深度相機可以快速取得拍攝範圍內，面對鏡頭之物件表面3D點雲輪廓。由於對後續應用之方向不同，使將點雲解析的方式有所差異。對於現場運作的工業機器手臂而言，將工件取與放之前，必須能即時認知爪夾可以靠近並抓取之部位及其3D輪廓，或是物件預備放置處的3D輪廓概況。然而機器手臂的抓取方式受限於爪夾造型，因而在抓取前只需辨認有限數量之夾持特徵是否存在，以及其位置與方向是否爪端能到達即可。一般而言，工作物件之外觀可視為由數個夾持特徵依據固定之拓樸關係所組成之集合，若能辨識到越多特徵則對物件的類別及方位將更為清晰，亦表示可達到如同人員般的智慧識別能力。本研究將探討深度攝影影像之點雲解析方法，藉由分析辨識凌亂擺放工件之3D點雲是否存在特定夾持特徵來求得即時夾持之方位，並可進一步推算出抓取搬移後的放置方位，以達到替代人員做3D視覺辨識之能力。

Depth camera can be used to capture 3D terrain in front of it in the form of point clouds. Many applications can then be developed by specific deciphering methods. One of the applications is to realize the existence of any 3D grasp features on a work part for an industrial robot, and then plan for grasping operations in realtime automatically. Since the way of grasping has everything to do with the shape/design of a gripper, the topology of a grasp feature is therefore limited. Moreover, if a part is graspable, then it must have at least one grasp feature on its outer surface, or a part topology must include the topology of the grasp features. Since point clouds represent 3D terrain, any grasp feature on the point clouds will reveal the position and orientation for its grasping. In this research, a methodology of topology mapping from 3D point clouds is developed to discover the existence of grasp features from several disorderly placed work parts. A grasp operation can then be created using inverse kinematics based on the feature position and orientation. It is believed that the research setup may adequately replace human vision on part handling tasks.