基于移动激光扫描点云数据栅格化的交通标线分割算法

Traffic Marking Segmentation Algorithm Based on Rasterized Urban Mobile Laser Scanning Point Cloud Data

摘要:

针对目前交通标线提取研究对于语义信息丰富的标线精细化分割研究不足的现状， 通过移动激光扫描数据提出了一种精细化交通标线的语义分割算法。该算法先从道路点云中提取出路面，并根据点云的强度信息将其转变为栅格图像并对其优化，通过二值化处理提取出路面标线；在此基础上根据矩形度筛选将所有标线对象粗分为两类，并采用不同的语义分割策略进行识别，最后获得了10种符号标线以及非符号标线对象的精细化语义信息。实验选取了上海8组不同区域的城市道路点云数据进行验证，结果表明该算法具有96.04 %的精度与96.92 %的召回率，综合评价指标F达到96.48 %，为高精地图提供了更加丰富的交通标线语义信息。

Abstract:

At present， the research on traffic marking extraction is insufficient for refined segmentation of traffic markings with rich semantic information. This paper proposes an algorithm to refine traffic marking semantic segmentation based on mobile laser scanning. The algorithm first extracts road surface from the point cloud， then transforms the surface into a raster image according to intensity information， so as to extract traffic markings through binary processing， based on which， all traffic markings are roughly divided into two categories by rectangularity filtering， and different semantic segmentation strategies are applied for recognition. Finally， the refined semantic information of 10 kinds of symbolic and non-symbolic traffic markings is obtained. A study was conducted by 8 groups of urban road point cloud data acquired in Shanghai， China， whose results demonstrate that the proposed workflow and method can achieve a Precision， recall， and F of 96.04 %， 96.92 %， and 96.48 %， which provides more abundant semantic information for a high-precision map.