Research And Implementation Of Independent Detection Of Driver Fatigue

With the rapid development of our economic and the unceasing improvement of thelevel of people’s material life, motor vehicles are increasing, at the same time, the trafficaccident due to fatigue driving also gradually increasing because of work intensityincreasing, it’s a serious threat to people’s personal and property safety, how toeffectively prevent fatigue driving and implement autonomous detection of fatiguedriving is a hot spot in the transport agency and related research and developmentinstitution. Autonomous detection of fatigue driving is real-time to determine whetherthe driver is driving in fatigue, and take timely measures when the driver into the stateof fatigue, thereby, reduce traffic accidents to some extent, and make people’s travelmore secure. So, implement independent of fatigue driving detection has considerableimportant social significance.The thesis chooses fatigue driving as the research object, the fatigue detection basedon image processing and pattern recognition is to detect whether the driver is fatigue byprocessing real-time video images, the method just only needs one camera, not interferewith the driver’s normal driving, and it has a higher operability, with strong real-timeand intelligent at the same time, the key technologies of the fatigue detection based onimage processing include face detection, eye location and eye feature extraction. Thereare currently so many algorithms be used for face detection and eye positioning. Thereare advantages and disadvantages for each algorithm, The thesis puts forward improvedalgorithm used suitably in the high-speed driving environment while illuminationchanges drastically through the comparative analysis of various algorithms, andproposed a fatigue criteria for multi-feature fusion, realize and ensure that the system iscapable of automatic warning when the driver is found in a state of fatigue, and remindsdrivers to slow down or stop to rest, so as to effectively improve traffic safety.Firstly, The thesis uses Adaboost algorithm which based on the Haar feature to trainclassifier for face detection, the method is insensitive to illumination change andsuitable for high speed driving environment, in addition, the testing result is good; Then,Taking geometric feature method to realize eyes’ coarse positioning by inherent characteristic values between face and eyes based on the result of face detection, at thesame time in order to eliminate the deviation caused by head deflection, the thesisproposed auxiliary Frame Difference method which commonly used in target tracking todetermine whether the eye area by coarse positioning is a big deviation, and implementfurther accurate location for eyes by using clustering technology to eliminateinterference around the eyes part; and then, use the connected region labeling method toextract eyes’ multiple feature value; Finally, using self-organizing competitive networkwhich without teachers’ guidance and have strong learning ability to give differentweights for several eyes characteristic values, and to make decisions for the open-closestate of each frame image of real-time video together, a fatigue criterion formulti-feature fusion based on PERCLOS parameters was proposed to determinewhether the driver is in a state of fatigue, to implement the independent detection ofdriver fatigue.After the accomplishment of fatigue driving detection algorithm, then we analyzereal-time ability and the accuracy, the results shows that the algorithm basically reachesthe expected effect.