logitech facial features tracking

asked 2014-07-08 07:49:11 -0500

for my application i want to track the facial features. i have tried some methods but none of them provided the required robustness , i have seen the Logitech avatars and their facial feature tracking was accurate and robust.any ideas how they are doing ?. it will be helpful....

edit retag flag offensive close merge delete



you could at least mention, what you tried before...

berak gravatar imageberak ( 2014-07-08 07:56:16 -0500 )edit

Have you seen this book (Chapter 6)?

thdrksdfthmn gravatar imagethdrksdfthmn ( 2014-07-08 08:06:14 -0500 )edit

@berak the first approach 1. haar based face detection , 2.canny edge detection , 3.contours finding and key point detection in this approach the points changes drastically . second approach is using flandmarks but 7 points doesn't fulfill the requirement

Dinesh_Kannan gravatar imageDinesh_Kannan ( 2014-07-08 09:02:58 -0500 )edit

@thdrksdfthmn is any other way to reduce the landmark points in muct database.because i don't want 76 landmark points.

Dinesh_Kannan gravatar imageDinesh_Kannan ( 2014-07-08 22:35:36 -0500 )edit

Yes, when you train the detector, you specify only X points (landmarks). Look a little in the code. I think you should train your own detector

thdrksdfthmn gravatar imagethdrksdfthmn ( 2014-07-09 02:24:17 -0500 )edit

When I have tested it I have got about 10 ms for computer (Win 7 ... I've forgot the other specifications)

thdrksdfthmn gravatar imagethdrksdfthmn ( 2014-07-09 02:25:41 -0500 )edit

Not to disappoint you but Logitech is a company that makes profit from safeguarding their algorithms and keeping them all to themselves. So it would only be guessing how they actually do it, probably based on a combination of existing algorithms which were then patented as a combination.

Stop trying to create sortlike behaviour and just look at existing papers who published usefull work on this like here

StevenPuttemans gravatar imageStevenPuttemans ( 2014-07-09 07:12:47 -0500 )edit

@thdrksdfthmn thanks !. i have seen the book , obtained the landmark points of eyebrows,eyes,nose,mouth with the help of asm (totally 24 landmarks ) using these points applied 3d reconstruction on a face model but rotation is not working , all other transformations are working fine. for head pose estimation i have referred this[http://www.morethantechnical.com/2010/03/19/quick-and-easy-head-pose-estimation-with-opencv-w-code/]... any idea ?

Dinesh_Kannan gravatar imageDinesh_Kannan ( 2014-07-10 00:29:17 -0500 )edit

In fact, the rotation is done by the tracking. If it's lost and rotated, the new init is going to be when a face is found (not rotated).

thdrksdfthmn gravatar imagethdrksdfthmn ( 2014-07-10 06:08:40 -0500 )edit