Abstract 

A method for a real-time vision system that automatically detects a user's blinks and accurately measures their durations is introduced.  The system is intended to provide an alternate input modality to allow people with severe disabilities to access a computer.  Voluntary long blinks trigger mouse clicks, while involuntary short blinks are ignored.  The system enables communication using "blink patterns:" sequences of long and short blinks which are interpreted as semiotic messages.  The location of the eyes is determined automatically through the motion of the user's initial blinks.  Subsequently, the eye is tracked by correlation across time, and appearance changes are automatically analyzed in order to classify the eye as either open or closed at each frame.  No manual initialization, special lighting, or prior face detection is required.  The system has been tested with interactive games and a spelling program.  Results demonstrate overall detection accuracy of 95.6% and an average rate of 28 frames per second. 

CVPR 2001 Paper (5.03MB) 

The following are short sample outputs of the system described in this work.  A red dot appearing over the eye denotes that a short blink was detected, while a blue dot denotes that a long blink was detected.  The initial blue rectangles surrounding the eyes and face indicate the results of the motion analysis phase.  The following light blue rectangle around one eye indicates the tracker's search space, and the inner purple rectangle represents the best match for the open eye template. 
  
  
Demo 1 : Shows long / short blinks and reinitialization of tracker 

Demo 2 : Shows user with glasses 

Demo 3 : Includes some quick head motion 

Demo 4 : Two heads compete for tracker 

Demo 5 : Includes some background motion distractions