Design And Implementation Of Fatigue Driving Detection System Based On Kinect

The research of fatigue driving detection system is one of the hotspots in the field of intelligent transportation. Since the fatigue driving detection technology has been going on for many years, a number of achievements have been obtained. However, so far no one mature environmental-adapted product has come out due to the difficulty caused of individual differences and environmental complexity. With the rapid increasing of urban populations and vehicles, many companies and researchers’ attention is attracted by the fatigue detection products because of their broad market.In this thesis, a real-time and non-contact fatigue driving detection system is designed and realized.Kinect is used as the input end of the video of the system. The implementation platform is based on VS+OpenCV. This thesis is based on the idea of stepwise segmentation, reducing the interference, locating the human eye accurately, analyzing the open and close state of eye and judging the fatigue according to the PERCLOS principle.The system consists of six parts, including distance-based segmentation, face detection, face tracking, eye localization, eye state analysis and fatigue detection. Experimental results show that the system can judge the open and close state of eye in real-time, detect the driver fatigue effectively, realize the function of fatigue detection, and reach the expected purpose.By comparing some classical algorithms, a suitable algorithm is selected according to the design requirement of the system. The distance-based segmentation is realized by using the feature of Kinect. The interference caused by people sitting in back row or the scenery outside the window is avoided successfully. Using the method of combing face detection and face tracking, the real-time face detection is realized. On the basis of reaching the function of eye location, the accuracy of eye location is further improved by optimizing the eye location algorithm. The problem caused by shadow gusty peak of eyebrow and eye to lacate eye is solved. Considering the image features of Kinect, the thesis designed a method for recognizing the eye state based on multi-threshold pixels. The judgment of the eye state is realized under certain conditions.Finally, the main work of this thesis is summarized, and the shortcomings of the system and the future research directions is analyzed.