Observing and Classifying the Activity of a Vehicle Driver
The goal of this project is to classify the focus of attention of a vehicle driver, using images from a camera mounted inside the vehicle. An obvious use of this information is for drowsy driver detection (a recent survey suggested that 3% of highway crashes or 100,000 crashes per year in the US are due to tired drivers). But this is only one of the ways to utilise data about the driver's activity. The information can be used in conjunction with readings from external sensors on the vehicle to alert the driver to possible collisions if his or her attention seems to be elsewhere. On a more sophisticated level, it may be possible to learn a driver's characteristic habits, and thereby to anticipate the execution of certain maneuvers. This could be used, for example, to initiate automatic changes to the suspension of a vehicle to facilitate a turn, when observations have indicated that the driver is about to execute that maneuver.
Background & Objective: Today's vehicles have increasing amounts of on-board computer power. Utilising part of that power to aid the driver - to make driving safer and easier - raises a problem however. The driver is already skilled at the task of maneuvering through traffic and crowded urban environments, so a visual or auditory warning to indicate every possible collision would be superfluous and annoying. What is needed rather is the ability to interact with the driver in an intelligent way, offering information only when it is appropriate. This requires observation of the driver, and the ability to understand where his or her attention is being directed. Similar techniques may be of use for other applications such as observation of computer monitor users, enabling hospital patients or those with physical disabilities to use head and eye motions to communicate, and to identify the patterns of attention of consumers in retail stores.
Technical Discussion: The algorithms used were developed to cope with the constraints in a vehicle driver application: Once head pose has been determined, processing is focussed on the subject's eyes. Prior models of skin colour are used to identify and segment the eye. The shape of the segmented area is used to determine if the subject is looking forwards through the vehicle windscreen, or downwards at the dashboard.
Technology Areas:
Computer Vision
Graphics
Modification Date: June 13, 2008