none
Multiple Face Detection/Tracking using Face SDK

    Question

  • Currently, I am trying to develop a Windows Phone application that displays a speech bubble above a person's head once a face is detected. I was able to do it with your SDK. However, I am being asked to implement multiple face detection and up to now, I am not able to do it.

    Using your SDK, can you please help me or give some advises on how I can tweak the code to support multiple face detection and tracking? What we want to achieve is that our app can track many faces and show what each person is thinking (by displaying a text inside a speech bubble) - it also practices augmented reality.

    I have actually found another API as an alternative. It works for the multiple face detection but it works much slower as compared to yours.

    Please advise.

    Thank you very much.

    Monday, October 8, 2012 4:24 AM

Answers

  • The tracker shipped in the SDK is targeted for single face tracking only. You have to implement your own multi-tracking algorithm in this case, starting with face detection.

    Face detection for each frame is unacceptable for performance consideration. If in your scenario, the person are always looking at the camera (frontal face), you can try the frontal face detector instead of the multi-view version. Then for each detected face, you can apply some tracking algorithm (either feature based like KLT or color based like camshift). You may then need to periodically re-detect faces to correct the tracking result after a while.

    Anyway, real-time multi-face tracking is hard on phone (also managed code is slower). You would do lots of optimization work and design clever algorithms.

    Thanks.

    Thursday, October 11, 2012 1:16 AM
    Owner

All replies

  • The tracker shipped in the SDK is targeted for single face tracking only. You have to implement your own multi-tracking algorithm in this case, starting with face detection.

    Face detection for each frame is unacceptable for performance consideration. If in your scenario, the person are always looking at the camera (frontal face), you can try the frontal face detector instead of the multi-view version. Then for each detected face, you can apply some tracking algorithm (either feature based like KLT or color based like camshift). You may then need to periodically re-detect faces to correct the tracking result after a while.

    Anyway, real-time multi-face tracking is hard on phone (also managed code is slower). You would do lots of optimization work and design clever algorithms.

    Thanks.

    Thursday, October 11, 2012 1:16 AM
    Owner
  • Could you please tell how could you detect faces using Microsoft fsce sdk ?

    please provide me the other API name too

    Monday, February 17, 2014 10:13 AM