Driver Fatigue Detection System Based On DM3730

With the development of social and economic, the standard of living is raising rapidly, and the amount of c are also increasing gradually. Although cars improve people’s quality of life, they also led to frequent traffic accidents, the death toll has been high, and the problem of the traffic safety is more and more serious. Many a good portion of traffic accident is caused by driver’s fatigue driving, especially at night. Therefore, a way to monitor whether the driver is fatigue and prevent they drive a car any more is an effectively way to avoid traffic accidents. This technology has important significance to improve the traffic safety. By using machine vision method to detect fatigue is becoming a research focus both at home and abroad. A real-time and non-contact machine vision method has more advantages than other methods, so it becomes an important method for fatigue detection.Based on the study of relevant research at home and abroad, in this paper, we designed an enhanced platform by using machine vision method to calculate the fatigue level of the driver all the time, which is small size, low power consumption and can work at day and night and all kinds of weather conditions. It also is a real-time equipment and can be used in the process of car driving. The main contents are listed follows:Firstly the inter-image difference between different frames which are under different kind of near-infrared light irradiation are calculated in system. Then the difference image is binarized and eye features are exacted to help us judge car driver’s fatigue level. We eliminate the influence of external light source through a light filter. Feature extraction methods are unified under different illumination by controlling near-infrared LED lights. The odd and even frames under different illumination have different feature, one has ’red eye’ effect and the other not. The difference value is the different on eyes. This eye location method is quick and accurate.Secondly, Otsu method is used in system to threshold difference image adaptively, and this system is able to appropriate different illumination environment. Without the environment implication, the pupil feature extraction becomes convenient and accurate.Thirdly, Kalman filter is introduced into the system, the location of pupil on next frame can be predicted based on priori knowledge, this greatly reduce the size of processing image and improve the operation speed, while also improving the feature extraction accuracy by excluding inappropriate characteristicsFourthly, this system select the TI Davinci chip as system platform, and we integrate our algorithm into the Codec, so Server can call for it.Finally, improved PERCLOS value and duration to make system more sensitive on fatigue drive detection.