Study On All-day Driver Fatigue Detection And Warning System Based On Facial Features

In recent years, with the rapid development of auto industry, the vehicle usage is increasing rapidly and the traffic accidents occurred more and more frequently. Many traffic accidents are closely related to the driver’s fatigue. Therefore, the study on detection and warning system of the driver’s fatigue has very wide application and is currently a hot topic for vehicle safety technology.Most of existing detection systems of the driver’s fatigue have low precision and work only with single daylight or single dark condition, so the detection and warning system of the driver’s fatigue with daylight and dark conditions is studied separately in the paper. Mainly from the aspects of the driver’s facial features localization and the corresponding recognition of the driver’s fatigue state, the paper use hemispheric integration camera SZA002-8 to realize the all-day detection and warning of the driver’s fatigue. The main contents in this paper are as follows:In daylight condition, first, realize the driver’s face detection by AdaBoost algorithm and on this basis, the paper propose a new rapid search method to realize the eye accurate localization by AdaBoost algorithm, then realize the mouth coarse localization by using the domain division method. Based on the facial features localization, the paper propose an improved ASM(Active Shape Model) algorithm based on feature points expansion and gray features extracted by PCA (Principal Component Analysis) to realize the contour localization of the eye and the mouth. The experimental data show that the average localization error is decreased by more than 38% in the improved ASM algorithm, and the system’s recognition efficiency of the eye state and the mouth state also improve greatly. Simultaneously, realize the driver’s fatigue state in daylight condition based on PERCLOS method and yawn characteristic. When the value of PERCLOS is larger than 0.4 or the length of yawn is larger than 5 seconds, give alert to the driver.In dark condition, hemispheric integration camera SZA002-8 that the paper adopts will produce the active infrared glow whose wavelength is about 850 nanometers to realize the collection of the driver’s near infrared image. According to the characteristic of the near infrared image, the paper presents an image enhancement method which is combined with wavelet de-noising and top and bottom hat transformation for the image pretreatment. Realize face detection of the infrared image by Otsu threshold segmentation algorithm, morphological filters and labeled algorithm of connected region, and on this basis, the paper obtain the eye area based on a new constraint algorithm. Then the paper locates Purkinje spot position based on Harris corner detection algorithm and detects the driver’s fatigue state in dark condition by the detection method based on Purkinje spot. Experimental results show that when the driver is in fatigue state, the system almost can’t detect Purkinje spot for a while and the value of PERCLOS is larger than 0.4 or significantly larger than 0.4, so the system will give alert to the driver.In summary, the paper studies the key technologies in the study on all-day driver fatigue detection and warning system based on facial features. The system has accurate detection and all-day detection characteristics, and has very wide application in the SAD (safety assist driving) field.