Vision-based Driver Attention Detection And Judgement Research

Driver distraction is one of the chief causes of traffic accidents.In recent years,vision-based driver attention detection has become a hot research topic.However,most of the existing solutions focus on only one feature of the driver's distraction,such as fatigue detection based on the driver's facial information,which usually ignores some other important information.In this paper,as for driver's attention detection and judgment research,we propose a vision-based comprehensive solution that fuses two kinds of visual data both from driver's face image and road frontal image.First of all,considering the overall framework,this paper designs a comprehensive vision-based driver attention detection scheme.This scheme uses the driver's facial information and road traffic information and comprehensively consider s the driver's fatigue status,head deviation,gaze direction and road hazard target matching situations,to effectively analyze the driver's attention.At the same time,we collected multiple sets of experimental data to screen and organize the data,the proposed scheme are demonstrated,which proves the feasibility of the proposed solution.Experimental results show that the proposed solution can effectively detect w hether the driver's attention is concentrated in the actual scene.Second,this paper proposes a parameter learning method to improve the program's adaptability.For different drivers,statistical methods are used to study the changes of visual parameters of different drivers.In the face detection and analysis module,a parameter learning phase is introduced to learn different parameters adaptively,such as the head deflection angle,the size of the eyes and the mouth,and so on.At the same time,simple and effective fatigue detection,head deviation detection,and matching scheme of gaze direction and road hazard target are given.Finally,an on-road dangerous object detection algorithm based on monocular vision is proposed.The algorithm combines optical flow method,ORB feature point matching between three consecutive frames,and k-means clustering.The detection result is not limited to the target category.The algorithm is based on monocular vision only and does not require complicated dual-target determination and image correction processes.The detection results are not limited to the target category,nor are they limited to still or moving targets.