Driver Status Detection Based On Multiple Facial Information

In recent years,the incidence rate of road traffic accidents rises synchronously with the rapid development of transportation industry.Among the top ten causes of death in the world published by WHO,Road traffic injuries rose from the 10 th to the 7th place in the past 16 years,while the other nine causes were related to diseases.There are many inducements of road traffic accidents.Human are the most complex and uncertain factor in the whole driving chain.The influence of the factor in traffic accidents is as high as 90%.The main reasons are not suitable for driving and the driver's lack of concentration.Therefore,the use of modern scientific and technological means to effectively identify the driver's bad state and timely warning is an effective measure to reasonably prevent road traffic accidents,and also an important topic in the current road safety management research.However,the individual differences between drivers,the complex driving environment and the limitations of existing driver state detection methods make it difficult to propose a technology of driver state detection with high accuracy,high real-time and strong robustness.To solve the above problems,this paper will launch the research of driver state detection based on multiple facial information.Based on the non-invasive data acquisition,with the goal of comprehensive recognition of driver's state and multiple facial information as the analysis object,this paper will use big data analysis and machine learning technology to mine the representation index of facial behavior.Through the recognition and intensity estimation of facial action unit,it realized the comprehensive detection of driver's emotion,fatigue state and distracted state in the process of driving.The specific research contents are as follows:(1)Considering the problems of the existing facial action unit database,a Chinese facial multiple information database with single action unit,basic expression and fatigue state is established.And the spatiotemporal characteristics of facial action unit are analyzed.(2)Constructed the geometric features of human face by using Facial landmarks.Realized feature selection and feature extraction by using Otsu binarization method and the improved principal component analysis method.Trained the intensity estimation model of facial action unit by using radial basis function neural network model.Improved the efficiency of model construction by using dynamic neuron addition method.(3)Based on the analysis of big data,the representation of facial action unit under basic expressions is mined.Considering the characteristics of high-dimensional and small-samples data,combined with the spatiotemporal characteristics of expressions,the recognition of driver's transient expressions is realized by using support vector machine algorithm,and the driver's emotion index is proposed to realize the continuous expression of transient facial expressions,and finally the objective evaluation of driver's emotion is realized.(4)Build driver status collection platform.The videos of normal,fatigue and distracted states were collected during the drivers driving,and the facial features which can represent the driver's fatigue and distraction were mined.Considering the differences of people in the fatigue state,the driver fatigue state detection is realized by using the method of feature matching.Based on the geometric features of the driver's face,the driver's distracted state is detected with the face orientation as the representation index.The experimental results show that the detection algorithm proposed in this paper has a high recognition accuracy,and can detect the driver's emotion,fatigue state and distracted state in real time and accurately.The results can be applied to driving assistant system,driver suitability detection,driver admission and other fields,And the results can improve driving safety,reduce or avoid driver's bad mood or distraction.It is great significance to reduce the road traffic accidents.