LiDAR-based HD Map Localization using Semantic Generalized ICP with Road Marking Detection

在GPS无法使用的情况下，对于自动驾驶来说，一个强大的环境感知和定位系统变得至关重要。本文开发了一种基于激光雷达的在线定位系统，将道路标线检测和高清地图注册相结合。在我们的系统中，提出了一种实时性能良好的道路标线检测方法，首先引入自适应分割技术来隔离与道路标线相关的高反射点，提高实时效率。然后，通过聚合历史激光雷达扫描数据，形成了一个时空概率局部地图，提供了一个密集的点云。最后，生成了一个激光雷达鸟瞰图像，并应用实例分割网络来准确标记道路标线。对于道路标线注册，设计了一个语义广义迭代最近点（SG-ICP）算法。将线性道路标线建模为嵌入2D空间的1流形，缓解了沿线方向的约束影响，解决了欠约束问题，并在高清地图上实现了比ICP更高的定位精度。在真实世界的场景下进行了大量实验，证明了我们系统的有效性和鲁棒性。

LiDAR-based online localization system for autonomous driving in GPS-denied scenarios

The proposed system incorporates road marking detection and registration on a high-definition map, using an adaptive segmentation technique and a spatio-temporal probabilistic local map to enhance real-time efficiency and accuracy.

The system uses a LiDAR bird's-eye view image and an instance segmentation network to accurately label road markings, and a semantic generalized iterative closest point algorithm for road marking registration. Extensive experiments in real-world scenarios demonstrate the effectiveness and robustness of the system.

Related work includes research on LiDAR-based localization systems, road marking detection and registration, and autonomous driving in GPS-denied scenarios. Some relevant papers include 'DeepLiDAR: Deep Surface Normal Guided Depth Prediction for Outdoor Scene from Sparse LiDAR Data' and 'Robust Road Marking Detection and Recognition Using Deep Learning and Geometric Constraints'.