Lane Line Detection Based On Unsupervised Clustering And Departure Warning System

With the advent of the intelligent information era,intelligent driving has received widespread attention from industry and academia.As advanced assisted driving being an important branch of intelligent driving,more and more companies and scholars have invested in related research.LDW(Lane Departure warning)is one of the most important functions in advanced assisted driving technology and is of great research value,which can effectively reduce the probability of traffic accident caused by fatigue.In recent years,lane line detection based on deep learning has greatly improved the detection accuracy,but it relies on a large amount of training data and requires much computing power from hardware.The high equipment cost makes it difficult to be widely used.The traditional lane line detection method based on Hough-transform has a small computational cost,but the algorithm is greatly affected by the environment,resulting in a high rate of false detection.Aiming at the problem that the traditional Hough transform method for detecting lane lines has high false detection and poor robustness,a lane line detection method based on unsupervised clustering is proposed.First,the road plane is mapped into a bird’s eye view through inverse perspective transformation,and the shape of the lane line itself is restored to the greatest extent.Then the bird’s eye view blocking mechanism is innovatively proposed to vertically divide the bird’s eye view into blocks.According to the lane in the small image block,the straight line feature,the paired line feature and the color feature of the line are used to extract the lane line sub-segment.Finally,according to the coordinate position of the lane line sub-segment on the bird’s eye view and the angle information in the horizontal direction,the DBSCAN clustering algorithm is used to divide the different image blocks.The upper sub-line segment is clustered to get a complete lane line.Experiments show that the use of paired line features can greatly reduce the false detection rate of lane lines,and the bird’s eye view block mechanism makes the overall lane line feature extraction more efficient.At the same time,the clustering method can make the detection results closer to the shape of the lane line itself.A deviation early warning model based on the intersection distance of the axis is proposed.The distance between the front wheel axis and the lane line is quickly calculated by calculating the intersection of the front wheel axis and the lane line.Experiments prove that the model can effectively play a warning role for lane departure.In order to verify the effectiveness of the lane-line detection algorithm based on unsupervised clustering and departure warning model,a lane-line detection and departure warning system was designed and developed.The system was implemented using C ++ language and Opencv.A large number of driving videos are used as experimental data to verify the effectiveness of lane line detection algorithms based on unsupervised clustering and departure warning systems.