Car accidents happen in the blink of an eye, and the more that eyes blink or wander, the greater the chance for an accident. According to carmaker AB Volvo of Göteborg, Sweden, driver fatigue and inattention are responsible for up to 56 percent of all accidents. But a machine vision system that watches drivers could help to ensure that they are watching the road. A machine vision system that monitors drivers could prevent accidents caused by fatigue or inattention. Called faceLab, the system identifies key facial features and determines the blinking rate and the gaze direction, indicators of a driver's attentiveness to the road. Researchers at Seeing Machines used techniques developed at Australian National University in Canberra to create faceLab, which monitors where a person is looking and how often she or he blinks. "What we've built is a sensor which is useful for traffic safety applications," said Alex Zelinsky, a professor of systems engineering at the university and chief executive officer of Seeing Machines. "We call it a human performance measurement tool. So, in other words, understanding how people work in a nonobtrusive way." FaceLab, which was developed with funding from Volvo, tracks drivers with two inexpensive commercial cameras. Algorithms running on a standard computer identify 32 facial features, including the eyes and their irises, and update their positions 60 times per second. From these points, faceLab constructs a three-dimensional model of the person's face and monitors indicators of driver attention such as gaze direction and blink rate. If the driver seems to be fading, the system triggers a warning. High accuracy According to Zelinsky, faceLab successfully tracks the head or face of a driver more than 95 percent of the time. At night, the system uses infrared illumination, and if facial features are obscured, such as by turning the head, it dynamically substitutes new estimated features. If the driver is wearing sunglasses, the system cannot measure blink rate because the eyes are hidden. However, it still can determine where the driver is looking by following the position of the nose. The company plans to develop an embedded version of faceLab for automotive applications, shrinking both system size and cost. Several car manufacturers, including Volvo, have expressed interest, but no commercial products or timetables have been announced. Seeing Machines also markets faceLab for other applications, such as controlling the game play in video games.