The purpose of this paper is to address the problem of robot posture estimation during augmented virtual scene construction. In this paper, a multi-cylinder segmentation algorithm based on point cloud curvature features is proposed for cylindrical articulated robots. First, 2D images and Point Cloud Data(PCD) from multiple viewpoints are captured by an RGB-D camera, and each viewpoint contains the PCD of robot links and ChAcUco marker in the environment. Secondly, the multi-view PCD are fused, and processing operations such as Gaussian noise filtering, pass-through filtering, and point cloud surface smoothing are performed. Third, the PCD of each robot link is separated from the fused PCD based on the point cloud curvature feature. For the segmented link's PCD, the point cloud is cylindrically fitted based on the point cloud normal and the distance of the point to obtain the link direction. Finally, the robot joint angles are solved based on the link orientation. The experimental results show that the algorithm is robust to the robot posture, and has high computational efficiency and accurate computational results.