Research On Key Algorithms For Vehicle Lane Departure Warning System Based On Machine Vision

Lane Departure Warning Syste m,as an important component of I nte lligentTransportatio n System and from the vehic le active safety,applies modern computervisua l infor mation,auto matic contro l techno logy and so on to provide a alar m to thedriver when the ve hic le is in la ne departure.The research and deve lopment o f LDWSgreatly enhance the safety and comfort o f ve hic les. This paper ma inly researched andachie ved the camera calibration, the lane detectio n and tracking,and warning decis ionalgor ithms for LDWS based on machine vis io n.This paper first introduces the research background and meaning of this projectand summar ized the research status at home and abroad.And the n obtained the roadima ges fro m AVI video file and for mulate Monocular pinho le ca mera model.Improved Tsai two-stage algor ithm to achie ve the ca mera calibratio n by adoptingnonlinear optimization a lgorithm to optimize the ca mera parameters which have beenfigured out.Color-ima ges were preprocessed with gray sca ling by weighted averagemethod,de-nois ing a nd edge enhance ment.P roposed a adaptive thres hold wave letde-no is ing method based on the Donoho wave let shr inkage de-nois ing method,whichadopted a adaptive thresho ld method to solve a threshold that can isolate no ise fro mthe use ful infor mation in every la yer mostly.And it has good performa nce.Researchedthe princ ip le of edge enha nceme nt and chose Sobel algor ithm to enhance the edge bycontrasting and ana lys ing the princ iple and the processed images.According to the vehic le status and the results of the camera ca libration,setted upthe region of interest lane and proposed b lock-class ify algorithm in ROI to extract thefeature po ints of the la ne.Next,formulated a linear-parabolic lane model whoseparameters of r ight and le ft road models are solved by the least square methodaccord ing to the extracted feature points.Adopted Kalman to pred ict the la ne ROI ofthe current fra me based on the lane slope and intercept state values of the last fra meima ge.It realized real-time tracking of the lane and updated the ROI. Theexperimenta l results show that the la ne detection a nd tracking based on block-classifyfitting and Kalma n a lgor ithm ha ve good perfor mance for the straight a nd curve lanefitting.And the n for mulated la ne departure warning decis io n algor ithm based on symmetr ical angle and lateral intercept change, syste m can be decided warning or notthrough the overa ll judgme nt of the both. This algor ithm improved the accuracy of thewarning and overcame the low warning accuracy of the curve lane through adoptingsymmetr ical angle and lateral intercept change.Lastly,the la ne departure warning system is designed based on Opencv and VCin the Windows platfor m.The s imulation exper iment hao been carried to ver ify theaccuracy and valid ity of the a lgor ithms from the results o f the lane detectio n trackingand the lane departure judgme nt.