Research On Classification And Recognition Of Road Target Imaging Using Laser Rada

With the advantages of strong anti-interference ability,high spatial resolution and strong environmental adaptability,Lidar is widely used in intelligent transportation fields such as intelligent traffic monitoring,vehicle-road coordination,and autonomous driving.This thesis focuses on the classification and recognition of lidar road target imaging.The main tasks completed are as follows:Aiming at the lidar point cloud imaging in the road scene,firstly,the three-dimensional imaging principles of different types of lidars are introduced,and their characteristics are explained separately.Secondly,the ground point cloud removal method is researched.Aiming at the problem that the RANSAC(Random Sample Consensus)plane extraction algorithm cannot accurately fit the uneven ground,the second-order polynomial surface model is used to replace the plane model and the initial point selection method and the error threshold setting method are carried out.Then,in view of the problem that the point cloud denoising algorithm cannot take into account the large-scale noise and the small-scale noise at the same time,a joint denoising algorithm based on statistical filtering and normal vector correction bilateral filtering is proposed,which realizes the simultaneous detection of different noises in the denoising process.Next,the point cloud segmentation method is researched.In view of the large amount of calculation in the point cloud processing of the traditional clustering algorithm and the poor effect of the target segmentation for the uneven density of the point cloud image,the preprocessing Meanshift clustering method based on two-dimensional projection is adopted.On the KITTI data set,the point cloud target segmentation Io U obtained by the improved algorithm is 73.2%,which is 9.0% and 5.8% higher than the area growth algorithm and the spectral clustering algorithm.Finally,for the problem of setting specific parameters when using SVM(Support Vector Machine)to classify point clouds,an adaptive parameter selection method is proposed,and the target classification accuracy of the improved algorithm reaches 96.63%,compared with the traditional support vector machine algorithm,it has increased by 4.17%,and the mean square error of the recognition rate has reached 1.6%,which is 1.2% lower than the traditional algorithm.The above research results have certain reference value for the application of lidar in road scenes.