As much importance has been paid to the quality control and project management in recent construction projects, the need to acquire accurate three-dimensional (3D) spatial and shape information has increased. Studies have been actively conducted to raise the construction work efficiency and quality by utilizing state-of-the-art technologies such as Terrestrial Laser Scan (TLS), Unmanned Aerial Vehicle (UAV), unmanned Ground vehicle (UGV), and Mobile Mapping System (MMS), in which 3D as-built models in a large scale construction site are acquired as a form of point clouds, which are then processed according to the use purpose. In particular, technologies of recognizing earthwork environments where construction equipment works for automation, remote and autonomous control of the equipment and implementing them by as-built models in an earthwork phase where many pieces of construction equipment such as excavators, loaders, and dozers are used are essential source technologies. Because the 3D information acquisition technology used in earthwork environments acquires all topographic information of surrounding environments using a wide-area data acquisition technology, it has limitations that data cannot be acquired or blind spot occur due to environmental limitations such as installation of surveying equipment and static, e.g, scaffolding, temporary construction and stopped equpment or dynamic obstacles, e.g, moving equipment and workers, or due to technical limitations TLS has a high accuracy of the acquired data, but has a limited shooting angle, transmissivity and reflexibility in the acquisition process. UAV can easily acquire data, but there are many factors, for example, camera resolution, stability of gimbal, processing accuracy that affect the accuracy of the data. To remove the blind spot, a scan of object from various angles is needed, but it is cumbersome and time consuming to move a scanning place in a vast construction site. Furthermore, even if scanning work is done at multi angles, it is difficult to cover all blind areas which occur on the spot. The purpose of this study is to acquire data by utilizing TLS and UAV technologies and generate optimized 3D world model by complementing heterogeneous data to overcome the current limitations of 3D spatial information in an earthwork phase of the construction project. To do this, this study proposed a hybrid (TLS+UAV) point cloud data based improved methods for increasing accuracy and optimizing data in the whole cycle from the acquisition, referencing, registration, merging, and up to optimization. There are differences in data density and accuracy occurred due to technical difference between TLS and UAV point cloud data. In order to efficiently combine the two data with these differences in accuracy and density, various studies have been conducted to raise the point cloud density and accuracy of UAV. For example, studies on flight optimization (altitude, photo overlay, and shooting angle etc.) equipment operation, improvement of equipment stability, and layout and number of ground control point (GCP) target in the data acquisition phase, or improvements on key-point extraction algorithms at the data processing phase have been conducted. For key-point extraction, algorithms are focused on accurate and rapid recognition of matching points between images and based on this, calculation of 3D depth. This study improved the feature detection performance for key-point extraction of acquired 2D images in the 3D reconstruction phase by maximizing the contrast of images through pre-processing based on Adaptive Histogram Equalization (AHE) and Local Contrast Enhancement (LCE). For TLS and UAV data registration, an algorithm that combined the Iterative Closest Point (ICP) algorithm and point matching was proposed and a method where the low density distribution detection was applied to the Statical Outlier Removal (S.O.R) filtering method was proposed to increase the efficiency of the noise removal in each data set. Once data registration is complete, merging into single data is needed, in which optimization work is needed to remove duplicate or unnecessary data. In this study, data were merged in the manner of complementing only the blind areas of TLS point cloud data with point cloud data generated through photogrammetry, and voxelization-based Blind Area Detection (BAD) algorithm was developed: 1) Removal of unnecessary duplicate points between two data sets, 2) Detection of point density-based valid data for recognition of blind areas (non-created point region): matching of TLS point cloud-based voxels where points are not present and UAV-photogrammetry point cloud based voxels where points are present, and 3) Detection of active and inactive data within the identified data (points existed between points of regions where points are densely present without blind areas) For lightweight merged data, data are optimized based on individual voxel method that divides the entire point cloud space into single voxels, and voxel octree method that has a hierarchical structure by dividing the entire point cloud space into eight areas by the level The lightweight degree and accuracy change were analyzed according to individual voxel's dimension, voxel octree's level, and dimension and level that satisfied the error range determined in this study were derived. The TLS and UAV data of earthwork data in two sites were acquired to verify the algorithm and technique proposed in this study and applied and comparatively analyzed for each step. For the surveying error range of the earthwork sites, the ±10cm criterion, which was the public surveying error range specified in the specification of the National Geographic Information Institute and the ±15cm criterion, which was the earthwork error range for slopes, were reflected. In order to evaluate the error of registration, each Ground Control Point (GCP) Target-Marker was installed in six places and Global Positioning System (GPS) was used to obtain the coordinates values of GCP. The ICP+PM registration method proposed in this study and ICP algorithm was compared and analyzed. After applying each registration method and GPS coordinate values of GCP markers for error verification, comparison between coordinate values extracted from the GCP was conducted. The results exhibited that the registration error value was reduced based on the GCP. The results of applying the proposed optimization technique showed that the data amount was reduced by about 90% and the RMS satisfied the error range. In addition, the elevation analysis on Z value in the optimization and original models was conducted after converting to the TIN DEM model, which was widely used in earthwork design and construction, and the analysis results exhibited RMS 0.149488 (individual voxel) and 0.117795 (voxel octree), which satisfied the reference allowable error value.
