Indoor objects 3D modeling based on RGB-D camera for robot vision

The 3D-object modeling has become one hottopic in robot vision field for robots locating, grabbing and dropping operation. In this paper, we propose an efficient method of object modeling based RGB-D camera (Kinect) that can help the robot to carry out operation task. Our method is divided into three stage: firstly, point clouds of scene containing object are captured from the camera and plane-segmentation to extract clusters of object; secondly, the preprocessed method is used to filter noise and outlier points; Points pair-wise alignment by way of our proposed pairs rejection strategy, global consistent alignment of the complete object and object-centered representation easy to manipulate. Besides, for the well-known loop closure problem, we optimize the pose and eliminate the accumulations base on the ICP improved algorithm. Finally, the proposed method is validated by the modeling of several household objects.