空載光達點雲過濾與檢核機制之評估

An Inspection Approach for Airborne LiDAR Data Filtering And Qualification

郭庭瑋(Ting-wei Kuo)

指導教授 ： 張哲豪

國立臺北科技大學/工程學院/土木與防災研究所/碩士(2013年)

https://doi.org/10.6841/NTUT.2013.00655

摘要

空載光達利用雷射多重反射回波的原理，在極短時間獲取地表與覆蓋物的高解析度三維空間坐標，形成點雲資料。透過對點雲資料的過濾，可將建築、橋梁、植被等地表覆蓋物分類為非地面點，同時判釋出點雲中適當點位為地面點，進而產製數值地表模型(DSM)及數值高程模型(DEM)。根據目前實務，多以點雲與DEM兩階段檢核。點雲檢核的方式，主要以點雲、航照圖、粗略地形模型等綜合資訊進行地形判識，以判定點雲分類正確與否。但點雲檢核工作傾向主觀認定，同一區域由不同檢核人員作點雲判定，往往會得到不同結果。DEM的檢核方法，主要是以地形類別、地表植被覆蓋情形即地表植被平均高等資訊進行DEM判釋，而判釋已有量化的標準。地面點分類的成果會直接產製DEM品質。根據現行規範表示，不同坡度點雲密度已達每平方公尺1.5至2點，平面密度非常高，縱使地面點分類略有缺漏的情況，最終DEM產品仍保持相當精度。綜合上述，本研究擬建立品質評估作業流程採用現行DEM規範來推估合理分類正確率，以在既有DEM規範下，達到相當的分類成果作為研究主要之目的。研究參考此品質評估作業流程，可將實驗分為內部品質評估並以此評估方法得到經驗精度，再以外部相對誤差評估方式，利用參考值驗證經驗精度合理性。
本研究針對各階段點雲分類結果對於DEM影響進行分析。實驗採取用100，經過濾前後兩階段點雲，將此內插為DEM，並檢驗之間高程差對於DEM容許誤差影響。統計實驗成果可得經驗精度，當點雲分類影響已小於經驗精度，便判定往後過濾結果已對DEM影響不顯著，便能停止過濾。為提升經驗精度可靠度，實驗利用參考值與通過經驗精度資料，進行誤差矩陣分析與高程比對方式，驗證經驗精度之可靠度，參考值驗證結果顯是合格率達到百分之百。此成果可供檢核參考之依據，並藉此提高作業效率。

關鍵字

空載光達 ； 點雲過濾 ； 品質評估

並列摘要

Airborne LIDAR captures the high-resolution 3D spatial coordinates of the earth’s surface and coverings and forms point cloud data in a very short period of time by illuminating targets with a laser and analyzing reflected and multiple echoes. By filtering the point cloud data, surface coverings such as buildings, bridges and vegetation can then be classified as non-ground points while appropriate points in the point cloud can be interpreted as ground points, thereby generating digital surface models (DSM) and digital elevation models (DEM). In current practices, two-stage validation that includes point cloud and DEM is the most common. Point cloud validation interprets terrains mainly based on collective data from point cloud, aerial photos, and rough terrain models to determine the accuracy of point cloud classification. However, point cloud validation tends to be subjective because results often vary when different people conduct point cloud validation on the same area. DEM validation interprets using DEM mainly based on terrain categories, coverage of the earth’s ground by vegetation, i.e. the average elevation of vegetation coverage on the surface, and there are already quantification standards for such interpretation. Results of ground point classification are directly linked to the quality of DEM products. According to current specifications, point cloud density for different slopes can reach 1.5 to 2 points per square meter, so final DEM products remain fairly accurate even if ground point classification has slight gaps. In view of the foregoing, the intent of this study is to establish a quality assessment process that estimates the reasonable classification accuracy in accordance with current DEM specifications, based on which considerable classification achievements can be expected. Based on this quality assessment process, our experiment consists of internal quality assessment, by which empirical accuracy is obtained, and external assessment of relative errors, which uses reference values to verify the rationality of the empirical accuracy.

並列關鍵字

Airborne LiDAR ； Filtering ； Qualification

參考文獻

[9] 黃俊翔，空載光達於內插方法與不同地形類別精度分析，碩士，成功大學，2012。

[4] 劉囿維，應用空載光達資料產製數高程模型之品質評估，碩士，國立成功大學，台南，2010。

[1] 何心瑜，空載光達作業流程及品質管理之研究，碩士，國立交通大學，新竹，2006。

[3] 何心瑜，空載光達作業流程及品質管理之研究，副研究員，工研院能環所，2008。

[12] Fenstermaker. L. K., "Remote sensing thematic accuracy assessment: A compendium", American Society for Photogrammetry and Remote Sensing, Bethesda, Md. ,2002, pp.413.