An improved region growing point cloud segmentation algorithm was proposed to solve the problem of difficult segmentation in the process of target parts recognition in unstructured environment due to the stacking of parts. Firstly, the point cloud obtained is preprocessed by pass-through and statistical filter to obtain the point cloud after removing redundant data. Secondly, octree is used to establish the topological relationship between point clouds to improve the efficiency of subsequent point cloud segmentation. Finally, the curvature calculation method based on local radial basis function was selected. The minimum point of curvature was selected as the initial seed point through calculation, and the range of spatial threshold was set for regional growth. Experimental results show that the proposed method can effectively solve the difficult segmentation problem caused by the stacking of parts in the unstructured environment, and the segmentation accuracy is high and the efficiency can meet the real-time needs of industry.