Abstract: Aiming at the problem of scaffolding of column parts placed in disorder, a classification and grabbing system for scattered column parts based on cylindrical fitting is proposed. First, through-pass filtering, outlier removal and voxel filtering downsampling are used to preprocess the point cloud data, then the plane fitting algorithm is used to remove the point cloud, and then the region growth algorithm is used to separate the point clouds of different cylinders. Because there is a lot of noise in the segmented cylindrical point cloud, the random sampling consistency method and the least square method are combined to remove the noise, the cylindrical surface is fitted, and finally the center of gravity of the point cloud is projected onto the cylinder axis to obtain an accurate center of the cylinder. The experiment proves that the algorithm has good robustness and high precision, and is suitable for the grabbing of column parts.