| Summary: | Vector maps are of great importance for both autonomous driving and robot path planning. Vector maps directly record the precise geometry of roads, boundaries, and other elements, which are important for path planning of robots and self-driving vehicles. By analyzing this precise geometric information, the system can quickly find optimal paths, avoid obstacles and navigate. However, current vector map construction methods rely heavily on manual post-adjustment, which greatly increases the workload. This dissertation focuses on generating outdoor vector maps for robot path planning using 3D point cloud data offline. The study involves processing point clouds through a sequence of steps. Firstly, traditional cloth simulation filter is applied for ground extraction. Next, kd-tree is used to search the nearby point around the ground, as these points can represent the part of the obstacles that we are really interested in. After that, Euclidean segmentation is used to get the cluster of the point cloud and then fast triangulation and B-spline are used to describe the point cloud outline. These processes enable the accurate capture of outdoor environments, which are essential for developing reliable maps that support autonomous robot navigation. By extracting key features from complex outdoor terrain, the proposed method enhances the precision and efficiency of robot path planning. Experimental results show that the vector maps generated offline by this workflow can effectively represent elements in the map over geometric shapes (e.g., points, lines, and polygons), providing a robust framework for future outdoor autonomous systems.
|
|---|