Active Wearable Vision Sensor —Detecting Person’s Blink Points and Estimating Human Motion Trajectory—

To realize versatile real-time man-machine interactions based on understanding human intention and activities, we develop an active wearable vision sensor. The sensor consists of the detector of person’s viewing lines and two active cameras. First, we establish a method for calibrating the sensor so that it can detect person’s blink points accurately even in a real situation such that the depth of blink points changes. We formulate errors of the viewingline detector in terms of the depth in blink points from the person and employ the stereo algorithm to correct the errors. Secondly, we propose a method, the binocular independent fixation control, for incrementally estimating the motion trajectory of the person wearing the sensor. In this method, while the person moves we control the two active cameras independently so that each fixates its optical axis to its own fixation point.