@mikesapi I see. Thanks for the quick reply. I'm doing something similar myself in college.
So you're using template matching? How does it perform with scaling? E.g. you initialise the templates of the eyes at a certain distance from the camera and then during the tracking you move away from the camera?
@kosomonova This kind of template matching worked sufficiently well for this application, in which the user is seated, facing a webcam and does not 'scale' much from the initialized position. If you need robustness to scale, you may search for templates at various scales and transformations. Another option is to allow the system to reinitialize automatically/dynamically update the templates..
Haar feature detection is used for automatic initialisation, after which the detected face features are tracked using NSSD.
If you send me your email address, I will let you know when the source is released, thanks.
mikesapi 11 months ago
@mikesapi I see. Thanks for the quick reply. I'm doing something similar myself in college.
So you're using template matching? How does it perform with scaling? E.g. you initialise the templates of the eyes at a certain distance from the camera and then during the tracking you move away from the camera?
Email address PMed.
kosomonova 11 months ago
@kosomonova This kind of template matching worked sufficiently well for this application, in which the user is seated, facing a webcam and does not 'scale' much from the initialized position. If you need robustness to scale, you may search for templates at various scales and transformations. Another option is to allow the system to reinitialize automatically/dynamically update the templates..
mikesapi 11 months ago
Hi,
Just wondering how you're detecting the features here? Are you running a Haar detector on each frame?
Also, is there source code or an executable for this available anywhere?
kosomonova 11 months ago
Great job!
owenmalta 2 years ago