Key points are not available for this paper at this time.
In the application of vision-guided robotic grasping, 2D position/orientation and 3D depth information are needed. This paper proposes a method for detecting the 4-DoF grasping of unknown objects, which using the combined RGB image and 3D point clouds data by RGB-D camera. The 3-DoF grasp information is obtained by segmenting the target region from the RGB image. Then the depth is converted into a point cloud image by 3D reconstruction. By using an improved RANSAC to fit point cloud planes of the background and object, the height difference between object surface and the planar bottom is determined. The information of Z-axis grasping depth for achieving 4-DoF grasping is determined by RANSAC plane fitting. Experiments are conducted to verify and validate the effectiveness of the proposed method in automated robotic grasping.
Tang et al. (Mon,) studied this question.