Study On All-day Monitoring Driver Fatigue And Distraction State Based On Visual

With the rapid increase of vehicle conservation and the enhancement of highway level, traffic accidents occurred more and more frequently. Many highway accidents are closely related to the driver's fatigue and loose alertness. Therefore, all-day monitoring research of the driver's fatigue and distraction state has the significant significance.This paper used the system which not only can monitoring driver fatigue and distraction in the daytime, but also can monitoring driver state at night. When light insufficient at night, we have used the near-infrared assistance illuminator, it had not any disturbance to driver eyes. The system can monitoring real-time driver face characteristic based on the machine vision method in all-day. The system should be warning when the driver occur fatigue and distraction state. It can provide the technical support for the reduction traffic accident occurrence. For the research of all-day monitoring driver fatigue and distraction state, this paper mainly contains following content.1. Driver face image division is the basic of driver face characteristic extracting and driver eye region locating. According to the infrared spectrum and driver infrared face image characteristic at night, the infrared image pretreatment used by the median filter algorithm. This paper detected the driver face region by Otsu segmentation algorithm, and located the driver face region by projection.2. Driver eye region accurate location is an important basis for the driver fatigue studies based on vision. This paper located the driver eye region by the ellipse fitting method, and used the Kalman filtering method real-time tracking driver's face region and the eye region.With the help of driver face region localization and the geometry position relations of face apparatus. This paper established the possible eye AOI (AOI-Area of Interest) by the upper part in an infrared image. In the AOI, the binary image used the morphology filter method. Then the image marked by using the region labeling algorithm. The driver eye region located through the constraint conditions eliminating disturbance. Driver eye edge detection used the outline track algorithm. The eye's edge points fitting by ellipse fitting method. Then the driver eye regions locate accurately. The driver fatigue and distraction system was enhanced quickly by the fast accurate tracking method. This paper used the Kalman filtering method real-time tracking driver's face and the eye regions at night. The experiment has confirmed the Kalman filtering tracking method efficiently and accurately.3. According to driver eye pupil reflexe characteristic with the near-infrared assistance illumination at night. After driver eye region located, the pupil and Purkinje spot position detected accurately. It can provide the theory foundation for the line of sight direction estimation. The line of sight direction estimate not only monitoring driver's fatigue state, but also detecting driver's distraction state.In order to detecting the pupil position, by using of two-dimensional histogram segmentation method divided up the eye region. Eye pupil edges extractiond through Canny edge examination algorithm. The pupil position located by Hough transformation method and ellipse fitting pupil edge points method. The paper located Purkinje spot position based on Harris corner detection algorithm. By Purkinje spot method theory, line of sight direction estimation based on the ellipse fitting centroid algorithm and based on Harris corner detection algorithm. At last this paper has analysed precision of driver'sight direction.4. The infrared assistance illumination can automatic shut-off when driver driving in daytime. The daytime characteristic of driver face image extracted by partial SMQT (Successive Mean Quantification Transformation) method. The daytime driver face located based on SNoW (Sparse Network of Winnows) method.The daylight disproportion brings some difficulties to the daytime image division. The daytime driver image has strengthened by cone-shape stretching algorithm. Then the daytime eye regions divided by use of the maximum-entropy segmentation method. Finally with the help of the eye constraint condition, the daytime driver eye regions location accurately. Similarly, this paper used the Kalman filtering method real-time tracking driver's face and eye regions in daytime.5. The driver fatigue state monitoring by using the Bayesian network fusion method at night, and driver distraction state detected based on the line of sight direction estimation.First of all, the paper separately have used the PERCLOS method, based on the GAZEDIS line of sight distribution method, based on the PERSAC method and based on Purkinje spot method for monitoring the driver fatigue state at night. Then the driver fatigue state judged by the Bayesian network fusion algorithm with four fatigue methods at night. Driver distraction state has detectd by the line of sight direction estimation and the line of sight distribution condition. Moreover the driver daytime fatigue state monitoring also maked use of the PERCLOS method. The daytime driver distraction detected by the face sway angle algorithm.In summary, many systematic and scientific researches have carried on in this thesis, which are the key technologies in Vision all-day driver fatigue and distraction state monitoring. The achievements will be provide the theory and the technical support in the SAD(safety assist driving) field, and acquire the obvious society and economic efficiency.