Investigating Data Fusion from Three Different Point Cloud Datasets Using Iterative Closest Point (ICP) Registration

A dataset from a single survey technique, such as terrestrial laser scanning, hardly generates a complete 3D model. Therefore, data fusion is usually used to integrate various data to overcome the technique limitations. For multisource point cloud data, Iterative Closest Point (ICP) registration can be used to combine the data. However, at the end of the ICP registration process, there are gaps between the point clouds. This paper contains an initial work to investigate the fusion quality of three point clouds (LiDAR point by drone, TLS point, and image-based point) using ICP registration. In this case, the TLS point is used as reference and target as well, while the other point clouds are the source data. This experiment observed the fused model of the ICP and inspected the roughness of each data for further assessment. Then, the M3C2 distance tool in the CloudCompare software was used to show the existence of distance between the point cloud models and the voids as well. Based on the results, the fusion of the three datasets completes the 3D model, except for a few holes caused by obstructions, such as vegetation and roof shadow. It is noted that the image-based point provides a more substitute data, while the LiDAR point by drone has a minor role to complete the model. Hence, for future work, the use of data from another technique can be considered for data fusion and distance refinement is important in this case.