Driver Fatigue Monitoring Research Based On Machine Vision And Grip Force Fluctuation On Steering Wheel

The world vehicle population was reported to have surpassed the 1 billion-unit mark in 2010(240 million in U.S.and 78 million in China).The increasing number of car drivers accompanied with more and more traffic accidents,which almost 20% of them involve driver fatigue that founded by Sleep Research Centre in UK.Privious research about driver fatigue detection technology are almost based on visual detection and Electroencephalography(EEG),Heart Rate,Electrooculography(EOG).But big part of the literature is video recognition techniques because of its non-intrusive application.However,visual detection that has some limitations in the condition of wearing glasses and lighting changes,in which circumstances traditional Binary segmentation dose’t work and bring low detect accurancy.Few researches on developing drowsiness detection systems based on steering grip force for drivers have been reported in the literature that prove grip froce can be a effective feature in detecting driver’s fatigue.To overcome those problems,this paper proposes a way to detect fatigue for drivers through video images and grip forces on steering wheel.The simulated results show that use of multiple visual parameters combined with steering grip force can effectively detect the driver’s fatigue and much robust than using single visual source.The main work of this paper as follows:(1)Simulated driving platform and driver fatigue monitoring system have been developed.In order to simulate the real driving operation as much as possible and to verify effectiveness of the algorithm,extraordinary hardware including a steering wheel and foot pedal been used that almost bring perfect driving exprience.The software we build can simulate all kinds of weather and road conditions.Driver fatigue monitoring system was constructed by camera and grip force sensor detection.(2)The original signal is collected by grip force sensor,which characterization ability is very limited.The original data need to be transformed and filtered.The smoothing method of the linear dynamic system model with better real-time capability is applied in the filtering process.After the pretreatment,the variance is selected as the grip strength feature in the time-domain features of the grip force.(3)The face detection algorithm based on Cascade structure is researched in human face localization.Adaboost algorithm based on Haar feature is applied to make face localization more accurate.All the visual process algorithm including Adaboost and improved active shape model have been realized on the platform of Visual Studio 2010 combined with OPENCV.(4)The 77 landmarks have been searched by using improved active shape model algorithm after locating the face.To overcome scale variability,we use the ratio of height and width to define openness.Finally,algorithm’s accuracy and robust have been tested in the circumstance of subject wearing glasses and normal state.(5)Driver fatigue detection algorithm based on the feature fusion of face and grip force is proposed.Percentage of eye closure,degree of mouth openness and grip force variance been used as input to design fuzzy rules.A significant amount of experiments have been done to adjust the fuzzy rules to meet the requirements of cross-consistency and consistency.Compare to other algorithm,the robust and accuracy of the proposed algorithm have been verified on the platform of MATLAB.