基于多约束因子图优化的无人车定位与建图方法

针对目前在特定场景下应用的低速无人车定位系统极度依赖全球导航卫星系统（GNSS），存在定位精度不高、漂移误差大、受环境影响严重等问题，提出一种低成本、高精度的无人车定位与建图方法。该方法基于三维激光定位与建图（SLAM）技术。首先，使用点云主成分分析（PCA）实现基于特征匹配的激光里程计；其次，将GNSS位置信息、点云分割聚类得到的地平面和点云聚类特征作为位姿约束分别加入图优化框架，消除激光里程计的累积误差；最后，得到最优位姿和大规模场景的点云地图，以实现无人车的自主定位导航。利用包含大型户外城市街道环境的KITTI数据集对所提出的SLAM算法进行了评估，结果表明:系统在3km运动距离情况下定位偏差可控制在1.5 m以下，在局部精度和全局一致性方面均优于其他里程计系统，为无人车的定位提供了新思路。 

Unmanned vehicle positioning and mapping method based on multi-constraint factor graph optimization

Aimed at the problem that the current low-speed positioning system of unmanned vehicle extremely relies on the Global Navigation Satellite System (GNSS), which has low positioning accuracy, large drift error and serious environmental impact, a low-cost and high-precision positioning and mapping method is proposed. This method is based on the three-dimensional laser Simultaneous Localization and Mapping(SLAM) technology. First, the point cloud Principal Component Analysis (PCA) is used to implement laser odometry based on feature matching. Then, the GNSS location information, ground plane and clustering feature of point cloud obtained by point cloud segmentation and clustering are added to the graph optimization framework as pose constraints, and the cumulative error of the laser odometry is eliminated. Finally, an optimal pose and large-scale scenes point cloud map is obtained to achieve the unmanned vehicles' position navigation. The proposed SLAM algorithm is evaluated using the KITTI dataset containing large outdoor urban street environments. The results show that the positioning deviation of this system can be controlled below 1.5 m at a movement distance of 3 km, and both in terms of local accuracy and global consistency, it is superior to other odometry systems and provides new ideas for the positioning of unmanned vehicles.