物體辨識與姿態估測是機器視覺應用中重要的處理工作。近年來，由於RGB-D攝影機的廣泛使用，三維物體辨識技術也越來越受到重視，因其在雜亂的場景中，不但具有較高的強健性與辨識率，且能夠更加容易且精準的估測物體三維姿態。因此，本論文設計一個基於RGB-D攝影機的三維物體辨識與姿態估測系統。所設計之系統首先經由RGB-D攝影機擷取色彩點雲影像，並提取點雲影像特徵點。接著使用紋理以及形狀特徵以建立CSHOT特徵描述子，並將特徵點資訊進行匹配以找出對應點。利用對應點資訊輸入霍夫投票演算法，濾除錯誤的特徵匹配點，以輸出物體初始姿態資訊。最後經由RANSAC演算法優化初始姿態，再利用假設驗證步驟將物體辨識中錯誤的物體假設去除，以取得最佳的物體辨識以及姿態估測結果。在實驗結果中，證實本論文所設計之系統不僅能夠成功辨識場景中物體，且對於物體的平移以及旋轉都能精準估測姿態。 Object recognition and pose estimation are important functions in applications of computer vision. In recent years, RGB-D cameras become more and more popular and 3D object recognition technology has got more and more attention as it not only has a higher object recognition rate in a complex environement, but also is able to accurately estimate the 3D pose information of the object in the workspace. Hence, this thesis presents a novel design of a RGB-D camera based 3D object recognition and pose estimation system. First of all, The proposed system takes colored point cloud data and extracts keypoints of the scene from the RGB-D camera. Then, the existing Color Signature of Histograms of Orientations (CSHOT) description algorithm is employed to build descriptors of the detected keypoints based on texture and shape information. Given the extractd keypoint descriptors, a matching process is performed to find correspondences between the scene and a colored point cloud model of an object. Next, a Hough voting algorithm is adopted to filter out matching errors in the correspondence set and estimate the initial 3D pose of the object. Finally, the pose estimation stage employs RANSAC and hypothesis verification algorithms to refine the initial pose and filter out poor estimation results with error hypothesis. Experimental results show that the proposed system not only successfully recognizes object in a complex scene, but also is able to accurately estimate the 3D pose information of the object with respect to the camera.
