Eyetracking w/ low-cost webcams - Six test subjects (Demo)

This video shows eye tracking performance for six of our test subjects. Note that this is raw frame-by-frame estimate without Kalman filtering or any other multi-frame averaging method that could be used to improve the accuracy further (MSE versus target less than 0.032 for each test subject (1.0=screen height/width)). Frames without found iris or with less than four facepoints were removed. For calibration, we used just a linear homography from eye gaze space to display space (because the used cameras were not fixed relative to the screen) and no higher-order polynomial model. Runs faster than realtime on most current machines (except some subnotebooks). At present this seems to be the most accurate eyetracking using only USB cameras (no IR glints, no extra hardware) on YouTube.

Category:

Science & Technology

Tags:

• low-cost
• eye tracking
• usb webcam
• test subjects