none
Neutral Face in Kinect HD Face Tracking RRS feed

  • Question

  • Hi!

    I am currently working on my thesis using HD Face Tracking of Kinect for Windows v2. I am using the Animation Units from HD Face as features of my data. I need to find out if these AUs are not sufficient to represent my data properly and if I should go back to extracting the points and computing the features. I would like to know how the neutral face is calculated. From a tutorial with Ben Lower, I have learned that the average shape of human face is deformed to match the face of the user. But I didn't get how the neutral face was calculated. From what point to what point of the 3d video is averaged to get the neutral face? Since it is getting the neutral face in real time, does it use the first frame up to something like 'current-few' frames? Are Eigen vectors used? If this is not how it works, can you explain how?

    Thanks!

    Wednesday, March 22, 2017 1:19 PM

Answers

  • http://www.sciencedirect.com/science/article/pii/S0262885614001310

    I believe this is the paper that introduces what they use in HDFace.

    https://www.youtube.com/watch?v=8_yMPLSZSEs

    Something along the lines of Active Appearance Models are used on a training set of faces to get a mean face shape. The live user then faces the kinect and the 2D AAM works and is constrained by the 3D data.

    • Marked as answer by pauriver Wednesday, March 29, 2017 8:54 AM
    • Unmarked as answer by pauriver Wednesday, March 29, 2017 8:58 AM
    • Marked as answer by pauriver Wednesday, March 29, 2017 9:24 AM
    Wednesday, March 22, 2017 1:59 PM

All replies

  • http://www.sciencedirect.com/science/article/pii/S0262885614001310

    I believe this is the paper that introduces what they use in HDFace.

    https://www.youtube.com/watch?v=8_yMPLSZSEs

    Something along the lines of Active Appearance Models are used on a training set of faces to get a mean face shape. The live user then faces the kinect and the 2D AAM works and is constrained by the 3D data.

    • Marked as answer by pauriver Wednesday, March 29, 2017 8:54 AM
    • Unmarked as answer by pauriver Wednesday, March 29, 2017 8:58 AM
    • Marked as answer by pauriver Wednesday, March 29, 2017 9:24 AM
    Wednesday, March 22, 2017 1:59 PM
  • Hi!

    Thank you for this! This is exactly what I've been looking for.

    Cheers!

    Wednesday, March 29, 2017 11:07 AM