| 摘要: | 影像匹配為一廣泛應用於數位航測之影像處理技術。該技術藉由半自動或自動化程序匹配影像中對應目標位置，建立影像間幾何關聯性，其應用面涵蓋幾何校正、影像拼接與模型重建等。於航測幾何處理中，易受影像紋理特徵明銳強度所影響，故當遭遇遮蔽區或灰階均調區時，易存在匹配錯誤。因此，為提升匹配之可靠度，本研究針對高重疊航測影像建立多向核線影像匹配技術，同時考量幾何與紋理特性，產生正確匹配成果，並應用於房屋模型重建。主要工作包含特徵萃取、核幾何建立、影像匹配、模型重建與紋理貼圖。本研究時程共計三年，其中，各年度工作目標包含(1) 結合單向高重疊航照影像與大比例尺向量圖重建三維房屋模型，(2) 整合雙向高重疊航照影像與大比例尺向量圖重建三維房屋模型，(3) 整合三維房屋模型與高重疊近景影像敷貼牆面紋理。各年度工作內容包含，第一與第二年均以二維房屋向量圖外部輪廓線輔助提供初始目標位置，其中，第一年使用影像匹配技術與單向多核線幾何約制，使用單向高重疊航照影像尋找共軛目標，建立房屋模型；第二年以多向核線約制與影像匹配技術，利用多向高重疊航照影像重建房屋模型，並評估該項技術應用於近景影像處理之可行性。最後，第三年藉由前述處理技術與重建成果，進行牆面紋理偵測、遮蔽區分離與牆面紋理補償，以產製完整牆面紋理並敷貼至三維房屋模型。研究成果為使用航照影像與近景影像重建數碼城市，研究評估以人工處理成果為驗證參考。 Image matching is a powerful technique that builds relationship between images to find conjugate points. Matching results may be used for geometric correction, image mosaicking, object modeling, etc. However, the complex textures may lead to ambiguous results in digital photogrammetry. This investigation, thus, uses multi-epipolar image matching to find the reliable conjugate points. Those matching results are also used to reconstruct building models from highly overlapped aerial images. The major works contain feature extraction, restitution of epipolar geometry, image matching, object reconstruction, and texture mapping. This study conducts for three years. The milestone for each year is (1) 3D building modeling by the integration of highly overlapped single-strip aerial photos and large-scale vector maps, (2) employing highly overlapped multi-strip aerial photos and large-scale vector maps for building reconstruction, and (3) texture mapping for the reconstructed building models using highly overlapped close range images. The planimetric position of each building outline is extracted from the vector maps. In the first year, the image matching locates the conjugate points from single-strip aerial photos. The building boundaries and epipolar lines help to refine the matching results. Then, building models can be generated by the 3D points and edges. In the second year, the investigation reconstructs building models using multiple epipolar constraints for multi-strip aerial photos. We will also analyze the feasibility of texture mapping using multi-overlapped close range images. In the third year, we employ the reconstructed building models to perform façade texture mapping. Those unwanted objects in front of walls would be detected, removed, and compensated. The validation will compare experimental results with the reference data that will be measured manually. 研究期間：9908 ~ 10007 |
