As an essential processing step in unordered picking tasks, point cloud segmentation directly impacts the subsequent accuracy of object recognition and pose estimation. To address the problem of inadequate segmentation performance of the traditional LCCP algorithm in complex object stacking scenarios, an improved LCCP point cloud segmentation algorithm that incorporates Gaussian curvature information is proposed in this paper. Initially, an enhanced VCCS algorithm is employed to partition the point cloud into super-voxel, and by integrating Gaussian curvature information, the issue of super-voxel easily crossing object boundaries is further addressed. Subsequently, concave-convex connectivity among adjacent super-voxel blocks is determined, followed by the merging of all convexly connected super-voxel to form the final segmentation results. The experimental results demonstrate that the method improves segmentation precision by 3.1 % to 22 % compared to LCCP and CPC, with a noticeable enhancement in overall algorithm performance.