车载移动测量系统可采集高精度道路三维点云数据，为道路边界自动化提取提供了支撑。为解决车载激光点云中城市道路边界点云提取困难问题，本文引入局部二值模式LBP（Local Binary Pattern），针对各类城市道路边界特征，设计了高度LBP、高程离散度LBP和空间形状LBP 3种改进算子；构建多元LBP特征语义识别模型，实现了道路路缘石空间几何与分布特征的量化分析；最后通过道路方向约束进行聚类去噪，提取道路边界。对4种不同的城市路段点云进行实验，实验数据的提取完整率为92.0%，准确率为95.8%。结果表明，该方法可以准确地提取不同环境下的道路边界，具有较强的适应性。

As an advanced surveying and mapping system, vehicle-borne mobile mapping system has several advantages, such as high precision, high efficiency, active, and non-contact measurement. This system can quickly collect high-precision road 3D point clouds, which are important for road boundary automatic extraction, and has become important in road information acquisition and update.To address the difficult and inaccurate extraction of urban road boundary point clouds in vehicle-borne laser point clouds, this paper introduces the Local Binary Pattern (LBP), which is an efficient and effective image processing method, to automatic point cloud classification. First, to take full advantage of the characteristics of various urban road boundaries, three improved operators were developed, including height, elevation dispersion, and spatial shape LBPs, which make full use of the three-dimensional shape, spatial geometry, and distribution characteristics of curbs. Statistical analysis was also performed on the feature image pixel values of the three LBP improvement operators. The statistical results are consistent with the spatial distribution and geometric characteristics of different objects, such as road boundary and road surface. Then, a diverse LBP features semantic recognition model, which can realize the quantitative expression of the spatial geometry and distribution characteristics curbs and pavements, was built. Finally, the road boundary point clouds were extracted by cluster and denoised with the road direction as the constraint.The point clouds of four different urban sections were tested. Results show that the extraction completeness rate of the experimental data is 92.0%. The method we developed can extract the main road and sidewalk boundary point clouds under different road environments. In terms of accuracy, 95.8% accuracy was achieved from considering the spatial distribution and geometrical characteristics of the curb. The results indicate that our method can accurately extract road boundaries in different environments and has strong adaptability.