ChipSight Slides

May 1, 2011


ChipSight is:

– a camera interface peripheral for microprocessors that outputs visual information including object shape and location.

– aimed at high volume products where small size, low power, and short development time for new features are critical.

– a collection of algorithms that perform the pre-processing (pre-vision, pre-cognitive) functions required by computer vision applications. ChipSight algorithms are efficient, have low memory requirements, and are designed to be used together as a microcontroller peripheral.

The advantages of the ChipSight peripheral are:
1. Eliminates most image processing for the developer
2. Eliminates FPGA
3. Eliminates frame buffers for region processing
4. Makes computer vision available to more developers and OEMs
5. Enables computer vision for consumer products
6. Scalable down to $1 cameras and $0.25 microcontrollers
7. Standardizes the image information interface


Example of changing megabytes of pixel data into a few bytes of information.

Input image

Feature encoding

Lane departure warning

Internal functions of ChipSight.


Adjustable precision extents. Only as much information as the application requires.

Region extent for object location.

Low resolution shape information.

High resolution shape information.

Input image.


Reading the values of currency.


Multi-touchless projector demo. Real gesture recognition, not phone rubbing.


How it works.



Dor May 3, 2011 at 1:37 am

Thank you for launching this great website.
Hope to learn more on embedded computer vision application.
Good luck

Craig Sullender May 3, 2011 at 1:48 pm

Thank you Dor!

I am shopping for the CPLD to use on the 4Cam, and comparing the different WLC-type cameras.

tmk May 26, 2011 at 3:43 am

< $5 computer vision periperhal sounds great!

When can i get one? :)

I wanted to try my hand at some video processing, and got a 60fps consumer USB camera and a 1ghz arm box, but the cpu seems to burn all its time processing buffers, and the linux drivers for the camera aren't so hot… I spent a lot more than $5 on the rig too :)

Probably trying to solve the problems at too high a level.. based on this blog, it seems like the arm box should have enough juice to handle it.

Any tips?

Craig Sullender May 26, 2011 at 4:47 am

Hi tmk, thanks for commenting.

Yes that’s what happens, chewing through pixels uses up the processor. Add an OS on top and you might not be able to get much done.

Another commenter recommended the STMicro STM32 and the CMUCam.

I am building a camera module with pixel-processing (ChipSight). Subscribe to this blog via RSS or Twitter and I’ll keep you posted on developments.

Tom May 26, 2011 at 3:47 pm

Can you say at this point when it will be available?

Craig Sullender May 26, 2011 at 4:02 pm

The camera module? Several weeks at least. A month at best for the hardware, then a month for drivers and tools. I wish I had 10 engineers.

Steve Battazzo May 27, 2011 at 7:05 am

Hi Craig, this is a pretty interesting project.
Are you sure the line buffers will fit on a CPLD? Have you written up your HDL and tried to synthesize it for a target device?

I don’t really know much about what’s in the Altera CPLDs, but I can pretty much guarantee that it would not fit in even the largest Xilinx CoolRunner-II (and those are very expensive anyway). A smal FPGA with even just a few block RAMs configured as FIFOs could do it.

Craig Sullender May 27, 2011 at 12:30 pm

Hi Steve!

Yes I have put it into FPGAs a few times, not CPLDs. An 18kbit BRAM is enough for video lines over 1k pixels long. The CPLDs I am looking at have plenty of BRAM, not that I wouldn’t take more if I could get it in the same package and price. Check out the Lattice MachX02.

Comments on this entry are closed.