Segmenting the point cloud from the 3D point cloud data of a subway tunnel is a key step to automatically detect the damage of the subway tunnel and reconstruct a 3D model of the tunnel. The collected 3D point cloud data are inaccurate for calculating the normal vector and curvature of the point cloud because of the structural characteristics of an automated detection system. This renders some common 3D point cloud segmentation algorithms, such as an improved region growing segmentation method, unsuitable for the point cloud data collected by the detection system. To segment the 3D point cloud data collected by an automated detection system, an algorithm based on density clustering was designed and implemented. This algorithm avoids the use of inaccurate normal vector and curvature, overcoming the limitations of an automatic detection system. Finally, we compared the segmentation results of the region growing segmentation method with those of the designed segmentation algorithm based on density clustering using the actual 3D point cloud data.