how do these people do it?

[ahack3r]

hey Trossen Robotics, I just joined in the hopes of learning more about computer vision robots. I want to know how autonomous motion tracking robots like http://zero-op.com/mercenary and
http://www.paintballsentry.com/ do it? I've read what they say on their sites, but how can someone else get at least a similar result with open source products?

After hours of research, I have narrowed my choices to these:

software possibilities:
http://code.google.com/p/weaponsgradekavelaars/
http://code.google.com/p/airsoft-sentry-gun/
http://opencv.willowgarage.com/wiki/Welcome

hardware possibilities:
http://www.trossenrobotics.com/p/des...ot-turret.aspx
http://www.trossenrobotics.com/p/pha...ot-turret.aspx
http://www.robotshop.com/arduino-exp...ion-kit-9.html

In your opinion, what are the best two choices for a fast, autonomous, open source, motion tracking device?

thank you for any help and information.

[animeunplugged]

you could look into roborealm

[ROBOTMAN]

Animeunplugged is right. Roborealm is a great way to track things and I believe a lot of people use it for sentry gun projects. I wrote a simple ball tracking program years ago here "its very poorly designed but demonstrates the concept" I also posted the old free version of roborealm here.

[defwheezer]

Roborealm does a good job at highly contrasting ball tracking (well lit red or green ball, like motion capture type), but not so much with face tracking, imo. I am also still looking for a good solution to face tracking or object tracking (blob).

[ROBOTMAN]

Yes it is difficult to track faces or differentiate say between a human and a wall but there is a "motion sensing" module that looks for changes in pixels that signify motion. If you used this to look for motion then position the turret in the general direction of that motion you could create a simple auto sentry. However there would be no distinction between a real target and the family dog .

[ahack3r]

my goal is open source so that I can later develop on it myself...Roborealm does not provide that.

[lnxfergy]

my goal is open source so that I can later develop on it myself...Roborealm does not provide that.

Honestly, RoboRealm is about 50x easier to get up and running with than OpenCV. Based on what I can gather about your experience level (from this and other threads), I would suggest at least prototyping your ideas in RoboRealm -- and then reimplementing in OpenCV if you desire to. OpenCV really is targeted at experienced programmers, and furthermore, at programmers who understand computer vision -- RoboRealm is much more approachable by beginners.

-Fergs

[animeunplugged]

Roborealm does a good job at highly contrasting ball tracking (well lit red or green ball, like motion capture type), but not so much with face tracking, imo. I am also still looking for a good solution to face tracking or object tracking (blob).

Face API works really well for face tracking
http://www.seeingmachines.com/product/faceapi/

[defwheezer]

Face API works really well for face tracking
http://www.seeingmachines.com/product/faceapi/

Wow- that really does work well for tracking head direction and tilt, and facial features like mouth and eye brows, and the non-commercial free license is sweet too.!

Pretty cool stuff, though without examples in C# (the current examples are in C++) I have no clear idea yet how to get data out to control, for instance, pan-tilt servos.

Any chance you have messed with this in C#?

[ahack3r]

Wow- that really does work well for tracking head direction and tilt, and facial features like mouth and eye brows, and the non-commercial free license is sweet too.!

Pretty cool stuff, though without examples in C# (the current examples are in C++) I have no clear idea yet how to get data out to control, for instance, pan-tilt servos.

Any chance you have messed with this in C#?

in the FAQ's, they say a C# version is coming soon...so I guess we will just have to wait.

Honestly, RoboRealm is about 50x easier to get up and running with than OpenCV. Based on what I can gather about your experience level (from this and other threads), I would suggest at least prototyping your ideas in RoboRealm -- and then reimplementing in OpenCV if you desire to. OpenCV really is targeted at experienced programmers, and furthermore, at programmers who understand computer vision -- RoboRealm is much more approachable by beginners.

-Fergs

well, you guessed right, I'm a noob and I agree, I think I'll just go with the Mosquito I/O based desktop turret first, and gradually build my own based off the Arbotix, my own platform and my own OpenCV based software.