Laplacian Fusion Approach of Multi-Source Point Clouds for Detail Enhancement

Occlusion problems cause many holes on scanned object surfaces in terrestrial laser scanning point clouds. The existing simulation methods, based on the properties of structural continuity, cannot recover the missing details. Previous studies have focused on co-registration in terms of using multi-source. However, the repairing accuracy of co-registration is insufficient to support subsequent reconstruction. Thus, one of the main contributions of our work is to present an effective fusion method to enhance details. A multi-view-projection-based vacancy filling strategy is leveraged to repair the integrity of the details. Subsequently, the multi-source data are accurately connected by the Laplace differential domain fusion approach. Specifically, in the first step, the proposed method detects 3D holes individually from the 3D and 2D interconversions, and then extracts the corresponding complementary point clouds from different sources. Subsequently, the Laplace differential coordinates are exploited to describe the repairing data in order to recover the associations as accurately as possible. The practicality of our method is demonstrated on realistic point clouds from multiple sensor surveys. The performance of our method is impressive in terms of the repairing completeness, with average values up to 82%. These positive experimental results demonstrate the feasibility, practicality, and potential of the proposed solution.