Firmware Hack Allows Video Analysis On a Canon Camera

An anonymous reader writes "Researchers from the University of Liege in Belgium have been able to perform real-time video analysis on a regular Canon digicam (video link) without any hardware modification. The results are shown directly on the digicam's screen. They use a hacked version of a popular open-source alternative firmware for Canon cameras: CHDK. This is a proof-of-concept that computer vision algorithms can now be embedded on regular Canon digicams with little effort (CHDK is coded in C). What other popular vision algorithms could be implemented? For what purpose?" You can get some idea about ViBe from this abstract at IEEE; basically, it allows background extraction in moving images.

Zoneminder

[b0bby]

I've been playing around with Zoneminder a bit & this could be a way to use decentralized cheap cameras to send events to a ZM server. Pretty neat.

Re: CHDK

[0100010001010011]

What's the 'newest' Camera that supports CHDK? Mine got stolen recently and I'd like something to replace it for times I don't want to lug around my SLR.

Does CDHK support the DIGIC IV (720p, better face recognition). I played around with one of my friends and face recognition, the fact that you can zoom in and scroll through all faces to make sure they're in focus. Lots of cool stuff, but CDHK seems to mostly support older stuff and Canon's numbering scheme sucks.

Interesting...

[fuzzyfuzzyfungus]

What I find interesting about this is not so much that the code can be loaded(since the CHDK project already did that job, and has had it working for some time now); but that consumer digicams would have enough general purpose punch to run anything much more than trivial scripts that more or less emulate series of button presses(which can be extremely useful, for time lapse, auto bracketing, etc, etc.).

Given the sheer number that are produced, and the fairly tight battery life constraints, I would have assumed that most of the heavy lifting(crunching raw sensor data to .jpeg form, or encoding video) would be done with largely fixed function hardware, with just a little bit of general purpose computer slapped on to handle UI, user input, and tweak the settings of the encoder units. Apparently, the general purpose units have more punch than I thought.

Why Bother Rewriting the Wheel?

[eldavojohn]

OpenCV [sourceforge.net] has C interfaces [willowgarage.com] and there are more that have some C code libraries [cmu.edu]. Really the coding challenge would be building the wrappers to utilize those libraries with your camera's hardware (I assume provided through CHDK APIs). My vote is for a nifty KLT implementation that allows me to take a video and extract a huge wide pan image in post processing on the camera.

Re: CHDK

[TubeSteak]

I played around with one of my friends and face recognition, the fact that you can zoom in and scroll through all faces to make sure they're in focus.

Maybe someone can explain the multi-focus thing to me.
Does the camera pick out all the faces and then pick an average focal point?
Or does the software (as opposed to the lens) actually tweak each face into focus?

Re:Not alternative firmware

[Anonymous Coward]

CHDK is an add-on to the existing firmware, that works by piggibacking on its OS

If it would have been an Apple camera, ten thousand lawyers would have been busy suing the developers.

CHDK made me buy a Canon, and Canons approach of "live and let live" will make my next purchase a Canon 50D.

Thank you Canon!