 |
Face tracking
Ulli from Germany [6 posts] |
14 year
|
Hello,
i'm trying desperately tracking a face with RoboRealm.
try with color filters (red) to date are very bad. It would surely be better, the eyes and the mouth to recognize and realize then face persecution. Unfortunately I have no idea how am I supposed to make me RR. anyone can help me or provide a demo program?
Regards,
Ulli
|
|
|
|
Ulli from Germany [6 posts] |
14 year
|
|
No one has an idea how one can realize a face tracking with RR?
|
|
|
|
Samuel Rosenstein from United States [30 posts] |
14 year
|
I think that you could use the skin filter. You could then track each eye (now black circles) using the blob filter. If you didn't care so much about the perfection then you can just use the skin filter with COG. I haven't tried it, but it might work.
Sam R.
|
|
|
|
Ulli from Germany [6 posts] |
14 year
|
Hello Sam,
many thanx for request.
I test it with Skin and Blob Filter, the result is not so goog.
The Picture is only with Skin Filter and value 3.
Hm... what can i do?
Best regards,
Ulli
|
|
|
|
Samuel Rosenstein from United States [30 posts] |
14 year
|
Ok. Maybe you could use shape matching, but I'm not sure if that would work. If you don't care about putting a fiducial on your head that would work really well. You could find rotation on the x and y axis and also the rotation of your head.
Sam R.
|
|
 |
from United States [214 posts] |
14 year
|
Hi Ulli,
Could you tell us a little more about your requirements? For example, will you always be sitting in the position shown in the image you included? If so, try Magenta in the RGB filter rather than Red. There is too much red in the wood behind you (which also gets picked up by the Skin filter.) Also, do you need to get back the actual shape of your face or just a blob in the right location? If all you need is a blob, you can follow the Magenta filter with Segment Colors. Then Auto Threshold the result and use it as a mask over the original image to see your face. I have included a sample .robo file. Using the slider controls on the Segment Colors filter, you can expand or contract the mask around your face. Of course, this approach won't work under general conditions where your background or lighting changes.
Another approach would be to use the Movement filter. If you are the only one in the image, then the only moving pixels will tend to belong to your head and shoulders. If you set the Movement filter to be very sensitive you'll pick up even subtle movements. If there are moments where you are perfectly still, you can use the last known position. If you want to filter out shoulder pixels, use the Blob filter to focus on those pixels above the Y coordinate near the base of your neck. I've included a second .robo file that takes the movement approach (but I don't filter the shoulders).
--patrick
P.S. I like Sam's idea of using a Fiducial. Maybe put it on a hat. :-)
 program.robo
|
|
|
|
from United States [214 posts] |
14 year
|
Looks like my second .robo file did not upload. Here it is.
program.robo
|
|
|
|
Ulli from Germany [6 posts] |
14 year
|
Hi, this program working quite well and is a good approach, many thanks! I built a robot that will direct his head always on the person (X, Y position) standing in front of him. I have already made some experiments with motion detection, but unfortunately the results are not so great. ) The background (wall) may of course vary depending on the surroundings of the robot.
The picture shows the robot ;-)
Best regards,
Ulli
|
|
|
|
from United States [214 posts] |
14 year
|
Cool looking robot! Looks like you are set up for stereo vision. By the way, if you want to actually detect and track faces, have you tried OpenCV? There is a C++ version at:
http://opencv.willowgarage.com/wiki/
and a .NET version (called EmguCV) at:
http://www.emgu.com/
If I recall, OpenCV uses Haar classifiers to detect the structural properties of a typical face.
--patrick
|
|
|
|
Ulli from Germany [6 posts] |
14 year
|
Hello,
thank you for the Emgu link, is very interesting.
I have tried with the opencv face detection and I'm excited.
It would be nice if Steve would add a face detection with Tracking and based on Haar classifiers. There are enough examples to this on the Internet.
Pls Steven add a Face detection to Roborealm :-)
This is the only handicap for RR in the moment, the remains works very good!
Greetings from the snowy germany ;-)
Ulli
|
|
|
|
Ulli from Germany [6 posts] |
14 year
|
Hello Guys,
really has no idea how to make a good face detection and tracking with RR?
Steven wrote 3 years ago:
Hi,
Yes, that the problem with the skin color tone is that many walls have a similar hue. The best and easiest way to detect faces is to wait for us to release the face detection module which is still in the works and looks quite promising. Hopefully that will be out in a couple weeks.
If you haven't already subscribe to the what's new RSS feed and that will let you know when that module gets released. We typically release new modules on Mondays so you can also check back on the website periodically.
STeven.
Hm...
Greats Ulli
|
|
|
|
EDV [328 posts] |
12 year
|
|
|
This Thread
