Research On The Real-time Image Processing System For Driver's Fatigue Monitoring

At present, the traffic accidents lead to millions of vehicle crash and fatal casualty and the driver's fatigue is one of important causes of vicious traffic accident. Fatigue is a physiological phenomenon, which is begotten by overage manual or mental work is, the normal physiological orderliness. Whether real-time detecting the driver is fatigue or not, whether reacting in time to avoid the accident as soon as possible when the fatigue appeared, that is very significant. "PERCLOS" (percentage of eyelid closure over the public over time) is a fatigue detecting method based on machine vision, which could realize non-contacted, real-time and round-the-clock detecting without disturbance to driving, and satisfy the particularity of driving drowsiness prevented application. The FHWA (Federal Highway Administrator) consider that PERCLOS is the most promising method in all real-time driver's fatigue warning techniques.In order to carry out PERCLOS—the fatigue measurement which based on image processing, we set up the real-time image processing platform on the basis of DSPs (Digital Signal Processors). In the interest of the real-time property, the DSPs we select the TMS320C6416 chips produced by TI (Texas Instruments) Corporation, and develop the real-time operation system (RTOS) under the CCS (Code Composer Studio) 6000 2.0 Integrate Development Environment.On the program design we according to the C6000 software development flow proceed the code writing. After carry out the base function of the program, we impose the debug and profile tools to analyze the performance of C code, and then according as the hardware property of the DSPs chips take relevant method to optimize the program, improving the parallel utility grade of hardware resource, consequently elevating the code's running speed and content the real-time requirement. The key of the usage of PERCLOS to the real-time monitor of fatigue driving is accurately tracking the drivers' eyes and judging the state of the eyes. By the way of carry out the round-the-clock detecting, we collect the video signal by the infrared CCD camera, assure that whether the daytime or the night could monitor regularity. When the light is abundant, the images we get are colored images, while it is gloomy; the images we collect are gray images without any colors. To the different two quality images is difficult to take the same image processing algorithm real-time accurately tracing the eyes. Via analyzing the two images respective peculiarity, utmost using the information included in the images, take the most feasible method to tracing the eyes. As for colored image, we apply to the complexional clustering property to classify the pixels of those images. In this way, it is swift and dependable to filter eyes from the complex facial region. Later on the driver's eyes can be tracked and ascertained through applying hybrid projection function in the region of driver's face. For gray images, we can't detect the people's face through the characteristic of complexion clustering. But under the infrared illumination, the gray value of the eyes is less than that of the face and has a clear circle area. So we can detect the circle by boundary detection, binarization and then Hough Transform and finally track the eyes. This system can accurately track the driver's eyes in real time, so that place the basis of the PERCLOS fatigue detecting method.