空載光達資料中地面點選取及房屋偵測

Ground Point Selection and Building Detection from Airborne LiDAR Data

邵怡誠(Yi-chen Shao)

指導教授 ： 陳良健

國立中央大學/工學院/土木工程學系/博士(2007年)

摘要

本文旨在研究自空載光達資料中選取地面點與房屋偵測。在第一個主題中，先陳述空載光達的系統、誤差來源及資料特徵，回顧並比較數種主要的過濾方法及其特性。本文在考量地形的高度、斜率及區域特徵後，提出一個創新的斜率式「爬升及滑行演算法」來選取光達資料中的地面點，該法具有區部搜尋又能涵蓋全區的優點。本文為提昇處理的效率及資料精度，建立一個在虛擬網格式初始表面模型上選取地面點的架構，並配合一個再搜尋步驟來獲取更細節的地形特徵。其中考量橋面點易遭誤判為地面的一部份，也增加了一個偵測並除橋面點的步驟。
第二個主題是自地面以上的點群中偵測房屋區塊。首先依據房屋的高度、尺寸及面積等先驗知識偵測房屋候選區，此時候選區塊主要為房屋及樹木二大類，隨後基於屋頂面具有斜率連續特性的假設，分析10個以斜率差為基礎的區塊紋理，最後採監督式及監督式分類來偵測房屋區。
本文使用二組資料測試驗證選取地面點的架構是否有效。第一組資料是取自國際航遙測學會的第三工作群第三小組，第二組資料則涵蓋了南台灣數種不同地形特徵。文中評估了選取地面點的錯誤量及參數的敏感度，也與數個知名的過濾法比較成果，同時也分析了地形特徵的保留程度。實驗成果顯示該處理架構可適用於多種不同地形特徵。為驗證房屋偵測架構的有效性，文中同時使用在台灣的都市及鄉村二個實驗區。實驗成果顯示，均調及微小斜率差機率等二個紋理特徵均可適用於房屋偵測。

關鍵字

分類 ； 房屋偵測 ； 數值高程模型 ； 雷射掃描 ； 空載光達

並列摘要

In the dissertation, ground points are selected and building regions are detected from airborne Light Detection and Ranging (LiDAR) data. In the first part of the paper the system, error sources, and the data features of the airborne LiDAR are described after which several major filtering algorithms and their characteristics are reviewed and compared. A novel slope-based climbing-and-sliding (CAS) method is developed to select ground points from airborne LIDAR data which takes into consideration the features of height, slope and area of the region of bare earth. In the proposed method not only is a local search performed but the merits of a global treatment are preserved. A scheme is proposed to improve the efficiency and accuracy where the ground points in the initial surface model are selected based on a novel pseudo-grid. After this a back selection step is performed to retrieve detailed terrain features. Considering that bridges have tended to be misclassified as parts of the ground, an additional detection step is included to remove bridge points.
In the second part of the dissertation, building regions in the set of above-ground points are detected. Prior knowledge of such things as the height, size, and area of the buildings, is employed to first remove extraneous points or regions and to detect building candidates. Building and tree are two main dominate classes in the candidates. Based on the assumption that buildings roofs are piece-wise continuous, ten region-based textures based on directional slope difference are analyzed. At the last, an unsupervised classification is performed for the building detection.
The filtering error of the generated DEM is evaluated, as well as the test of parameter sensitivity. The processing results are quantitatively compared with several recognized counterparts in the literature. The presentation of the terrain features is also analyzed. The experimental results demonstrate that the proposed scheme can be applied to diverse terrain types. To validate the detection scheme, two data sets including urban and rural areas in Taiwan are tested. The experimental results indicate that two features of homogeneity and probability of a small slope difference preserve most information and thus are most suitable for the detection.

並列關鍵字

laser scanning ； airboren LiDAR ； classification ； building detection ； DEM

參考文獻

[1] Ackermann, F., “Airborne laser scanning — present status and future expectations”, ISPRS Journal of Photogrammetry & Remote Sensing, Vol. 54, pp. 64–67, 1999.

[5] Baltsavias, E. P., “A comparison between photogrammetry and laser scanning”, ISPRS Journal of Photogrammetry & Remote Sensing, Vol. 54, pp. 83–94, 1999a.

[6] Huising, E. J. and L. M. Gomes Pereira, “Errors and accuracy estimates of laser data acquired by various laser scanning systems for topographic application”, ISPRS Journal of Photogrammetry & Remote Sensing, Vol. 53, No. 5, pp. 245–261, 1998.

[8] Sithole, G. and G. Vosselman, “Experimental comparison of filter algorithms for bare-Earth extraction from airborne laser scanning point clouds”, ISPRS Journal of Photogrammetry & Remote Sensing, Vol. 59, p. 85–101, 2004.

[10] Henricsson, O., “The role of color attributes and similarity grouping in 3-d building reconstruction”, Computer Vision and Image Understanding, Vol. 72, pp. 163–184, 1998.