Determination of driver dangerous states using smartphone camera-based measurements while driving

The paper considers the reference model for dangerous state recognition system based on the camera readings describing vehicle driver’s facial features. The model includes the developed schemes for determination of potentially unsafe driver’s behavior, observable by a number of visual cues, at each moment of vehicle movement. These schemes are focused on recognition of driver’s drowsiness and distraction, and tracking from camera video stream with aid of image processing methods and describing a set of facial features, including eyes, gaze direction, head pose, etc. To early recognize and classify the particular dangerous states of the driver the reference model for face features recognition is proposed, built upon the general schemes and characterizing visual driving behavior in certain situations that can potentially risk a driver. The evaluation of the proposed reference model was tested with smartphone-based prototype mobile application and showed preliminary results that shows performance and efficiency improvement in recognition of dangerous driving states while driving.