INTEGRATION OF DIGITAL PHOTOGRAMMETRY AND LASER SCANNING TECHNIQUE FOR GENERATING HIGH-QUALITY 3D POINT CLOUDS

摘要

Light Detection and Ranging (LiDAR) technology is capable of acquiring point cloud data quickly with great accuracy and high density. However, for a complex scene, missing areas frequently occur as a result of occlusion, reflection, and other optical factors. In this study, an image-based approach for point cloud reconstruction and densification is presented. In the case where the missing areas can be identified from image data, the proposed method requires only one single image to reconstruct the point cloud of missing areas. First, the camera exterior orientation parameters were determined by manually selecting correspondences between LiDAR point cloud and an image and a single-image resection algorithm was applied. By assuming the planar geometry whose normal vector can be estimated using the neighboring points of missing area, high-quality object points were then created, filling up the missing area in the point cloud. In addition to filling the missing area, the abundant spatial and spectral information from an image can be used to density the raw LiDAR data, providing more details on texture information. An outdoor experiment on building facade has been conducted in this investigation. By applying the proposed, missing areas like window glasses or occluded walls have been successfully reconstructed. Moreover, this approach has improved the visual quality of LiDAR point cloud, achieving more realistic data for subsequent applications such as 3D modelling.