Research On Real-time Monitoring And Recognition And Behavior Evaluation Of Drivers Of Commercial Vehicles

In recent years,with the development of the socio-economic level and the increasing number of vehicles every year,road traffic accidents have occurred frequently,especially the accidents of operating vehicles will cause serious losses.The occurrence of traffic accidents is closely related to the bad behavior of drivers.The existing networked monitoring system can no longer meet the needs of intelligent,real-time identification of the driver’s dangerous posture.At present,the development of computer technology provides the possibility to dynamically recognize the driver’s posture.However,at both the level of theoretical research and market products,several common problems have been found.the research objects and scenes are too limited,the types of recognition posture are limited,and the superimposed postures cannot be effectively recognized.The research on the characteristics of driver postures is insufficient,and the lack of quantitative evaluation of driver posture behavior and warning methods is obvious.In this paper,the groundbreaking analysis of the driver’s posture has six characteristics of continuity,diversity,superimposition,similarity,transition,and mutual influence.The label processing method of the driver’s posture data set is designed according to the driver’s posture characteristics.A multi-cascade convolutional neural network driver posture recognition model was proposed to achieve 11 basic postures and superimposed postures,and a total of 53 types of driver postures were effectively recognized.The accuracy rate of the model in the daytime data set is as high as 98.68%,the recognition time of a single image is 405 ms while the accuracy rate in the night data set is as high as 98.03%,the recognition time of a single image is 362 ms.The accuracy and speed of this model are both infinitely perform better than the model without multi-cascade convolutional neutral network.This paper summarizes 8 qualitative indicators and 4 quantitative indicators by observing driving videos and questionnaires,and establishes driver behavior evaluation models respectively.Then combine qualitative and quantitative to form a comprehensive evaluation model of driver’s posture and behavior.According to the model calculation,the postures selected sorted from high to low according to the degree of danger as follow: play mobile phone> make a call> pick and place things>eat(smoking)> control the dashboard> control gear> one-hand driving> normal driving.In addition,this paper also conducted in-depth construction of a safety evaluation model for multiple posture stacking.Through the four universal constraints of posture superposition,a mathematical model is established to quantify the risk index of double postures,three postures and even more postures superposition.Finally,according to the quantified index value of the recognition gesture,a combination of three prompt methods of sound,vibration and light is proposed to warn the driver.Finally,this paper uses Py Charm + Py Qt5 to develop a real-time monitoring and recognition system for the driver’s posture of operating vehicles,and to achieve the core functions of cascaded convolutional neural network and driver posture behavior evaluation and early warning.Realize the core functions of cascaded convolutional neural network and driver attitude behavior evaluation and early warning.