使用立體視覺建立網路虛擬實境之地理資訊系統

Abstract

本研究延續前人所建立的地理資訊系統及網路地理資訊系統，從中再進行新功能的擴充，這些功能包含三維模型的重建、展示與網路虛擬實境的建立。在三維模型重建方面，主要是使用立體視覺來估算空間中的深度資訊，以及從攝影機校正所得的攝影機參數來做為模型重建的依據，再使用比對影像對所得到的像差陣列，藉由簡單的針孔相機模型幾何關係，以投影重建方法來將模型的三維座標重建出來。在重建的同時，利用像差值來做為景物分離的依據，以及建立像差在修補及濾波上之後處理工具，以利快速取得所需要的場景物件模型。多視角景物的合成是使用SIFT的方法來找出影像與影像之間的特徵點對應關係，藉由對應點的三維座標將不同視角之同一景物模型完成定位的工作。在不同視角之模型定位完成後，將各個模型的資料點投射到圓柱的曲面上，然後使用Delaunay triangulation演算法，將曲面上的資料點進行三角網格化，完成不同視角模型之資料點之網格化與合併。如此將三維空間的網格問題簡化至二維，便可用套用一般影像處理的方法，來處理二維圓柱曲面之三維模型網格問題。在網際網路的三維模型繪圖展示上，使用網路虛擬實境模型語言X3D檔案格式，將前人所重建的環場模型及使用圓柱投影像重建的單一模型擴展到網路上，做三維模型的即時繪圖及供人瀏覽。將建立好的X3D物件依照拍攝時所記錄的地理資訊，與前人所建立的網路地理資訊系統做連結，在提供導覽的同時，也提供使用者和場景模型的互動。

This research continues the development of the integrated Web-GIS created by former studies and expands new functions including the 3D model reconstruction, display and internet virtual reality. In the reconstruction of 3D models, we applied stereo vision to calculate the depth information, determined the camera intrinsic parameters by camera calibration, and reconstruct 3D models by projective reconstruction based on simple geometry of pinhole camera. The segmentation of 3D objects in a scene was based on the disparity estimated from image pairs. We also developed the post-processing tools for mending and filtering of the segmented 3D objects. The multi-view objects were merged by using the SIFT method to detect the corresponding feature points of images from different views, and the matched points are used for the registration and merging of the two 3D objects. Following the object registration of different views, the data points of 3D objects are mapped into the cylindrical surface. The Delaunay triangulation method is then applied to triangulate the points on the surface and subsequently completes the merged model based on the registered feature points. This procedure simplifies the 3D meshing problem to a 2D problem and thus 2D image processing algorithms can be employed for the computation of 3D reconstruction. To visualize the 3D models in the internet environment, the virtual reality modeling language X3D was used to render and display the 3D objects in real-time. The X3D models, together with the panoramic images, can be linked to the GIS system to form a virtual reality GIS (VRGIS) for internet access. The developed VRGIS thus allows users to interactively navigate through scenes in virtual reality.