项目名称: 基于张量投票的车载LiDAR数据的目标识别
项目编号: No.41501501
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 天文学、地球科学
项目作者: 管海燕
作者单位: 南京信息工程大学
项目金额: 20万元
中文摘要: 从高精度、高分辨率的海量车载LiDAR点云中自动识别城市地物是激光LiDAR数据分析与应用中亟需解决的关键问题。为了进一步提高当前自动处理车载LiDAR海量点云数据的效率、性能和自动化目标识别的可靠性,本课题以基于张量投票的城市地物目标感知与识别方法的研究对象。研究内容包括(1)融合高程、强度以及其它属性信息,利用参数化主动轮廓模型实现一般道路的自动提取,并通过张量投票计算非道路激光点与其周围的数据点之间的局部信息传递,实现各种几何结构类型的描述;(2)深入研究张量投票的投票尺度对几何结构关系,构建多尺度张量投票模型,提高对地物目标进行几何结构识别和提取的鲁棒性;(3)结合地物目标的语义信息,优化地物目标分割结果。本课题以感知编组理论为基础,结合计算机视觉、数值分析、计算几何等理论与方法,探讨一种有效的基于点云的地物目标识别策略,希望能够为改善大数据量点云数据目标识别与解译作出贡献。
中文关键词: 地面激光雷达;点云滤波;点云分类;三维重建
英文摘要: Automatically identifying and classifying urban objects of interest from mobile laser scanning (MLS) data is critical to interpret and analyze MLS data for various applications. Compared to advanced MLS technology, MLS data post-processing methods, including object recognition and three-dimensional reconstruction, are still in the early period. This study presents a tensor voting-based urban object classification strategy, in terms of computational efficiency, robustness, cost-effectiveness, and reliability. The strategy includes the following key components: (1) With data attributes, including height, intensity, and wavelength,a snake based road extraction is proposed; (2) Tensor voting, a perceptual grouping method, is explored to extract geometrical primitives from the extracted non-road data; and (3) Sematic information of the objects of interest is combined with the geometrical primitives for refining object classification results. The proposed strategy is expected to improve the quality and efficiency of interpreting and extracting objects from massive MLS data.
英文关键词: mobile lidar;filtering;classification;3D reconstruction