Development Of Simulated Driving Platform For Real Cars And Research On Fatigue Detection Algorithm Based On Driver's Facial Expression

With the improvement of people's living standards,cars are becoming more and more popular and traffic safety problems caused by them are becoming more and more serious.Fatigue driving is one of the main factors that induce traffic accidents so fatigue driving has become the focus of current research.In order to study fatigue driving,an experimental platform for fatigue driving recognition and intervention based on real cars is developed.This paper has conducted a lot of research on various fatigue detection algorithms at home and abroad,analyzed their advantages and disadvantages and proposed a fatigue detection algorithm based on driver's facial expression.Combined with facial local binary pattern features,a dual-channel convolutional neural network is built to determine the driver's fatigue.Finally,the driver will be warned and intervened based on the result of fatigue determination.The main research contents and innovations of this article are as follows:(1)Development of simulated driving platform for real cars.The experimental platform retains the measurement and control motherboard and virtual driving software of "easy driving star" car simulator,which is realized by transplanting the simulator function to the scrapped car.In order to realize signal collection of car's control mechanism,the signal relay board is designed to collect the operating information of steering wheel,shift lever,handbrake and other mechanisms and then match with the motherboard.Road condition simulation is realized by DC motor drivers driving the electric jacks supporting the car chassis.Finally,driver image acquisition system is built based on raspberry pi computer.The system is used to collect facial image data when the experimenter is driving on the machine and upload it to the PC.(2)A face detection algorithm combined with MOSSE tracking is proposed.The algorithm uses Adaboost algorithm based on Haar-like features to detect human face and uses MOSSE tracking algorithm to continuously track the driver's face.Aiming at the problem that filter updating rate of classic MOSSE algorithm cannot adapt to target and background changes,dimensional analysis method is used to establish filter updating rate prediction model,which can keep updating rate in the best state and improve the tracking effect.Compared with Adaboost algorithm alone,face detection combined with MOSSE tracking has greatly improved in real-time.(3)A fatigue recognition algorithm based on dual-channel convolutional neural network is proposed.Aiming at the problem of incomplete facial expression feature extraction in the single-channel convolutional neural network,a dual-channel convolutional neural network is designed to extract the driver's facial expression features.The inputs of the two channels are facial grayscale image and LBP feature map,final fatigue judgment result is output through feature fusion and classification network.Compared with single-channel fatigue identification,dual-channel fatigue determination performs better in accuracy and robustness.(4)Design of fatigue warning and intervention system.Main functions of the system include fatigue driving warning and fatigue intervention.When the driver is judged as fatigue driving by the fatigue detection algorithm,the system makes corresponding levels of warning and intervention response according to the fatigue driving duration.To ensure that the driver arranges the driving schedule reasonably,it plays a role in escorting the safety of life and property.