构建三维道路数字模型对智能车服务和道路管理具有重要意义。文中针对高速公路不同路段应用场景下车辆运行速度快、干扰噪声多、特征少和无回环检测辅助等一系列问题，提出一种以激光雷达信息为建模基础数据、激光雷达里程计与LOAM技术等多传感器融合的高速公路三维建模方法。首先，通过车载激光雷达获取道路场景的激光点云数据，使用激光雷达图像分割技术赋予每一个点有关构造物的标签，剔除道路上其他运动车辆的信息，减少建模噪声；其次，制定了一个精确的同步策略来对GNSS、IMU和激光雷达等传感器进行集成；在此基础上，结合惯性导航预积分结果、基于特征点云的位姿约束和RTK数据构建因子图，消除激光雷达里程计的累积误差，从而构建全局一致性的高速公路三维数字模型。为了保持姿态估计的有限数量，文中还引入了基于关键帧的滑动窗口优化策略。最后，分别采集高速公路场景中常见的3种路段（一般路段、桥梁和隧道路段）进行建模分析，结果表明，在具有挑战性的高速公路场景建模中，文中方法能够有效提高建模鲁棒性、精度以及模型有效性。

The construction of 3D road digital model is of great significance for intelligent vehicle service and road management. In this paper, to solve the problems such as fast running speed, interference noise, few features and no loopback detection assistance existing in different sections of highway application scenarios, a three-dimension highway modeling method with lidar information as the modeling data base is proposed, in which multi-sensor fusion based on lidar odometry and LOAM technology is adopted. In the investigation, firstly, the point cloud data in different road scenarios are obtained by lidar, and the lidar image segmentation technique is used to assign each point a label about the structure and exclude the information of other moving vehicles on the road to reduce the modeling noise. Then, an accurate synchronization strategy is developed to integrate the sensors such as GNSS, IMU and lidar. On this basis, by combining the inertial navigation pre-integration results, the position constraint based on feature point cloud and the RTK data, a three-dimension highway digital model with global consistency is constructed to eliminate the cumulative error of the lidar odometry. Moreover, in order to maintain a finite number of attitude estimates, a sliding window optimization strategy based on key frames is introduced. Finally, three common road sections (general, bridge and tunnel) in the highway scenario are collected for modeling analysis, and the results show that the proposed approach can effectively improve the robustness, accuracy and validity in the challenging highway scenario modeling.