以幾何特徵為基礎之光達點雲套合

Feature-based Registration of LiDAR Point Clouds

莊子毅(Tzu-Yi Chuang)

指導教授 ： 趙鍵哲

國立臺灣大學/工學院/土木工程學研究所/博士(2012年)

https://doi.org/10.6342/NTU.2012.01756

摘要

光達快速獲取高密度三維點雲之效益使其於數位城市建模、環境與工程監檢
測以及古蹟文物保存等應用受到廣泛關注，而光達點雲套合為點雲資料處理程序
中重要環節之一。就地面光達系統而言，為描述全部場景資訊往往需在不同視點
進行多測站掃描，並將不同測站之點雲資料套合至相同坐標基準；相對於已知方
位資訊的空載與車載系統，套合可視為不同航線間之最佳擬合程序以消除相鄰資
料間的微小偏差。此外，不同光達平台的資料整合亦在各相關領域中具有實務上
的需求。有鑑於此，本研究基於點、直線與平面三種空間幾何元件，針對特徵式
點雲套合三個主要程序(特徵萃取、匹配以及轉換參數估計)分別提出多重特徵偵測
元、RSTG 匹配法以及多重特徵轉換模式，藉由不同特徵整合之效益，建構高度自
動化、適合各類型點雲套合任務且可進行多測站點雲同時套合之特徵式點雲套合
模組。
以幾何特徵為基礎之光達點雲資料套合必須倚賴方便、明確、高品質及可靠
的特徵資料，由於平面特徵具有顯著的幾何辨別性與簡單的數學解析式，相關研
究多針對平面特徵為處理對象。然而，無論以何種光達平台進行點雲收集，平面
特徵的分布幾何皆相對單調，往往難以滿足套合作業所需的穩固幾何轉換或以平
面交會產製角點與直線特徵等應用。本研究建構的多重特徵偵測元藉由影像處理
與群聚分析方法，搭配由粗到細的執行策略，可有效率地基於原始點雲品質獲取
精確的直線、平面以及點特徵。不僅免除套合實務上人工量測或佈設控制特徵之
需求，接續之RSTG 匹配法更可由萃取成果中獲取共軛特徵，並同時估計測站間
的近似轉換關係，提供非線性多特徵轉換參數估計使用，達到高度自動化點雲套
合成效。
環顧現有的點雲套合相關研究成果，鮮少具備整合點、線、面三種幾何特徵
以及能適用於不同類型點雲套合任務之自動化程序。而本研究除了對於各階段建
構之數學模式以模擬及實際資料進行驗證分析外，亦針對多重幾何特徵整合之效
益、參數估計平差模式以及套合基準之選取等因素，對於套合精度品質之影響進
行探討，其成果有助於對於點雲資料處理方式以及各式幾何特徵於套合作業之效
用更深入的了解。除此之外，各階段程序所發展之方法或工具不僅可獨立使用，
行使其個別功能，亦可作為它項應用之部分元件，具有高度的後續發展性。

關鍵字

光達 ； 幾何特徵 ； 萃取 ； 匹配 ； 套合

並列摘要

As a developing 3-D surface measurement and mapping technology, LiDAR (light detection and ranging) continues to attract research attention in many application areas. Registration of overlapping LiDAR point clouds is typically required and constitutes an essential data processing step to form a comprehensive 3-D scene. Point clouds being registered may well be strips from aerial LiDAR, routes from mobile LiDAR or individual static terrestrial scans. Indeed, the required registration may also be between different LiDAR platforms, aerial and terrestrial LiDAR for example. In general, registration can be carried out with the aid of artificial markers, utilizing surface matching approaches or employing geometric features implied within point clouds. Research about integrating multiple feature matching that provides choices as to the optimal configuration against geometric scene restrictions is currently limited, and integrated schemes which comprise the extraction of features, matching and estimation of transformation parameters without the aid of artificial markers and initial approximations have yet to be reported. Moreover, most existing registration methods are designed for a single type of LiDAR and do not fully support cross-platform registration.

This research proposed a feature-based approach for LiDAR point cloud registration, which can efficiently handle registration between point clouds from both single- and multi-platform LiDAR without the aid of either markers or the provision of initial approximations for 3-D transformation parameters. Depending upon the geometric characteristics of the point clouds to be registered, primary features, including points, lines and planes, are employed. Each feature type can be used either exclusively or in combined fashion. The proposed working scheme comprises three kernels, namely a feature extractor for feature acquisition, a RSTG approach as a feature matching technique and, finally, a robust transformation model for the estimation of transformation parameters. It should be noted that these three individual parts are not necessarily tied to each other, that is, each means included in the study can be working independently in the scope of this research or cooperated with other applications. Furthermore, this study investigated into the effects upon registration quality in terms of the multiple feature integration, adjustment models for the estimation of transformation parameters and the choice of a reference frame. It would give an in-depth exploration of the optimization of data processing as well as the benefits of multiple features in a registration task. This study has been proven accomplishing a comprehensive, highly flexible and effective point cloud registration working approach.

並列關鍵字

LiDAR ； Geo-feature ； Extraction ； Matching ； Registration

參考文獻

Fusiello, A., Castellani, U., Ronchetti, L. and Murino, V., 2002. Model Acquisition by Registration of Multiple Acoustic Range View. In: Proceeding of the European Conference on Computer Vision, pp. 805-822.

Gelfand, N., Mitra, N.J., Guibas, L.J. and Pottmann, H., 2005. Robust Global Registration. Eurographics Symposium on Geometry Processing, pp. 197-206.

Ackermann, F., 1984. Digital Image Correlation: Performance and Potential Application in Photogrammetry. Photogrammetric Record, 11(64):429-439.

Ackermann, F., 1999. Airborne Laser Scanning ‐ Present Status and Future Expectations. ISPRS Journal of Photogrammetry and Remote Sensing, 54(2‐3): 64-67.

Akca, D., 2003. Fully Automatic Registration of Laser Scanner Point Clouds. In: Optical 3-D Measurement Techniques VI, volume 1, Zurich, Switzerland, pp. 330-337.