Automated Extraction Of Road Geometric Features From Mobile Laser Scanning Point Clouds

Recent advances in laser scanning technology have led to the integration of laser scanners, navigation sensors (typically including Global Navigation Satellite Systems and Inertial Measurement Unit), and other data acquisition sensors (e.g., digital cameras) with mobile mapping platforms. Mobile laser scanning (MLS) has become a cost-effective solution for capturing very dense point clouds along road corridors along large areas. Thus, the MLS data is widely used in numerous applications such as smart city,3D city modeling, highway construction, and civil engineering.Compared to advances in MLS hardware, MLS software and automated algorithms, especially the development of road environment interpretation algorithms, which are used for efficiently extracting3D curb and road marking of interest from point clouds, is relatively slow. This is because in urban street scenes, captured point clouds contain various types of objects, such as pedestrians, cars, poles, buildings, roads, etc. Additionally, holes often appear in the captured point clouds because of the occlusion of various objects. Meanwhile, compared with the higher objects such as buildings, trees, curbs and road markings with smaller size are not clear in term of shape, location and topology in the complexity of urban street scenes.Hence, this dissertation aims to fulfill three tasks, namely, automated detection candidate road regions from mobile laser scanning point clouds; automated extraction and delineation of3D road borders of street scenes from candidate road areas; automated extraction road markings from candidate road areas, In detail, the main research works of the dissertation are as follows:1) In the view of the road geometric features extraction from mobile laser scanning point clouds, the research background of the dissertation is introduced. We start from the review of the development history, and the constituent parts of the mobile lidar systems. Then some popular commercial mobile lidar systems are introduced. For the raw point clouds, we analysis the difficulties of extracting the geometric features of road in term of huge data, space distribution, the complex street scene and the completeness of data.2) The background and methods of extracting the road features from various geo-spatial data sources are reviewed in this chapter, In the research status of the road extraction from remote sensing images, some semi-automated or automated methods are introduced. Then we examined the studies of road extraction from raw airborne laser scanning data, or fused the auxiliary data such as GIS map or images and airborne laser scanning data. In the term of the mobile laser scanning data, many studies track the problem of road extraction based on generating points into images, scanning lines or cluster methods. The difficulties and prospects of road extraction from mobile laser scanning data are pointed out.3) In order to more cost-effective process massive point clouds over large areas, a novel point partition technique is proposed to divide point clouds into consecutive road cross sections according to GPS time or scanning angles of points. Candidate road regions are sorted by cross section for further road borders (curbs) and road markings detection using a moving window operator.4) In this chapter, we propose a high precision method for extracting3D roads borders from MLS point clouds. The proposed method accurately extracts roads borders by utilizing global road properties (topology and smoothness) and local road features (curb borders). In each road cross section, curbs are detected using three popular curb patterns that account for elevation difference, point density, and slope changes. Finally,3D road surfaces are refined and tracked by linear structure labelling and interpolation by Bezier. Compared with the previous methods, the proposed method can be used with large-scale point clouds without requiring a secondary dataset (e.g., images, GIS dataset), or point cloud color or intensity data. Additionally, the proposed method efficiently treats multiple types of road environments, including regular and irregular road shapes, which is critical for modeling urban environments.5) For point clouds contain a lot of implicit information besides the coordinates of laser points. Among them, the reflectance strength of laser point is closely related to the property of road surface materials, which helps detecting road markings. For the purpose of road marking extraction, we proposed a method to capture the goal in the dissertation. Firstly, multi-scale criterions are used to detect the road markings from the complexity street-scenes or high-way environments. By varying the diameters of the spheres, we monitor how the local cloud geometry behaves across scales to detect the road marking. In order to refine the results of road markings based on multi-scales criteria, then, the proposed method generates a geo-referenced reflectance strength image of road markings points, and labels the regions of road markings according to the shapes and arrangements of road markings. In order to improve the performance of road markings extraction, the proposed method incorporates the semantic knowledge (e.g., shape, pattern) of road markings, which is particularly helpful for extracting those road markings occluded or with incomplete shapes.6) Seven different datasets were selected to validate the performances of the proposed methods. Visual inspection and quantitative evaluation showed that the proposed methods are effective at extracting road borders and road markings from mobile laser scanning point clouds, even in complex urban street-scenes.