Design Of Machine Vision Based Detection System On Drivers' Rearview Mirror Watching Behaviors

Statistics show that 25%~30% of the traffic accidents is directly related to the driver's vigilance,and one of the main accidents is the driver's steering control,such as turning,merging,changing lanes,in which especially common when the driver did not pay attention to vehicle steering side rear traffic information.Real time detection and necessary warness of the driver's rearview mirror watching behavior is conducive to reducing the occurrence probability of such traffic accidents.This paper puts forward the detection method and the technical scheme based on machine vision and image processing technology to explore and research the behavior,and with a lot of experiments and examples verification,eventually empoldered the detection system on driver's rearview mirror watching behaviors.The main work and research achievements include:(1)On the investigation of a large number of related technologies and papers at home and abroad and in view of the problems that may be encountered in the subject,this paper presents a method of detecting the driver's rearview mirror watching behaviors only based on the driver's face and neck outside contour lines.Algorithm quantity is small,better real-time performance and moderate robustness.(2)Based on driver behavior characteristics,a driver's face and neck regions static identification algorithm was designed to complete the grayscale average learning work of driver's face and neck skin and static regions search identification in the current light when the vehicle engine was fired.Then a driver's face and neck regions dynamic identification algorithm was invented to accomplish the grayscale average learning work of driver's face and neck skin and regions fast tracking recognition when the vehicle was driving.Finally line scanning method was used to extract the contours of driver's face and neck,and an area ration between left and right parts of the contours separated by a vertical line passing through the base point of neck contour was defined as a characteristic parameter.The images processing results showed that the method has stable self-adaptability and was robust to some disturbance.(3)In view of different characteristic parameter's benchmarks not only caused by differences in driver's face and camera installation location,but also influenced by disturbance of driver's hairstyles,accessories and so on,a discipline called local peak value of the characteristic parameter's cumulative probability was uncovered,which help to build a threshold judging principle of the driver's rearview mirror watching behaviors.All parameters update cumulative probability in time in the process and parameters estimation renewed.The experimental statistics show that the effective data of rearview mirror behaviors checked can complete parameter's adaptive adjustment.(4)The detection system is small size and easy to popularize.System base on Raspberry Pi 3 microprocessor and embedded Linux system platform,QT was used for designing user interface,Open Source Computer Vision Library(OpenCV)was used for image interface functions.Test results showed that system has stable real-time performance and universally applicable ability.