基于车载激光点云的道路标线提取方法研究

Study on Road Marking Extraction Method Based on Mobile LIDAR Point Cloud

车载激光扫描近年来被广泛应用于道路环境信息提取及道路三维模型重建方面等工作,然而由于测量数据中包含目标种类繁多、点云数据量大,仍然需要大量的人工处理步骤以实现道路标线等目标的提取,已有研究所提出的自动化目标提取算法在实用性上仍不成熟。基于车载激光点云,利用标线点云的空间分布特点及反射强度信息,结合行车轨迹及标线的位置、高程特征,提出了一种能快速提取道路标线的方法。首先基于点云反射强度指标建立Logistic回归模型,以选取合适的阈值进行标线点分类。其次,利用行车轨迹及标线点的位置、高程特征建立圆柱体搜索区域,并且进行定步长迭代式的标线点搜索。最后,对搜索结果进行分段式最小二乘法拟合并将其转换为CAD三维多段线,从而得到道路标线提取结果。该方法中采用的Logistic回归模型兼顾了标线点的提取率和误判率两大指标,在保证足够提取率的同时,使得误判率尽量低。所提出的定步长迭代式的搜索方法具有极强的适应性,在标线不清晰甚至较大面积缺失的情况下也能准确地提取出道路标线。以Topcon公司生产的IP-S2移动测绘系统为例的试验结果表明:该算法能高效、稳定地提取道路标线、提取精度,以符合工程实践需要。

Mobile LIDAR has been widely used in road environment information extraction and road 3 D modeling in recent years,but due to the complex targets and enormous volume of point cloud data in measuring data, a large amount of manual work is still required for extracting road marking and other targets, current research of target extraction automation algorithms are still immature in practicability. Based on mobile LIDAR point cloud as preliminary data, by using the spatial location property and the reflection intensity of road marking point cloud,combining with the location and height characters of mobile trajectory and road marking, a method for rapidly extracting road marking is proposed. First, a Logistic regression model based on point cloud reflection intensity is built to classify road marking points using a proper threshold. Then, a cylinder shaped searching area is built using the location and height characters of driving trajectory and marking points for searching the marking points with fixed-step-value iteration. Finally, the searching result is fitted by piecewise least squares method and converted into 3 D multi-segment CAD line, thus the road marking extraction result is obtained. This method gives consideration to both true positive rate and false positive rate of road marking points through Logistic regression, reducing false positive rate as much as possible while maintaining sufficient true positive rate. The proposed fixed-step-value iteration search method has strong adaptability, which can extract road markings correctly when road markings are not clear or even large area absent. The test result using Topcon IP-S2 mobile LIDAR device indicates that the method can extract road markings efficiently and stably with a relatively high accuracy that meets engineering requirement.