Research On Key Technology Of Lane Departure Warning System Based On Machine Vision

Lane Departure Warning System(LDWS)is one of the important functions of Safety Driving Assist System(SDAS)in Intelligent Transportation System(ITS).LDWS obtains road information through sensors,and judges the trend of derivation from the current lane according to the vehicle’s driving state,warning time,distance and other relevant parameters.If the vehicle would deviate without the turn signal,the driver will be alerted by visual,auditory or tactile means.The in-depth study of LDWS and its wide application will vastly improve the safety of drivers and reduce the probability of traffic accidents.At present,LDWS research based on machine vision has become a focus at both home and abroad.In this dissertation,the key techniques and algorithms of LDWS are studied based on monocular Machine Vision.LDWS generally consists of lane detection,lane modeling,departure decision warning,etc.Whereas,lane detection technology under bad environment conditions still remains a problem up to now.It mainly focuses on the key functional modules of LDWS,including the following contents.(1)An adaptive threshold algorithm for lane line detection based on single-row sliding window is designed in the paper.In the normal weather of shade of trees,light haze and other circumstances,it proposes a sliding window in image detection and a dynamic detection area to identity the lane line in terms of double-peak spectrum of single pixel image.Through experiments,it is found that this algorithm has a high accuracy and robustness.Meantime,it has a significant suppressing effect on the single point noise which exists independently,or the continuous noise whose length more than 2 times of the sliding window.(2)An algorithm on lane line detection at night is specially described based on Fractional Differential and Fangi and Hessian.Aiming at the characteristics of weak light at night,high image noise and less lane information in a single frame image,the algorithm combines 3-4 consecutive images to offer the rich lane information in the image.The merged images are treated with noise reduction and enhancement,then the lane information is detected.Experiments show that the algorithm can detect lane lines quickly and effectively in the most night road images and achieve 76 percent accuracy.(3)A detecting algorithm of lane image on ridgelines is proposed for.In the weather of heavy rain,fog and haze,the accessed road images have many interfering factors,such as rain,fog,haze,water,surface water,and noise,which leads to a significant decline in contrast.Thus,this dissertation designs an enhancement image algorithm that is combined with detecting algorithm in ridge lines,skeleton extraction,gap junction,etc.The recognition results mention that it is a real time algorithm for detecting lane and achieves 71 percent accuracy.(4)The practical significance of the linear model are emphasized in this study,combined with "Highway Engineering Technical Standard"(JTG B01-2014)and "Intelligent Transportation System Lane Departure Alarm System Performance Requirements and Detection Method"(GB/T26773-2011).According to Kalman filter approach,the paper traces the identification line of lane and improves the detection efficiency.(5)An improved TLC model algorithm with FOD is put forward in the dissertation.It examines the advantages and disadvantages of each decision model,and analyze the velocity,rate and trend of the variation of distance between vehicle center and left and right lines under the static ROI.Combined with such studies and the driving characteristics,this new algorithm is designed.Through experiments,it makes false alarm rate of less than 5% and a missing alarm rate of less than 4%.