無人機攝影測量於陡峭崩塌地之應用

Abstract

山崩是形塑地表的重要作用之一，也是常見的自然災害，因此監測山崩非常重要。傳統上，在大地測量和地工技術的協助下，可透過多時期的觀測了解崩塌行為。近年來，無人飛行載具和運動回復結構的結合已成為地面測量的方式，透過其省力及低成本的操作流程，獲取地物的詳細表面資訊。該方法逐漸用於崩塌地的調查上，同時部分研究依攝影測量原理試驗不同的測量設計以提升量測品質。然而少有研究專注在陡峭崩塌地上，此類崩塌地常出現在活躍造山帶上，並且有較高的致災潛力，也是台灣地區常見之崩塌地型態。本研究目標為使用無人機與運動回復結構，發展一套低成本且有效偵測陡峭崩塌地細部變化的方法。本研究檢驗無人機的定位設備精度、航帶形狀、以及地面控制點分布對運動回復結構產製之模型的影響；另測試點雲密度、影像網型、以及地面控制點分布對變遷偵測的影響。本研究區位於台20線177公里處，包含兩個坡陡、表面不穩定、以及現場作業空間受限的崩塌地。本研究流程包含野外調查、運動回復結構、準確度評估、變遷偵測、以及體積估算。測試結果顯示在有地面控制點的情況下，無人機的定位設備精度非影響模型準確度的關鍵，航帶形狀亦無顯著影響，而地面控制點的分佈是影響模型準確度的關鍵。另外，點雲密度對偵測崩塌地表面變化無顯著影響，而影像網型對變遷偵測的影響非常明顯；在相對誤差可以保證的情況下，地面控制點的分佈並不是影響變遷偵測關鍵。

Landslides are one of the fundamental processes shaping the Earth’s surface and one major natural hazard, so it is important to monitor landslides. Traditionally, with the assistance of terrestrial geodesy and geotechnical methods, one can understand landslide behaviors via multi-temporal observations. In recent years, the combination of UAV and SfM have been a topographical survey method to acquire detail 3D morphological information of landforms for its labor-saving and low-budget operating procedure. The method has been progressively applied in landslide surveys, and some studies have experimented with various survey designs to improve the quality of the UAV-SfM survey according to photogrammetry principles. However, few studies focus on steep landslides, which commonly occur in active mountain ranges with high hazardous potential and a common landslide type in Taiwan. Therefore, this study aims to use UAV-SfM workflow to develop a low-cost, efficient methodology to detect a detailed morphological change of steep landslides. This study examines how UAV's positioning equipment, shapes of flight strips, and GCP geometries affect the SfM-derived 3D models’ accuracy. Also, it examines how point cloud density, image block, and the GCP geometries affect morphological change detection. The study area is located at 177 km of the provincial highway No. 20, and there are two landslides with high steepness, dynamic surface, and limited space for fieldwork. The whole workflow includes field survey, the SfM process, accuracy assessment, change detection, and estimation of volume change. The results of the examination show that the UAV positioning equipment is not the key to the model’s accuracy when GCPs are available. The shapes of flight strips have no significant impact. The GCP geometry is a decisive factor in the model’s accuracy. Also, the point cloud density is not significantly influential in the morphological change detection of the landslides. The image blocks affect the performance of the volume change detection very obviously. When the comparative accuracy can be guaranteed, the GCP geometry becomes less important for volume change detection.