Study On The Method Of Mouth Segmentation And Location Based On Monocular For Driver

Statistics in many countries show that death traffic accident is directly or indirectly due to drivers focused their attention, caused by fatigue or sleepiness. Driver's factors have been one of the most important causes of road accidents. Until now many research have focused on monitoring the driver's face, eye, pupil and so on to obtain his/her face rotation and orientation, eye activities, eye blinking rate, gaze direction, finally to determine his/her fatigue or distraction state. However, Most of researchers have neglected driver's fatigue state such as driver's yawning and his/her distraction like conservation and talking on a cellular phone while his/her driving. Driver monitoring has been a focus of Safety Driving Assist technologies research. In Which, fatigue driving and driving spirit scattered condition monitoring system will play an important role in lowering the accident rate. Machine Vision in real-time, accuracy, and applicability of economic and other aspects has greater advantages than other monitoring methods. Study on driver visual monitoring Using vehicle-mounted camera systems is the hot technologies today. Many researchers focused on tracking the driver through the face, eyes, the pupil, has been head rotation and direction eyelid movement blink frequency, driver fatigue monitoring the direction of attention or mental scattered. However, the driver Yawns and prolong conversations with others, so to speak, or use cell phones while driving or driving fatigue did not receive the spirit of scattered attention, We can also detected by the driver, the driver's mouth to fatigue and mental state decentralized monitoring. Driver's mouth detection and location technology has a direct impact on the state of the driver mouth detection. According to the analysis of the state of the driver mouth, this paper proposes Several methods of driver's mouth detection and location, which lays a foundation for driver monitoring technology for further study and provides reference information and support for driver monitoring technology of the integrated monitoring system.The research of the paper includes five parts : driver's face detection, mouth detection based on gray, lip detection based on lip color, comparison and analysis of mouth segmentation algorithm, driver mouth location.Driver's face detection applies human face skin color model. YCrCb color space and color Gauss model are chosen. using the Cr and Cb of different skin color with the same Gauss model characteristics, gray images normalized by similarity are segmentalized by the maximum variance of the similarity threshold segmentation method, then Using Connected component labeling algorithm locate driver's mouth. Using Projection and the relationship facial geometry to determine the regions of face in order to determine ROI mouth. Experimental results show that: This type of driver's face detection and location has high reliability, real-time. It has good dynamic positioning capability, and has better adaptability for different light, complex background, and the driver sitting positions. It lays a good foundation for mouth detection and monitoring.When driver is talking or Yawning, lips or the shadow of mouth cavity and the surrounding region of the skin has a high contrast. According to these features we analyze the characteristics of each type of image segmentation algorithm effectively. For the feature of the image histogram is bimodal, the ideal method of segmentation is the segmentation algorithm based on the histogram threshold; For low signal-to-noise and the more complex background images, the results of the maximum variance segmentation algorithm is better; According to the gray difference between mouth cavity and lips being great, while differences between Gray pixels in mouth cavity or background is small, Moment automatically adjusted based on gradient threshold is adopted, the algorithm proved effectively.In the face of driver's Lips detection based on color, according to the character that driver's lips are redder than face, HSI color space that can segment mouth is chosen. Then according to separation between hue and brightness and color, we analyze which segmentation algorithm may be effective for the type of image segmentation for the features of the region of interest of mouth using H. The segmentation algorithm of maximum variance in the two-dimensional is adopted for lower SNR and complex background images; The K-means clustering method is adopted for uneven illumination images. In contrast with the mouth segmentation algorithm analysis, several dynamic threshold segmentation algorithm based on gray and color. The results showed: the segmentation algorithm of the moment automatically adjusted based on gradient threshold has better effect than the segmentation algorithm based on histogram threshold and the maximum variance segmentation algorithm. It has the same better real-time with the maximum variance segmentation algorithm than the segmentation algorithm based on histogram threshold; K-means clustering algorithm has better real-time than the segmentation algorithm of maximum variance in the two-dimensional, whose segmentation effects for most images are valid. It has high robustness. The 2D maximum variance is more sensitive to strong light than K-means clustering algorithm.According to contrast between gray and color, the results showed: When mouth is open, image threshold segmentation methods method based on gray is better than based on color; When the lips close or there are some or all teeth in mouth, image threshold segmentation methods method based on color is better than based on gray.Driver mouth location adopts Morphological processing algorithms to remove the noise in the light of the mouth images after segmentation. Hole filling algorithms based on regional growth is adopted to eliminate the discontinuousness of mouth region after segmentation. Then connected component labeling algorithm is utilized to mark areas of interest images of driver's mouth. Using Constraints of location and shape to locate the driver's mouth, and finally projection is adopted to find mouth region. The results show that: the above methods could better locate the mouth under different circumstances.All necessary software is developed using Visual C++ and Window 2000. The software realizes various algorithm functions. The experiment results prove that the software is valid and available.