Design Of Driver Status Monitoring Based On Embedded System

In recent years,with the rapid development of Chinese economy,the amounts of cars in China has increased rapidly.However,the traffic accidents in China has also increased rapidly in recent years,and the current traffic safety situation is very serious.Nowadays countries all over the world are studying how to improve traffic safety.The advanced driving assistance system is an important research direction of safe transportation.It can be studied from two aspects of vehicle control and driver status monitoring.In the past,most of the researches obtains the surrounding environment information through various sensors from the perspective of vehicle control,analyzes it and gives the driver hints or directly controls the vehicle.However,according to relevant statistical data,the cause of most traffic accidents is that the driver does not operate the car in accordance with traffic regulations.Therefore,how to monitor the status of the driver has become a research hotspot in the field of advanced driver assistance.This paper designs an embedded system for the driver status monitoring to detect the dangerous behaviors of drivers that may cause a traffic accident and it will promptly issue an alarm to warn the driver.Thus,this system can reduce the possibility of traffic accidents from the source.This paper designs an embedded driver status monitoring system.The following five abnormal behaviors can be detected by it: closing eyes,yawning,bowing down,looking around and smoking.This system collects real-time images through the near-infrared camera,then determines whether there is an abnormal action at present through abnormal behavior detection.If an abnormal situation is detected,LED and audio alarms are issued.According to the performance of the embedded platform adopted by our system,we adopt the method which name is one millisecond face alignment with an ensemble of regression trees to detect face landmarks,the following detection methods according to the abnormal situation are proposed: facial motion detection method for detecting closing eyes and yawning,face orientation detection method for detecting bowing down and looking around,and cigarette butt detection method for detecting smoking.In order to improve the running speed of this embedded system,this paper makes two improvements for face landmarks detection: The first is face detection is performed by features with less computational complexity;and the second is initial landmarks location method is improved.In addition,the speed of the system is further improved by code optimization.And a near-infrared face sample library is built in order to make the system using near infrared images for detection more suitable for the actual application environment and improve the detection accuracy.The experimental results show that under the specified conditions,the face landmarks detection accuracy of the proposed system can reach 95%,and the detection accuracy for all abnormal conditions is above 92%.What's more,it also has good real-time performance,and all indicators meet the requirements of our partners.