2026 Volume 7 Issue 3 Pages 514-522
In recent years, various types of three-dimensional (3D) spatial data have been developed as fundamental data supporting the digitalization of urban spaces and social infrastructure. To utilize these 3D spatial data for road management and infrastructure inspection, accurate estimation of the camera pose from monocular images acquired in the field is essential. This study proposes a camera pose estimation method based on shape matching between a rendered depth image generated from a LiDAR point cloud and a depth image estimated from a monocular image. Corresponding points were identified using Fast Point Feature Histograms (FPFH) extracted from point clouds generated from the depth images, and the camera pose was estimated from the matched correspondences. Experimental results demonstrated improved alignment between the camera image and the re-rendered image, indicating that depth image shape matching using a LiDAR point cloud is effective for camera pose estimation from monocular images.