Octree-based obstacle representation and registration for real-time

Abstract

This paper presents a novel approach for obstacle representation method when robotic manipulation task system understand real indoor environment for 3D workspace modeling. When many objects scattering on the table in workspace, if the robot want to grasp only one object, the robot system should has a path planning to approach the object. In this case, the obstacle representation in cluttered-environment is an important role for robot system. The research area of 3D workspace modeling, the research of step by step accumulating environment observation method is more difficult than the entire environment observing method and the research is not reported sufficiently. In this paper, we can contribute the two issues for real-time 3D workspace modeling to using the sequential input stereopsis scenes information. First, we can suggest a method to estimate the transformation matrix from using SIFT feature and Epipolar Geometry Constraint characteristics in the continuous process of accumulating sequential input stereopsis scenes for more fast and accurate than recently research. This method guarantees a feasible transformation matrix result better than the using traditional ICP and general SIFT method. Second, we can suggest a method for octree-based obstacle representation and we can also suggest an octree update method for real-time. It is faster than entire workspace observation method, and if the robot doesn't know about the objects and obstacles information in workspace, we can help the robot to understand environment himself from practical information. Taking the obstacle information from above method can help the robot system possible to do path planning for robotic manipulation task in 3D workspace. Through the experimental result, we can show that our method is well-performing and well-modeling the obstacle in 3D real environment workspace modeling in real-time.