- 學位論文
未排序近景影像之相對方位
Relative Orientation of Unordered Close-range Images
摘要
有別於絕對方位模式，影像相對方位之解算可不需引用或不全然藉助物空間控制資訊下進行，其計算所需之觀測量可藉由影像匹配技術以自動化或輔以人工量測之方式提供，而利用具備相對方位之影像，仍可行使在模型坐標系下之物空間三維資訊測繪，因此具有活絡的應用層面。 近景攝影時為使影像內容盡可能地涵蓋完整場景，必須挑選適當位置並從不同角度拍攝之，造成攝像幾何複雜以及影像內容成像比例尺變差大，因此建構方位解算之共軛對應量測以及方位初始值估計儼然成為一大挑戰。 有鑑於此，本研究以增進排列不規則或是缺乏序列資訊之近景影像在計算相對方位上之處理效能為目的，首先根據點位之分布以影像匹配技術約制保留或補充幾何上重要的特徵點位，依此建立以像對為單元之影像關係，並進一步擷取多重疊之共軛點，利用階層式結構串聯影像重疊關係並統一像對間之尺度及模型坐標系統，解算之策略上則結合線性系統及共線式完成影像之相對方位。從模擬及實際資料的實驗成果展示本研究所設計流程能發揮特徵式匹配在建構影像間連結關係的實用性，而特徵分布、多重疊點位及階層式處理考量及策略則能提升影像間連結幾何與物像對應品質。
並列摘要
The model of relative orientation of images is independent of the object coordinate system, which is different from absolute orientation. In addition, because the observations can be measured automatically or measured semi-automatically by using the technique of image matching, it has more flexibility than the model of absolute orientation. In the other side, people usually have to take the close-range images of some scene from appropriate positions and orientation in order to cover the scene. Because of the complexity of the image geometry and the diversity of the image scale, it’s a challenge to measure the correspondences and estimate the approximate value of relative orientation obviously. As a consequence, the goal in this research is to propose an appropriate procedure to construct the relative orientation of the close-range images, which is part of irregular image block or lack of the information of sequence. At the first we use the technique of image matching to supply the observation in the essential regions and construct the relation of images pair by pair. In addition, we determine the n-fold points and using the hierarchical structure of images to integrate the unified scale and coordinate system. Finally we combine the method of liner system and the bundle adjustment to complete the relative orientation of images. Both the simulated and experimental test results reveal that the proposed approaches are applicable, and the strategy about the distribution of feature points, determination of n-fold points and the hierarchical structure of images can promote the relation of images and the quality between images and object.
參考文獻
延伸閱讀
- 周彥榕（2012）。基於最小平方法與邊界前瞻技術之非失真影像序列編碼〔碩士論文，國立臺北科技大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0006-1908201208395100
- Yeh, C. T. (2015). 未經相機校準下之立體影像校正及光學失真修正 [master's thesis, National Tsing Hua University]. Airiti Library. https://doi.org/10.6843/NTHU.2015.00362
- Zhang, Y. W. (2004). 未校正立體像機之自動化影像對正技術 [master's thesis, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU.2004.02121
- Chang, H. Y. (2019). Unsupervised Content-Preserving Example-Guided Image-to-Image Translation via Domain Mapping [master's thesis, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU201903078
- 胡彥同（2013）。Resolving the Chaotic Optical Image That Comes From Different Directions and Overlaps〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU.2013.00630