三維地理資訊系統因環境變遷，資料定期更新維護屬必要。考量成本及效率，相較於模型全部重建，僅針對變遷的部分做更新為較佳的策略。影像應用於變遷偵測之首要工作為方位重建，其所需之控制資料以點與線為主。未變遷的三維向量模型可用來做為線控制資料以重建方位。著眼於此，本研究使用影像及地物向量資料，經影像及模型間的套合求解影像方位。 本研究內容可分為兩個主要部分：(1) 線控制方位求解數學模式建立，考慮透鏡畸變對線控制影響，結合影像直線參數式及共線條件式比較分析從原始影像萃取特徵及從透鏡畸變校正後影像萃取特徵之空間後方交會兩種求解模式。(2) 特徵自動套合方位重建，採方位初步修正及方位精密求解兩階段式套合，先利用四邊形的幾何特徵將初始外方位修正，再利用多條直線控制精化方位。 實驗資料有模擬資料及真實資料兩種。模擬資料用於測試透鏡畸變對線控制之影響。真實資料包含近垂直攝影影像、傾斜攝影影像及近景室內攝影影像，用於分析控制特徵形式與數量對方位求解的影響及特徵自動化套合方位重建。 實驗結果顯示，使用原始影像萃取特徵模式當透鏡畸變大時，成果誤差較大，使用透鏡畸變校正後影像萃取特徵可提升方位求解精度。所提出之線特徵自動套合策略共軛特徵選擇正確率為81.4 %。近垂直攝影及近景室內攝影測試例特徵幾何配置優良且控制線長，方位求解精度較佳。幾何配置較弱，且控制線段長度差異大的傾斜攝影測試例之方位求解誤差大。 ;Data updating and maintenance of three-dimensional Geographic Information Systems (GIS) are necessary due to environmental changes. Instead of reconstructing entire models, the more economic and efficient way of updating is to focus on the changed parts. The preprocessing of change detection using imagery is orientation modeling, the needed control information might be categorized as point-based control and line-based control. Unchanged three-dimensional vector data in GIS can be used as control data of orientation modeling. Therefore, this study registers images and object vector data for orientation determination. This study contains two major parts: (1) line-based orientation model construction and (2) feature registration. Considering the lens distortion effects of line controls in the first part, two space resection processes, Feature extraction from Raw Image (FRI) and Feature extraction from Compensate Image (FCI), are compared. For the second part of the study, the first step is to pull-in the initial exterior orientation parameters using the geometric information of quadrangle objects. The following process is to refine the orientation using more line controls. Experimental data includes simulated data and real ones. Simulated data used for analyzing lens distortion effects of line controls. Real data contains vertical, oblique and close range photographs. Real data were used for analyzing effects of control configurations and orientation modeling with automatic feature registration. Experimental results showed that the large lens distortion gave large error of FRI. FCI indicated better results than FRI. The proposed automatic feature registration procedure made 81.4 % correspondence feature selections correct. In the case of good configuration, including vertical and close range photographs, orientation modeling accuracy is high. On the other hand, the oblique photograph with weak configuration and shorter control lines yield lower accuracy.
