Analysis of vehicle surroundings and driver status from video stream based on a single PAL camera

Driver distraction and fatigue are important factors that cause accidents. We developed a novel machine vision system which can provide the driver's face pose and eye status information as well as the driver's viewing scene. The driving original video information was obtained simultaneously by a panoramic annular lens (PAL) camera in the system. The PAL camera can capture its surroundings with a field view of 360-degrees horizontally and -10~+60 degrees vertically without any scanning parts. The PAL camera was properly configured on the vehicle to not only capture the scene in front of the vehicle but also the driver status information including face and eye. A rapid face detection algorithm was proposed based on Bayes classification of the skin color. Once the face detection had been completed, template projection was used to detect eyes in a narrow area of the face. Besides, a cascade yielded from training was used to detect the vehicle object and a novel object localizing algorithm was proposed which is insensitive to the shake of the vehicle. Our contribution was that it was possible to research the vehicle and driver information at the same time with a single panoramic camera. Meanwhile the system was very simple.