Biography
Enrollment Date: 2011
Graduation Date：2014
Degree：M.S.
Defense Date：2014.05.27
Advisors：Xiang Xie
Department：Institute of Microelectronics，Tsinghua University
Title of Dissertation/Thesis：Segmentation method for sparse 3D point cloud based on mobile mapping system
Abstract:
With the rapid development of science and the improvement of living standard, people are not just satisfied with models in 2D, the images. Owing to the accuracy and integrity, the 3D reconstruction exists as the one of the most rapidly developing fields. Especially the large-scale 3D reconstruction, which can be widely used in urban planning and digital city mapping, is definitely the research hotspot in recent years. During the point cloud processing, segmentation acts as the prerequisite and foundation for the reconstruction. By the points representing different objects being separated, the reconstruction of the objects further gets studied.
The point cloud data are processed as unorganized in the existing segmentation methods, while the feature of order is neglected. This paper proposes an efficient two-step segmentation method for large-scale 3-D point cloud data collected by the mobile laser scanners. First, a new scan-line-based ground segmentation algorithm is designed to filter the points corresponding to the ground with high accuracy. Second, we propose a self-adaptive Euclidean clustering algorithm to further separate the off-ground points corresponding to different objects. The main innovation points of this dissertation are as follows.
In the scan-line-based ground segmentation algorithm, it utilizes the inherent ordering of LiDAR scan lines. And since the ground is always lower than other objects and its points are continuous in the outdoor environment, our algorithm utilizes the gradient and elevation as the judgment criterion to detect the ground sequences. Each scan line is processed independently. What’s more, it is effective for ground with slopes and bumpy roads without the need for assuming that the ground is located in the lowest. Such scenes are hard, if not impossible, to solve using the previous methods. In experiments in different scenes, the method runs at an error rate of 0.674% and a computing throughput of over 20 million points/sec.
In the cluster-scale based self-adaptive threshold Euclidean clustering algorithm, the self-adaptive threshold is designed by the area of cluster and utilized to further segmentation. In the proposed system based self-adaptive threshold Euclidean clustering algorithm, by modeling the process of collecting data in the mobile mapping system, we derive the distance between any point and its neighbor points. The distance, i.e. the threshold of the point, is described by the parameters including the distance from LIDAR, the velocity, the acquisition frequency and LiDAR’s angular resolution. Based on the segmentation result of ground, the objects which are floating can be detected and combined to solve the over-segmentation problems. Also the large-sized clusters are combined to avoid the over-segmentation due to occlusion or sudden change of the scanning direction.