基于深度卷积神经网络的激光雷达图像轮廓线提取方法  

LiDAR image contour extraction method based on deep convolution neural network

作　　者：熊丽丽 冯丽佳 苑柳青 XIONG Lili;FENG Lijia;YUAN Liuqing(Hebei Polytechnic Institute,Shijiazhuang 050091,China)

机构地区：[1]河北工程技术学院,石家庄050091

出　　处：《激光杂志》2023年第10期94-99,共6页Laser Journal

基　　金：河北省教育厅科学技术研究项目(No.1803114)。

摘　　要：为提高地物轮廓线提取精度,提出深度卷积神经网络的激光雷达图像轮廓线提取方法。采用深度卷积神经网络进行目标点云识别,利用激光雷达点云数据构建不规则三角网,采用不规则三角网法向量方法检测激光雷达图像边缘点,搜索初始角点,通过角度阈值判断是否为伪角点,确定最终角点集合,采用变权重最小二乘法拟合激光雷达图像轮廓线。实验结果表明:该方法提取的激光雷达图像轮廓线误差小,轮廓线可以含全部边缘点,角点坐标最大偏差和占地面积相对误差低、点云贡献率高,轮廓提取效果突出。In order to improve the accuracy of surface feature contour extraction,a deep convolution neural network method for lidar image contour extraction is proposed.The deep convolution neural network is used to identify the target point cloud,the irregular triangulation network is constructed by using the laser radar point cloud data,the irregular triangulation normal vector method is used to detect the edge points of the laser radar image,search the initial corners,judge whether it is a pseudo corner through the angle threshold,determine the final corner set,and use the variable weight least square method to fit the contour of the laser radar image.The experimental results show that the error of the lidar image contour extracted by this method is small,the contour can contain all edge points,the maximum deviation of corner coordinates and the relative error of floor area are low,the contribution rate of point cloud is high,and the contour extraction effect is outstanding.

关 键 词：激光雷达图像 轮廓线提取 不规则三角网 轮廓线拟合 最小二乘法 

分 类 号：TN209[电子电信—物理电子学]