線與角點為建構二維面與三維面的基本元件，這些特徵的幾何資訊可供相關應用比較分析的依據，如資料套和、影像匹配、模型重建與變遷偵測等。既有研究主要藉由調整門檻與建立基本元件萃取特徵，然而，此舉需依目標複雜度經門檻測試以達到成果最佳化。由於線特徵具明確的幾何特性與特徵連續性，故本研究提出位相連續性法則分析局部差異性以萃取特徵，並維持特徵之連續性，此舉可取代門檻調整過程及提升處理自動化程度。本研究在影像處理中，使用單一固定門檻進行高斯平滑處理，此門檻為依據既有研究之經驗值予以設定，以同時萃取線與角點特徵。於光達資料處理中，本研究省略面偵測與面交會過程，直接使用單一固定高程差門檻從空載光達點雲中萃取三維線特徵，此門檻為依據最小牆高值給定。此外，本研究之處理目標亦包含屋頂女兒牆之線型。在驗證部分，萃取之特徵經人工產製的參考資料比對後，顯示本研究可經由較高之自動化處理過程獲取無失品質的特徵成果。; Edges and corners are important components in constructing planimetric planes and 3-dimensional facades. Feature geometry can also support related processes for advanced analyses including data registration, image matching, object modeling, and change detection. Conventional methods have focused on the threshold operation or primitive comparison to identify features. Because these predefined constraints may require optimization according to prior experience for various targets, however, a new, alternative for feature detection without threshold selection is needed. Because one edge should suffice for the specific geometry and connected elements, the proposed scheme in this study analyzed the local relief with topological connection criteria instead of threshold operation to improve the detection ability and automation. In image processing, one constant threshold was used for Gaussian smoothing (a method adopted in related works) to detect edges, and image corners were specifically addressed. In comparison, LIDAR (LIght Detection and Ranging) processing uses one fixed relief threshold to detect 3-dimensional lines, including parapets on rooftops, from airborne LIDAR data without surface determination and intersection. This relief threshold was related to the minimum wall height, and in the validation process the detected features were compared with references identified through visual perception. Based on these comparisons, the proposed scheme achieved higher automation without losing detection quality.
