A contour detection method for bulk material piles based on cross-source point cloud registration

Abstract

In the field of automatic bulk material loading, accurate detection of the profile of the material pile in the compartment can control its height and distribution, thus improving the loading efficiency and stability, therefore, this paper proposes a new method for pile detection based on cross-source point cloud registration. First, 3D point cloud data are simultaneously collected using lidar and binocular camera. Second, feature points are extracted and described based on 3D scale-invariant features (3DSIFT) and 3D shape contexts (3DSC) algorithms, and then feature points are used in progressive sample consensus (PROSAC) algorithm to complete coarse matching. Then, bi-directional KD-tree accelerated iterative closest point (ICP) is established to complete the fine registration. Ultimately, the detection of the pile contour is realized by extracting the point cloud boundary after the registration. The experimental results show that the registration errors of this method are reduced by 54.2%, 52.4%, and 14.9% compared with the other three algorithms, and the relative error of the pile contour detection is less than 0.2%.