none
Issues with Face Tracking (using the C++ API) and Near Mode RRS feed

  • Question


  • Is it just me or is anyone else having problems with face tracking in near-mode (using Kinect for windows of course but just to be clear)

    These have been my observations (even in the sample apps).

    1) Face tracking in default distance mode works very well. Detection is quick. Tracking is fairly robust and it is continuous and consistent.

    2) In near-mode, detection does not work very well. Detection takes a long time, if at all, and the difference is clear between detection in near mode and detection in normal mode. I find that I have to go to "normal" distance to be detected and then go near to be in the near region with tracking still on (if it doesn't fail while moving). So I have to move back and forth to be detected. However, in the WPF samples, detection and tracking in near mode seems to work just fine.

    3) In addition to issues with detection  there seems to be a divider (~0.8m) between the near and normal regions where face tracking stops or goes awry (i.e. tracking will continuously turn on and off.) Again, this problem is observed in SingleFace.exe (c++) but not in the WPF samples.

    Can anyone help me understand why this is happening and what I can do to fix it? My app uses the C++ API and I experience the same things as with SingleFace.exe.

    Thanks!

    Friday, November 9, 2012 9:39 PM

Answers

  • I figured this out, btw. The problem was that IFTFaceTracker.ContinueTracking was being given a NULL headpoints value and consequently, IFTResult.GetStatus would give a failed status (not S_OK). It was NULL because the function that provides it (GetClosestHint in the SingleFace app) uses the skeleton data but seated skeleton mode is turned off by default! (in SingleFace)

    So the solution is to turn on seated skeleton mode.

    :)



    Wednesday, November 14, 2012 8:43 AM