Study on the Technology for Driver Fatigue Detection Based on Video Information Extraction
|School||Tianjin University of Science and Technology|
|Course||Detection Technology and Automation|
|Keywords||Fatigue detection skin color model eye location face location PERCLOS|
With the development of society and increasing number of motor vehicles, traffic safety problems is more and more serious. A variety of traffic accidents each year have caused huge losses to society. Fatigue driving has become one of the main causes of accidents. Therefore, driver fatigue warning and detection system significantly affects the traffic safety and avoidance of traffic accidents.This paper mainly studies the non-contact and real-time detection of driver fatigue based on video image after considering the requirements of real-time and accuracy, the key technologies of video-based driver fatigue testing are studied based on the theory of video-image information processing and extraction, effectively optimization is done to the image processing algorithms in the context of good accuracy.The main contents of this paper include the following:(1) The video image extraction analysis. Pre-processing the original video image to make the sharp features more prominent in order to facilitate subsequent image analysis.(2) Face location. Based on the pre-processing image, confirm the face area in image background. Because the color of skin in the color space has good feature of clustering, with better real-time detection method based on skin color, use the projection method to reduce the amount of data in the calculation.(3) Eye location. As the eye location is a key of this project, the exact location of the eye position has a direct impact on the accuracy of the algorithm. Therefore, combine the integral projection function and the hybrid projection function and precise positioning the eye center. Then integrate Kalman filter algorithm into the experiment to realize tracking the location of the eyes in the changes of consecutive frames, to reduce the repositioning time and to improve real-time feature.(4) Analysis on the state of eye. Extract the eye feature on the basis of eye location. Analyse the main features which determine the eye state, including the corner of eyes, the iris and the upper eyelids. Modify the traditional eye model and use the height difference between the upper and lower eyelid to determine eye state.(5) Finally, use the PERCLOS fatigue analysis algorithm which is currently accepted as most effective method combining with the blinking frequency to judge the fatigue state. Carry out the real-time analysis on the premise of ensuring a good accuracy.The software systems of fatigue detection described above can real-time detect the eye state of driver and conduct fatigue identify. We have obtained good experimental results in laboratory conditions.