Optimal Environment for Face Tracking RRS feed

  • Question

  • I am currently able to capture both 3D and 2D data from the Kinect for Windows device. The tracks generally well enough. I am trying to get it to be as accurate as possible over the features of the face that I need to use (eyes, lips, nose, outline). For example I know that the lips track symmetrical open and closed with some additional events for emotions (smille, frown etc). I am looking for ways to improve the accuracy of the point data (either 3D or 2D). Are there code or environmental factors that I can use to improve the data? I would also be open to trying multiple Kinects or passing additional depth information to Face Tracking SDK.

    Currently using C# library with depth resolution at 320x240 and color resolution at 1280x960 in near mode. Also, I have Kinect placed above monitor with an array of standard incandescent bulbs around it. I can dim the bulbs, but have tried up to their maximum of 750W. Computer is an MSI GT70 with i7 processor. 

    Tuesday, January 8, 2013 4:22 PM