Extraction And Recognition Of Road Markings In Vehicle-mounted Point Cloud

With the rapid rise and application of the unmanned driving industry,related research represented by vehicle-mounted lidar systems has continuously entered people’s vision.The lidar system provides high-density and high-precision point cloud data for vehicles in real time.These point cloud data can record the rich attributes of roads and surrounding objects,including accurate three-dimensional coordinates,color information,and reflection intensity information.Driving,urban surveying and other related industries provide reliable and effective data sources.The extraction and classification of target objects in vehicular laser point cloud data is the key to many researches and applications.In this paper,the method of extraction and recognition of road marking lines in point clouds is designed for the road marking lines existing in vehicular point clouds.To deal with the problem of large and disordered data of the original vehicle point cloud,after the original vehicle point cloud is segmented according to the driving trajectory,a threshold line-based point cloud elevation filtering and K-means clustering algorithm are used to identify the line points Cloud extraction method.The algorithm first selects segmented vehicle point cloud data,analyzes the Z-axis(elevation)information in the data through frequency histogram,determines the optimal threshold,and completes the point cloud elevation filtering to obtain the filtered road point cloud,and then the road The elevation information and reflection intensity information of the point cloud are used as the main clustering basis,and the road sign line point cloud extraction based on the K-means clustering algorithm is performed.For the problem of direct point cloud data recognition is more complicated and more difficult,the road marking line point cloud is projected on a plane,the appropriate grid resolution is determined,and the road marking line point cloud in three-dimensional space is converted into a two-dimensional overhead image to It uses more mature digital image target detection and recognition algorithms.This paper designs a road marking line recognition method based on the YOLO v3 target detection model.Through mirror image flipping,adding noise,rotating random angles andrandom cropping and other data enhancement methods,the number of training set samples is expanded to meet the requirements of deep learning for a large number of training samples.At the same time,optimize the size of the prior frame,and use the transfer learning method to complete the training of the YOLO v3 network model.In this paper,we conducted multiple sets of experiments to verify that the trained YOLO v3 network model has a good effect on the detection and recognition of marking lines,and the recognition accuracy can reach more than 90%,which can provide research ideas for the recognition of marking lines in point clouds.The model and the future vehicle assisted driving have certain reference significance.