面向3D道路点云地图的杂波目标去除算法  

Clutter removal algorithm for 3D road point cloud map

作　　者：赵文娇 王浩勇 Zhao Wenjiao;Wang Haoyong(School of Communications Engineering,Shanxi Vocational University of Engineering and Technology Jinzhong,030619,China;School of Mathematics and Computer Science,Shanxi Normal University,Linfen 041004,China)

机构地区：[1]山西工程科技职业大学交通工程学院,晋中030619 [2]山西师范大学数学与计算机科学学院,临汾041004

出　　处：《国外电子测量技术》2021年第8期23-28,共6页Foreign Electronic Measurement Technology

基　　金：山西省软科学基金(2020041033-02)项目资助。

摘　　要：道路信息高精度点云地图通过移动地理信息系统(MGIS)构建包含道路周围设施的信息,在运行MGIS时获取除设施以外的非结构化形式目标数据已成为研究难题。将除道路信息外的杂波定义为杂波目标,以MGIS移动平台为基础,利用几何数学模型建立杂波目标在地图坐标系中的坐标,通过Microsoft公共(MS-COCO)数据集对掩模R-CNN算法的权重进行预训练,从而加快算法对图像的实例分割,结合零填充方法完成图像缓冲区域的像素填充。通过在帧级去除杂波目标,然后对干净点云帧进行累积得到干净点云地图。将该方法应用于Kitti数据集,去除杂波目标的总体精度为95%,并且整个点云地图的误检误差仅为1%。High precision point cloud map of road information is constructed by mobile geographic information system(MGIS), which contains the information of facilities around the road. It has become a research problem to obtain unstructured target data except facilities when running MGIS. In this paper, clutter other than road information is defined as clutter target. Based on MGIS mobile platform, the coordinates of target clutter in the map coordinate system are established by using geometric mathematical model. The weight of mask R-CNN algorithm is pre trained by Microsoft common(MS-COCO) data set, so as to speed up the image segmentation, and the pixel filling of image buffer area is completed by combining zero filling method Charge. The clutter target is removed at the frame level, and then the clean point cloud frame is accumulated to get the clean point cloud map. Applying this method to Kitti dataset, the average overall accuracy of clutter removal is 95%, while the false detection error of the whole point cloud map is only 1%.

关 键 词：MGIS 点云地图 图像分割 点云去除 杂波目标 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置] TN958.98[自动化与计算机技术—控制科学与工程]