OpenFace – a state-of-the art tool intended for facial landmark detection, head pose estimation, facial action unit recognition, and eye-gaze estimation.
I have a question about the head pose detection. I took the same video (approximately) with two camera positions and used it to estimate vertical head rotation (pose Rx).
First, I placed the camera on top of a computer screen facing vertically and second, I tilted the camera towards the face of the person (see the image below).
Then, I used FeatureExtraction to get the vertical head rotation for the two videos. I am getting different results from the two videos (about 15 degrees difference, which I guess is the angle of the camera tilt):
Do I understand it correctly that the head rotation is relative to the normal of the camera? If I want to have vertical head rotation relative to the world, I need to place the camera at the vertical position, right?
I have already collected videos for an experiment with the tilted camera position. Is there a way to correct for this and convert them to relative to the world?
Hi,
I have a question about the head pose detection. I took the same video (approximately) with two camera positions and used it to estimate vertical head rotation (pose Rx). First, I placed the camera on top of a computer screen facing vertically and second, I tilted the camera towards the face of the person (see the image below).
Then, I used FeatureExtraction to get the vertical head rotation for the two videos. I am getting different results from the two videos (about 15 degrees difference, which I guess is the angle of the camera tilt):
Do I understand it correctly that the head rotation is relative to the normal of the camera? If I want to have vertical head rotation relative to the world, I need to place the camera at the vertical position, right? I have already collected videos for an experiment with the tilted camera position. Is there a way to correct for this and convert them to relative to the world?
Thanks for the help!