A 3D Modeling Method of Indoor Objects Using Kinect Sensor

Abstract

Indoor 3D object modeling is often used for object recognition, location and other robot manipulating task. In this paper, we put forward a common method to build 3D model of household object for robotic grasping. The point clouds including object from different viewpoints are captured from the Microsoft's Kinect v2 sensor. A pixel filtering approach is used to process depth image and morphology algorithm is implemented to filter noise points in the point clouds of object. FPFH descriptor is extracted from each point. Sample Consensus Initial Alignment and ICP algorithm is used to register two adjacent point cloud accurately. Based on a closed-loop optimization method, the cumulative error from continuous registration is reduced. We build some 3D models of indoor objects through the proposed approach. The experiment results shows that the method is convenient and can meet the accuracy requirements.