A 3D line segment detection algorithm for large-scale scenes

A re-projection based 3D line segment detection algorithm is proposed to efficiently extract 3D contours from a large-scale reconstructed scene. For the reconstructed scene based on the 2D line segment detector and structure from motion approaches, the proposed algorithm first re-projects the spatial point clouds to image plane based on different camera matrices, then finds the best re-projected lines under the condition of weak matching. 3D line segments are derived by using re-projection index and 3D contour is obtained via line segment extension. To alleviate the matching errors caused by re-projection, a planar clustering algorithm of 3D point clouds is implemented. Experimental results show that the proposed algorithm can produce satisfied visual effects with high computational efficiency.