Suggestion to improve head tacking accuracy

[Henry Nash]

Hi I tried using the head tracker in your EyeX 4C. I noticed the head tracking isn’t exactly 1:1 when I map it to my view. It can wobble and is slightly tilted as I move my head back and forth. I have a suggestion to improve the accuracy of the EyeX head tracking. I think it has potential to match the TrackIR, and the user would not need to wear any hardware.

Enable Viacam http://eviacam.crea-si.com/ has a very smooth head tracking algorithm that works using a webcam. The motion has no wobble or tilt. I’m not sure how many features EyeX uses, but this one uses multiple that adjust as the user rotates his head.

He describes the algorithm here: https://sourceforge.net/p/eviacam/discussion/898005/thread/a52e22c7/

Once the face is located, the motion inside of the user’s face bounding area is estimated using the Horn and Schunck optical flow algorithm (using OpenCV). As a result, two velocity fields (one for each axis) of the same resolution as the input images are obtained. Both fields are then downsampled to a 15×15 matrix and modulus is computed for each cell of such matrix. To avoid static areas (i.e. background) only these cells with enough motion (i.e. whose modulus is above a threshold) are picked up and averaged to compute the resulting motion vector.

That’s a mouthful, but the algorithm is not too hard to implement and there is example code available.
https://github.com/cmauri/eviacam/blob/master/src/visionpipeline.cpp#L319

I’d appreciate it if you passed this suggestion to one of your product managers or engineers!

[Grant [Tobii]]

Hi @nash8350, thank you for input and keen observations. I will be sure to pass along your suggestion to our algorithm team for consideration! Best Wishes, the Tobii Tech Team.

[Author]

Posts