A novel octree partitioning approach based on camera parameters for unknown object representation

This paper proposes an improved octree representation method which is necessary for robot manipulation. To manipulate an object in a certain complex environment, robot needs simple, precise, and task oriented environment representations. Octree is a effective and broadly used three dimensional volumetric representation method to do this. But the problem of the conventional octree representation methods is that it generate many useless and incorrect cells because it generates the octree based on the number of 3D points. To solve these problems, we use the new 'Fill-Factor' concept. This paper introduces the concept and shows the experimental results to show the feasibility of the concept.