Jack000's profile - activity

say I have some frames of a CG animation and wish to apply motion blur as a post-processing effect (simulated by this waterfall)

I could use dense optical flow to generate a vector field, which might look like this:

now I'd like to blur the original image in the direction of the vector field for the motion blur effect. What's the best way to do this?

my first thought is to apply a convolution filter for motion blur, but the kernel will have to change based on the value of the vector field. I'm hoping there's some function in opencv that does this that I'm ignorant of, or I'll have to implement the convolution myself in a loop?

yeah that's what I'm doing now. I thought Haar -> CAMShift -> Kalman would have better performance and/or be more accurate, as suggested by this: https://code.google.com/p/camshift-ka... but maybe CAMShift is just not the right way to solve this particular problem.

I'm just getting started with OpenCV and for my first project I'm trying to do some face detection and tracking. The use case is pretty simple:

• find one face in my webcam feed with Haar Cascade
• track the center point of that face with CAMShift/meanshift

the face detection part works fine, but I find that CAMShift and meanshift tends to extend to the neck region, lowering my center point. Is there any way to tell CAMShift that I'm only interested in faces? Is this the wrong approach for what I'm trying to do - what is the best way to track the face, and only the face after the detection step?