Target Segmentation and Pose Estimation for Rapid Grabbing Using Depth Information

摘要

In order to realize rapid grabbing of stacked objects, we propose a quick and accurate method to realize target segmentation and pose estimation using only depth information. We use filtering to remove the noises in detection. Then we detect the contours and use the averages and gradients of depth to find the contour of the top plane. We calculate the centroid and normal vector of the top plane to get the pose estimation. Thus we can realize the rapid grabbing of the top object, then achieve rapid grabbing of stacked objects one by one.