Automatic 3-D Scene Model Reconstruction Using a RGB-D Sensor

[[abstract]]在本論文中，我們結合了現有的方法並提出一個自動三維場景模型重建系統。首先我們會使用階層式基於外觀貼合方法進行局部影像貼合，估測攝影機的方位。再來我們會利用散射變換偵測有重複場景的 RGB-D 影像，並用 SURF、Arun、RANSAC 等方法估測攝影機的平移和旋轉。將影像貼合網路以圖的方式表示，其中每個點代表了 RGB-D 影像和該時間點的攝影機方位，每個邊表示了兩個點之間的局部影像貼合結果。然後我們利用代數平均法將累積誤差隱藏。最後在實驗結果的部分會與最新的各種方法進行比較。[[abstract]]In this thesis, we integrate existing methods and propose an automatic 3-D scene model reconstruction system. First, we estimate the camera poses using heirarchical appearance-based registration which is a pairwise registration method. Then,we detect repeated scenes in the RGB-D images with Scattering Transform and estimate camera translations and rotations using SURF, Arun, and RANSAC. The registration network is represented as a graph. Each node of the graph represent a RGB-D image and the camera pose. Each edge of the graph represent a pairwise registration result of two nodes. We hide the accumulated registration error using Lie-algebraic averaging. Experimental results show that the proposed method is comparable to the existing state-of-the-art methods.[[note]]碩