| Author: | Chen, Pengxin |
| Title: | Scan registration, mapping and semantic segmentation using mobile laser scanning |
| Advisors: | Shi, John (LSGI) |
| Degree: | Ph.D. |
| Year: | 2022 |
| Subject: | Surveying Lasers in surveying Digital mapping Mobile computing Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Land Surveying and Geo-Informatics |
| Pages: | xvi, 177 pages : color illustrations |
| Language: | English |
| Abstract: | Point clouds registration, mobile mapping and semantic segmentation are closely linked steps to build geometrically-accurate and semantically-rich 3D digital replicas using Mobile Laser Scanning (MLS). They not only are the fundamental work in surveying and mapping tasks, but also contribute infrastructure to many downstream applications such as navigation, augmented reality and autonomous driving. However, previous works show limited odometry/mapping drift control and segmentation accuracy. Given above, the research objective of this thesis is to develop novel methods for building point cloud maps using MLS, from four progressive perspectives: (1) scan-to-scan registration, (2) scan-to-map registration, (3) graph SLAM and (4) semantic segmentation on point clouds. From each of the perspectives, new algorithms are proposed to solve specific research problems and the performances are evaluated by rigorous experiments. For scan-to-scan registration, this thesis addresses reliable scan matching for MLS that can traverse different environments and be robust against outliers and noise. An expressive feature coined the rectangle-flattening representation is introduced to enhance the scan matching reliability in multiple scenarios. First, a clustering method is proposed based on density, direction and flattening. It allows regions to grow in a "planes first, lines second, less flattened structures last" manner. This method can extract rectangles from environments where planes are scarce. Second, a squared point-to-rectangle distance function which is piecewise yet continuously differentiable, is developed to leverage the rectangle-flattening representation for scan matching. Unlike the traditional point-to-plane or plane-to-plane residual functions that rely on planar surfaces in other directions to provide translational information, the proposed point-to-rectangle distance function is intrinsically translation-aware. For scan-to-map registration, this thesis proposes a framework for odometry, mapping and ground segmentation using a backpack LiDAR system that can achieve both real-time and low-drift performance. First, a spatio-temporal calibration method is presented to carefully merge scans from the two laser scanners on a backpack. Second, a point feature extraction method is developed. It generalizes a point's geometrical characteristics as two groups (disjoint, continuous) and three types (edge, corner, plane). The extracted features are used in scan-to-map registration. Third, a fast ground segmentation method is customized for the backpack LiDAR system. Taking advantage of a customized backpack multisensorial platform, a MLS dataset is collected to evaluate the proposed algorithms. Moreover, this dataset is further improved by incorporating other sources of data such as panoramic camera, GNSS and IMU. The dataset is made publicly available at https://github.com/chenpengxin/PolyU-BPCoMa.git. It is hoped that this dataset will be useful to the vision and surveying community, and be helpful for developing algorithms for SLAM, mobile mapping and colorization. For graph SLAM, this thesis proposes a novel mobile mapping framework, coined the hysteretic mapping, to control the overall mapping drift. Essentially, a feature equalizer is the key to balancing the inhomogeneous feature points of buffered scans in a sliding window and make them uniformly distributed to the best possible extent. The feature points after equalization are more friendly to the feature-to-map registration process, so that the overall odometry and mapping drift in a graph SLAM method can be alleviated. Moreover, a self-supervised motion prior generator network is designed. This network consumes images of point clouds after spherical projection, and the single-track inference will directly provide an initial guess for the registration process. For semantic segmentation, the prior knowledge of spatial and intensity information is embedded into a point cloud segmentation network for corridor environments. The proposed semantic modeling algorithm consists of three main modules: point feature extractor, spatial distribution prior and intensity distribution prior modules. Specifically, the point feature extractor functions as the network backbone that contributes to obtaining the basic predicting results. Then, the two well-designed prior-information based modules are utilized to refine the basic results so that better network performance can be acquired. Extensive experiments are conducted to evaluate the proposed algorithms both qualitatively and quantitatively. Experimental results are compared with baseline and state-of-the-art methods on multiple datasets to ensure well generalizability. The results prove that the proposed method is competitive to others in the way of accuracy and robustness. In summary, this thesis serves as a systematic study on the mapping and modeling using MLS. It starts with scan-to-scan/map alignment, further dives into graph SLAM, and finally ends with the corridor semantic segmentation. A customized backpack multisensorial platform is designed to validate the proposed algorithms and the dataset collected by the platform is contributed to the public. The presented methods, dataset, and benchmark cover a complete process of building digital replicas using MLS. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12083