FaceVest Face Data Capture Platform
FaceVest Face Data Capture Platform
The FaceVest is a hardware platform for capturing multiple video streams of annotated face data to digital video tape. It features two different head mounted cameras for capture of video.
The vest's head mount display overlays a video stream from a head mounted camera onto the wearer's left eye. Calibration marks are inserted into the video display stream by a character overlay generator. These marks are manually aligned with subject's eyes in the display stream by the wearer.
Alignment of the marks with the eyes of a subject places the face in a roughly known configuration for easy extraction at a later time. Once the marks are aligned, the wearer presses a button to mark up the audio track of the video stream with a pure sine wave tone. These easily segmented audio bookmarks will later signal a human-recognized presence of a calibrated face in the data stream. Note that the overlay output is not fed into the capture device; all video is split with a distribution amplifier. This will give us a decent estimate of ground truth for comparison of automatic methods. Having a human help with segmentation may be appropriate (and efficient) for some applications. Furthermore, multiple cameras are easy to add. Each subsequent view, though not directly calibrated, is relatively calibrated (the cameras are mechanically coupled through the helmet).
Components used in the vest:
Links To More Pictures:
Why near
field Infrared(IR) is safe to use in our application/design, and how
it can be dangerous.
