): http://www.cudos.org.au/research/Astrophotonics/Firefly.shtml .
My YT Channel has a recent Video where regular glasses are coated with nanoparticles that change the frequency of light, making Nightvision glasses that operate without electricity or very much size and weight - no doubt there is a method to use nanoparticles to create a log-curve that reads the light at a particular pixel and controls the curve to prevent well filling (or emptiness, by photon multiplication).
The above system with mechanical mirrors and circuitry to calculate the size of the agile pixel sounds slow, I question the Frame Rate and slew (rolling shutter).