Guard Dog Robot Project

Well I stumbled upon this blog

http://www.redcode.nl/blog/2010/06/creating-shazam-in-java/

( whose author stumbled upon this blog )

http://laplacian.wordpress.com/2009/01/10/how-shazam-works/

and I guess audio is really one of the areas of multimedia computing I have not really explored much all those years..

Although I don`t like the language of implementation ( I mean Java ) of Roy van Rijn , a music/sound classifier should come in handy with the guard dog project.
I guess , with the correct microphones , it could be trained to detect from dogs barking to birds singing or rain , etc. This kind of basic functionallity ( not human speech recognition ) , but ambient waveform recognition would make a robot much more “human” because we have hardcoded built in the ability to distinguish sounds..

Also it is truly disgusting that the copyright holder for the Shazaam algorithm mailed a “threatening” message ( http://www.redcode.nl/blog/2010/07/patent-infringement/ and http://www.redcode.nl/blog/2010/11/patent-infrigement-part-2/ ) ..

Well I guess everything is explained here -> http://patentabsurdity.com/

I am now re-implementing feature detection , using second order intensity changes and writing the feature tracking algorithm.

The point correspondences will then be used with the Eight Point Algorithm

The system of correspondances will be then solved using the fantastic GPL licensed GSL library and from that result I will get the camera pose estimation !

After I know the camera pose estimation , have the disparity depth maps from the stereo vision they will be combined to start to make a mapping of the world around guard dog!

Then ( I think ) the project will be finally ready ( as for the guard functionality ) and from there on I will have to start implementing haar/eigenface face/object/pattern detection , the KnowledgeBase stack ( for A.I. ) , sound recognition etc..

It is kind of sad seeing augmented reality being used for such a petty cause ( for making people buy clothes and toys ) on the other hand I guess there is nothing wrong with applications like these , after all if this becomes a trend in stores it will ultimately boost interest in machine vision areas , and thats a good thing..

http://ewokrampage.wordpress.com/
http://www.robots.ox.ac.uk/~gk/

Kinect has a much better result for close ( 0.5m - 3m ) , low light distances