OpenCV + OpenGL + Immersive environment enhanced

[Djcytral]

nice!

[diem389]

really cool!!!

[HectorUch123]

Hello !

Your help was really usefull , i did that and used threads , now i posted a video showing de result, that video is also a answer for your one , i hope you like it.

Thanks once again.

[apanap556]

I tried something similar once, but instead of detecting the face on every cycle, I detected where the face was, found some features on the face and tracked them with Lucas Kanade (based on the lk example that comes with opencv), which is faster. It sometimes lost track of the face though (which I guess could be fixed by doing proper face detection every once in a while).

[apanap556]

Ah, I see it's already been suggested a year ago.. :) Anyway, nice demo!

[dannyxyz22]

Thank you,

this is the idea I used in the google code ehci project, but it uses POSIT to get the 6 degrees of freedom head tracking.

It needs a function to re-initialize near some known positions.

Thanks for your feedback,

Daniel

[HectorUch123]

Hello friend!

I tryied to do that with threads and your hint together , it really had speedup the application. Thank for the help!

I forget to comment the project: Really great!

Best Regards,

Heitor.

[dannyxyz22]

Well done!

Thank you and congratulations :)

[]'s

[HectorUch123]

oh thanks! Will this really improve speed? I put the face tracking code inside the opengl display function(that i also use on de idle func) . This right? or i will have to open in threads? thanks

[dannyxyz22]

This will improve detection speed which is generally the bottleneck, though you'll need to try and see if it's your case.

Kind regards,

Daniel

[bluesrunthegame]

Instead of face recognition can you just recognize a mass of pixels which change rapidly? That would define the head once the person moved. The dimensions of the blob could give a rough approximation of distance from the monitor, and do so without extensive feature extraction. Or maybe initial recognition could be performed with face detection and then tracking with the previous method, with intermittent face checks to verify the target?

[dannyxyz22]

Your idea of initia detection and then tracking is good. That's actually what's done in order to reduce computational costs, although it's not very robust due to error accumulation.

[dajoebomb]

This could be amaazing for games in the future!

[digitalmorphine]

Can't you track just the positions of both eyes to estimate the distance from the screen? I guess that's what Johnny Lee does, although with LEDs and an IR camera instead.

[dannyxyz22]

Yes, can you send me your e-mail?

I'll try to make it public...

Best regards,

Daniel

[HectorUch123]

Man, i am trying to do the same ,but its to slow, how didi u make it fast? Can you post the code?

[dannyxyz22]

Hi Hector,

try changing the scale parameter for the opencv haarcascade function cvHaarDetectObjects. Try maybe 2.0 or above and you'll notice some speedup. Scale is the fourth parameter and defaults to 1.1.

I hope it works,

[]'s

[temposer]

could you possibly post the app?

[gelmo81]

looks interesting =) good job.

[dannyxyz22]

Thank you :)

By the way, I enjoyed daydream very much!!!

Congrats!