| 摘要: |
| 传统点云构网算法(泊松算法、Delaunay三角剖分算法、Graph Cut算法等)能够处理多种激光雷达获取的点云数据，通用性较强，但都需要去噪、降采样、法向量提取等预处理，其实，在已知轮廓和方位角信息的情况下，可以充分利用已知信息进行直接构网。针对上述情况，提出了一种基于点云初始信息的单线激光雷达旋转扫描快速空间构网方法，该方法利用激光雷达旋转扫描形成的轮廓信息以及轮廓中每个点的方位角信息，将两条相邻轮廓上的点云按照方位角顺序进行构网，当某方位角点缺失时两条轮廓同时跳过该点，与下一个点构网，再将所有轮廓按照轮廓顺序依次构网并对缺失点产生的破面进行修补。实验结果表明所提方法在纸箱、房间及长廊中均能表现出较好的三维重建效果，平均误差SD(Standard Deviation)相比于传统方法更小，提高了三维重建的准确性，构网速度和鲁棒性也显著增加。 |
| 关键词: 激光雷达点云 方位角 轮廓 空间构网 |
| DOI： |
| 分类号:TN958.98 |
| 基金项目:国家自然科学基金项目（面上项目，重点项目，重大项目） |
|
| Rapid spatial mesh construction by rotational scanning with single-line LiDAR contour and azimuth information |
|
余继龙, 钟世龙
|
|
|
| Abstract: |
| Traditional point cloud construction algorithms such as Poisson algorithm, Delaunay triangulation algorithm, GC (Graph Cut) algorithm, etc., have strong universality and are suitable for processing point clouds obtained from various instruments. However, they all need to preprocess the point cloud, such as denoising, downsampling, normal vector extraction, etc. The algorithm complexity is high and the speed is slow. In response to the above problems, in the case of laser radar rotation scanning, the initial information (scanning contour, azimuth, etc.) collected by the laser radar can be fully utilized to directly construct the network. Therefore, a fast spatial construction method based on the initial information of the point cloud for single-line laser radar rotation scanning is proposed. This method uses the contour information formed by the laser radar rotation scanning and the azimuth information of each point in the contour, and directly constructs the network according to the azimuth order of the point cloud on the two adjacent contours. When a certain azimuth point is missing, the two contours skip this point at the same time, and then all the contours are constructed in the order of resolution and the broken surface caused by the missing point is repaired. Experimental results show that the proposed method can show good three-dimensional reconstruction effects in cartons, rooms and corridors. The average error SD (Standard Deviation) index is better than traditional methods, the average error is increased by more than 50%, and the network construction speed and robustness are also significantly increased |
| Key words: LiDAR Point Cloud Azimuth Contour Triangle mesh |