none
HDFaceBasics-WPF - Looking for Consistent Values from repeated captures of the same face RRS feed

  • Question

  • Hi there. I have been playing around with the HD Face Basics code from the SDK and trying to understand how everything works.  What I am currently looking for are values that would stay consistent between repeated captures of the same face.  That is proving difficult and was looking for some expert guidance on the subject.

    I am aware that of the vertices that come from the CalculateVerticesFromAlignment but they are simply points in Camera Space and aren't exactly the same even when capturing the same face in the same orientation (Pitch, Yaw, Roll) at the same distance.  More specifically with the vertices, I attempted to calculate the 3D vector length between vertex points given (with the capture done at similar orientation and distance) but the vector lengths were still not as similar as I would have hoped and were not different enough between different faces. Additionally, I have become aware of the 94 SU's from the FaceShapeDeformation enum and the 17 AU's from the FaceShapeAnimations enum but those don't appear to be exactly the same between different captures either.

    My next assumption was that there is a way to utilize the SU's and AU's in combination with the vertices (ex: the deformation SU's are to be multiplied by the vertices to obtain a new value that would be consistent) but I cannot find any documentation on how the SU's or AU's are actually applied to the mesh created.

    I did find a previous thread from 2014 in which Carmine Sirignano had requested the information from others at Microsoft but I don't see that any was ever obtained (https://social.msdn.microsoft.com/Forums/en-US/a8426e1c-cb53-429e-831a-ab6f390675c2/where-can-i-get-the-definition-of-the-mesh-in-ifacemodel).

    Does anyone have any information that might be relevant to what I'm looking for? In the end, I would like to be able to save data from a face capture and call it back in during another capture to compare to see if the face/person is the same.  I believe I can do most of this but just need to find the relevant data that will stay consistent.

    Evan Silverstein



    Tuesday, December 1, 2015 8:13 PM