Detection Method Of Distracted Driving Based On Visual Features And Behavior Data

In recent years,with the popularity of intelligent terminals like cell phones,the accidents of distracted driving keeps increasing.And distracted driving becomes one of the main factors influencing traffic safety.Therefore,scientist make a lot of research about distracted driving.However,domestic study of distracted driving starts late,did not form a complete system.Analysis of distracted driving characteristics only consider general characteristics like number of glances,fixation duration et.al currently,it cannot reflect changing process of visual distraction,this is one of the limitations of current research.At the same time,this research lack the study of quantitative rating method of distraction tasks,and distracted driving detection method of driving behavior with high accuracy.Firstly,the paper designs distracted driving experiment with several tasks,then set up data export and/or dynamic display system.Professional and non-professional drivers are participated the experiment,software are made to preprocess and cut out the visual data.Driver's visual data and behavior data are obtained through the system,it lay the foundation of distraction feature description and distraction detection method present.Secondly,visual features of several distraction tasks is description,observe general visual features of drivers with different profession.AttenD algorithm is used to express real-time visual features of distraction tasks,and finally a visual assessment method of distraction task is put forward,this method can rank complexity levels and risk levels of distraction tasks.Results can be applied to feasibility assessment of On-Board Equipment,and offer theoretical basis for related management department to set standards.In addition,the paper describes driver behavior of different distraction tasks and different profession.Based on AttenD algorithm,three distraction levels are divided,correlation between visual and behavior data is built and driving behavior under different distraction levels is described.Significant level in different distraction tasks,professional types,and distraction levels is obtained by using one-way analysis of variance.Driving behavior indicators to detect distraction can be got,it offers the idea of distraction detection.Finally,the paper puts forward a classification method of distracted driving behavior by using random forest.The method is based on driving behavior,and the classification criterion is developed with the analysis of accompanying visual features,it conduct correlation research between driver's visual features and diving behaviors of distracted driving,and realize classification of driving behavior in different distraction levels.The model will detect distracted driving at a higher accuracy and lower cost without the use of an eye tracking system,since the model input is entirely based on driver behavior data.This study not only describes the visual and behavior features of several distraction tasks,and puts forward an assessment method to evaluate distraction tasks.Based on correlation ship between visual and behavior is established,a new detection method is put forward to identify distracted driving behavior.Research has stronger application characteristics.On the one hand,the system built in this study can be applied to experience and education of distracted driving;On the other hand,assessment method of distraction tasks provide the basis for feasibility evaluation of in-vehicle facilities;At the same time,distraction detection method can provide technical support for prewarning system in real vehicle.