Research On Lane Recognition And Tracking Method In Complex Environment Based On Feature Point Clustering

Lane recognition is an important part of intelligent vehicle environment perception.Lane line can provide key information for vehicle positioning in the road.Different lane line types also correspond to different vehicle behavior constraints.Lane detection,type recognition and tracking have also become a common concern of researchers.Aiming at the problems that the current lane recognition is easy to be disturbed by the environment and difficult to take into account the real-time and robustness,this study studies the lane recognition and tracking in complex environment.The main research contents are as follows:(1)Graying and inverse perspective transformation,image denoising and Tophat contrast enhancement are carried out on the lane line image;The edge detection operator and the lane line edge and color features extracted by multi-color spaces binarization are fused,and the Yellow feature extraction threshold is set for feature fusion optimization to improve the real-time and accuracy of the feature fusion algorithm.(2)Combined with the distribution characteristics of lane lines in the aerial view,a lane line feature point clustering method based on RC-DBSCAN(Reclustering based on DBSCAN)is adopted.According to the clustering results,lane line fitting and type recognition are carried out.DBSCAN algorithm is used to cluster the feature points in the preprocessed binary image,and the clustering results are projected to the x-axis of the image coordinate system to obtain the corresponding projection cluster;The Euclidean distance between the centroids of different projection clusters is compared with the secondary clustering threshold,and the secondary clustering of feature point clusters is carried out;The lane line cluster is extracted by histogram peak value,and the lane line is fitted based on least square method and parabola model.In lane line type recognition,the image is white balanced to reduce the influence of illumination on color information;The lane line color is recognized in lab color space,and the lane line type is recognized by using the characteristics of whether the lane line cluster has secondary clustering or not;Combine the color and line type recognition results to obtain the type information of lane line.(3)Comparing different tracking algorithms,combined with the similarity of lane line positions of adjacent frames,a Kalman filter tracking algorithm based on trusted region optimization is proposed.Through the prediction and updating of lane line parameters,the interference of complex environmental factors on the detection results is effectively reduced.The experimental images are collected in complex multi scene roads and the algorithm is verified.The results show that the algorithm has good robustness for lane line detection,low false detection rate in multi scene,and its real-time performance can also meet the needs of lane line detection;The accuracy and real-time of lane line type recognition in structured roads can also meet the actual needs.