Research On Driver's Fatigue Detection Based On Eye State Using ELM

Recently, traffic safety problems have become a widely discussed topic around the world. Around 40% of traffic problems are caused by fatigue driving based on the statistics. Research on the driver fatigue monitoring methods has a very important practical significance to prevent the occurrence of traffic accident.This thesis analyzes the existing driver fatigue detection technology based on the video analysis and proposes an ELM (Extreme Learning Machine) technology based on the eyes state to classify the driver fatigue. At first, the skin color segmentation is used for the face detection. And the KLT (Kanade-Lucas-Tomassi) algorithm is adopted to track each frame from the video based on the detected data. Then the integral projection function is used to estimate the eye area and detect the eye opening or closing state. Meanwhile, the eye opening or closing degree distance is also calculated for a set of images or frames. After that, all the features are extracted (blinking frequency and eye closure time) based on the above data. Since multiple iterations are adopted on the video, data volumes will be very huge. In order to reduce the time complexity, sliding window technique is introduced for real time data processing. Finally, this thesis uses the extracted features for the ELM training to achieve the driver's fatigue detection and classification.In this thesis, we used a real time network camera for the video acquisition in the experiment. The results show that the proposed methods can satisfy the real time requirement and extract driver's eyes features to judge the driver's working states effectively.