基于车载激光扫描数据的高速公路道路要素提取方法研究

发布时间：2018-08-17 14:30

【摘要】：随着智慧城市的飞速发展,市场对数据采集设备从采集速度、数据精度、处理效率、受干扰因素等方面提出了新的要求,车载激光扫描技术因此成为人们研究的热点。车载激光扫描技术具有非接触、高精度、高速度、大范围、成果直观的特点,已经开始应用在城市规划、道路铁路检查等领域。虽然车载激光扫描技术在理论上能提供足够的商业价值,但因其成本高昂,数据成果巨大、数据分析和应用效率低等因素,导致实际的利用率较低。本研究从市场需求出发,主要研究自动、高效、准确的从高速公路的车载激光扫描数据中提取出道路要素的方法,从而促进车载激光扫描技术的发展和普及。本文简要介绍了激光雷达技术的发展,以及在激光雷达技术基础上发展起来的车载激光扫描技术,总结了当前车载激光扫描技术中系统集成、成果数据组织与管理、成果数据分析和应用等研究热点的发展现状。在这些研究基础上,依托现有的设备资源,集成一套车载激光扫描系统,分析其工作原理和成果误差来源,并设计实验对造成误差的主要参数进行了标定,最后分析了整个系统的成果点云数据的精度。确定成果点云精度改善达到具体的应用要求后,采集高速公路的车载激光系统点云数据,对数据进行分析,提出一套自动、高效、准确的提取高速公路道路要素点云的新方法,为实现高速公路规划和养护、公路路面和附属设施普查提供便利。本文提出的高速公路道路要素提取方法,简述如下:对于路面的提取,首先通过点采集时间间隔提取每条扫描线,基于车载激光扫描系统的集成原理得到扫描线上各点采集时刻的扫描仪发射中心投影下来正对的路面点坐标,称为路面轨迹点。通过分析扫描线上各点与对应采集时刻的路面轨迹点坐标之间的斜率提取出最靠近车载激光扫描系统的路面点,称为扫描线上的路面起始点。按点采集顺序分析从起始点开始扫计算描线上各相邻点的斜率和斜率差,查找到显著变化的位置,认为是路面边界点疑惑点,通过分析前后点斜率和斜率差的方法进行确认,最终提取出扫描线上的路面边界点。将扫描线上路面起始点和边界点之间的点作为最终的路面点云。对于道路标线的提取,是对已经提取出来的路面点云的每条扫描线,通过反射率滤波进行初提取,再通过基于4倍点间距的三维格网划分去除离群点的方法进行提取的。对于护栏的提取,首先利用扫描线上的路面边界点与护栏相邻的特点初步分割出带有杂点的护栏点云,由于护栏点云每条扫描线上的主体目标比较集中、杂点相对离散,采用投影到不同高度水平面上提取点状和线状密集区域的方式分别提取出护栏的支撑主柱和波形护栏板点云,最后合并一处。对于路灯的提取,首先利用扫描线上的路面边界点与路灯相邻的特点初步分割出带有杂点的路灯杆状部分点云,再通过提取路面上方空间点云的方法提取出路灯的灯头部分点云,合并一处作为路灯点云初步分割的结果。基于初步分割出的结果中主体目标分布相对独立的特点,采用聚类划分超级体素,计算体素特征并合并轴向相近的体素成为目标,通过统计目标所包含的体素特征选择同时包含水平面状和竖直杆状体素的目标为路灯。本文所提出的方法结合高速公路道路要素的分布特征对原始点云进行初步分割,减少了数据处理量。并且结合道路要素的几何特征对原始点云进行精细提取,保证了提取的准确度。经验证,方法简单、高效,极大提升了车载激光扫描系统的实用性。
[Abstract]:With the rapid development of Smart City, the market has put forward new requirements for data acquisition equipment in terms of acquisition speed, data accuracy, processing efficiency, interference factors and so on. Vehicle-borne laser scanning technology has become a hot research topic. It has been used in urban planning, road and railway inspection and other fields. Although in theory, laser scanning technology can provide enough commercial value, but because of its high cost, huge data results, low data analysis and application efficiency and other factors, resulting in low actual utilization rate. This paper briefly introduces the development and popularization of the laser scanning technology for vehicles, and the laser scanning technology for vehicles developed on the basis of the laser radar technology, and summarizes the current vehicle laser scanning technology. Based on these researches and relying on the existing equipment resources, a set of vehicle laser scanning system is integrated, its working principle and error sources are analyzed, and the main parameters causing errors are designed and tested. Finally, the accuracy of the point cloud data of the whole system is analyzed. After the improvement of the point cloud accuracy reaches the specific application requirements, the point cloud data of the laser system on the highway are collected and analyzed. A new method of automatic, efficient and accurate extraction of the point cloud of highway road elements is proposed to achieve high accuracy. Highway planning and maintenance, highway pavement and ancillary facilities survey to facilitate. The highway road elements extraction method proposed in this paper, briefly described as follows: For the extraction of the road surface, first through the point acquisition time interval extraction of each scan line, based on the integration principle of the vehicle laser scanning system to obtain the scan line of each point acquisition time. By analyzing the slope between the points on the scanning line and the coordinates of the points on the corresponding collection time, the pavement point nearest to the vehicle laser scanning system is extracted, which is called the pavement starting point on the scanning line. The gradient and slope difference of adjacent points on the scanning line are calculated by starting sweep to find out the position of obvious change, which is regarded as the doubtful point of the pavement boundary point. The pavement boundary points on the scanning line are finally extracted by analyzing the slope and slope difference between the front and back points. Pavement point clouds. For the extraction of road markings, each scan line of the extracted pavement point clouds is extracted by reflectivity filtering, and then the outliers are removed by three-dimensional mesh partition based on four-fold spacing. For the extraction of guardrail, the boundary points and guardrail on the scanning line are first used. Because the main targets on each scan line of the guardrail point cloud are relatively concentrated and the clutter is relatively discrete, the point cloud and the wavy guardrail point cloud of the guardrail are extracted respectively by projecting to the horizontal plane of different heights to extract the point cloud and the linear dense area, and finally merge into one. For the extraction of street lamp, firstly, the pole-shaped part of the street lamp point cloud with miscellaneous points is preliminarily segmented by using the characteristics of the boundary points on the scanning line adjacent to the street lamp. Then the point cloud at the head of the street lamp is extracted by the method of extracting the spatial point cloud above the road surface, and one point is merged as the preliminary segmentation result of the street lamp point cloud. In the cut-out result, the distribution of the main object is relatively independent. Clustering is used to divide the super voxels, computing the voxel features and merging the voxels with similar axial direction to become the target. The target containing both horizontal and vertical bars is selected as the street lamp by statistical voxel features. The distribution characteristics of highway elements can segment the original point cloud and reduce the amount of data processing, and extract the original point cloud precisely according to the geometric features of highway elements to ensure the accuracy of extraction.
【学位授予单位】：北京建筑大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495;P225.2

【参考文献】