Research Of Vision Attention Modeling And Unsafe Driving State Analysis

With the ever increasing amount of cars,the safety of traffic has aroused more and more attention.As an important factor of traffic safety,the behavior of drivers is one of the most important thing to consider about.Therefore,it can provide important information and prevent traffic accident by judging whether the behavior of the drivers is safe or not.In this paper,we conducted a research in vision modelling and developed a real time system to monitor the unsafe behavior of the drivers.This research has these achievements listed below,1 A saliency detection database has been developed,this database not only has a highly accurate eye movement information but also has saliency region labeled according to the eye movement information.This database also has annotation of language level to the saliency region.This database also collected the driver’s eye movement information and labeled the saliency region for the first time in this research field,benefit for the research in traffic scene.2 Developed an improved Intelligent Scissors algorithms to assist the process of labelling pictures,both the accuracy and efficiency are tremendously improved by this algorithm.3 By collecting and analyzing 16 state of the art saliency detection models,we unveiled the effect of color information to saliency detection models.4 This paper introduced a novel real time deep neural network based saliency detection model,this model is based on the state of the art ZF network to extract features from pictures,then employ RPN to generate ROI regions and extracted feature vectors from the feature map according to the ROI region,it combined the global information and region information developed from ROI to predict saliency region.This model can achieve a frame rate of 15 per second thus fulfillment the need of real time system and solid the foundation to develop driver unsafe state system;5.By employing the SDK of Tobii eye tracker,this paper introduced a driver unsafe driving state monitor system to judge the safety state.This system combined the saliency region predicted from scene picture and eye movement information got from eye tracker,by developing a driver attention judging algorithm,this system is capable of rising alarms by monitoring driver’s safety state.