Multi-View Reconstruction Of 3D Building Models Based On Airborne And Mobile LiDAR Data

The three-dimensional (3D) reconstruction of a building model is a hot topic of research in many fields such as geographic information science and remote sensing. With the growing application for 3D navigation, urban environment simulation, low-carbon city construction and disaster emergency management, requirements for the high accuracy, high detail and high integrity of 3D building models have become more urgent. Airborne and mobile Light Detection and Ranging (LiDAR) technology are important means for obtaining 3D space information due to the advantages of active laser detection, high detection precision, and short operation cycle. The integration of airborne and mobile LiDAR data for 3D building reconstruction has become a very worthwhile research topic and has wide application prospects.However, the integration of airborne and mobile LiDAR data, and the 3D building reconstruction based on both point cloud data have still been limited, relevant research approaches and methodologies still wait to be explored. Therefore, a technology solution by integrating airborne and mobile LiDAR data is proposed to reconstruct 3D building models with multi-views (top-view and side-view). The dissertation is focused on the reconstruction of 3D building models by taking complementary advantages of integrated airborne and mobile LiDAR data. The solution mainly includes three steps:integration of multi-source data, reconstruction of building roof models, and reconstruction of building facade models. And the contents are as follows:(1) Integration of airborne and mobile LiDAR data based on ground points. It is difficult to integrate point clouds from different platforms, such as airborne and mobile LiDAR system, due to significant differences of scanning angle, space coverage, spatial resolution, and scene complexity. Besides, signal-loss problems occur frequently in GPS navigation of mobile LiDAR system, which lead to the location excursion of scanning point cloud, and make it more difficult for the data integration. This dissertation presents a "coordinate transformation-global correction-local refinement" integration strategy, which preliminarily integrate the airborne and mobile LiDAR in large region, and refine the preliminary integration result in local region. First, a processing flow of coordinate transformation and geometric correction is adopted to rectify the partial deviations of mobile LiDAR data. And the airborne and mobile LiDAR data are preliminarily integrated based on the GPS information. Second, common ground point cloud is detected and used for the precise registration of preliminary integration result by adopting ICP algorithm. By then, the airborne and mobile LiDAR data in local region are refined. The experimental results shows that the proposed integration strategy can achieve fine preliminary integration result, and the integration accuracy in local region is further improved. The integrated point clouds can provide data support for the 3D building construction.(2) Reconstruction of 3D building roof models from airborne LiDAR data based on multi-scale grids. The characteristics of airborne LiDAR data, such as discrete distribution, huge data amount, and loss of spectral information, bring challenges for the detection and reconstruction of building roof models. This dissertation proposes a "building seed region detection-surface structure detection-building roof model construction" technical strategy, which reconstruct 3D building roof models with high correctness and completeness in a large urban area. First, in the large-scale grid, an nDSM is constructed to eliminate the influence of topographic relief on building detection, and a new algorithm is proposed to detect high accuracy building seed regions. Second, in the small-scale grid, a hierarchical morphological interpolation method is proposed to generate a high-resolution depth image, after then a full λ-schedule algorithm is used to segment this depth image, for detecting the precise and detailed building features. Third, by fusing the multiple scale results to realize advantageous complementarities, the high accuracy and high detailed 3D building roof models are reconstructed. The experimental results shows that the proposed method can ensure the technical applicability in a large urban area, and further improve the detection and reconstruction quality of building roofs.(3) Reconstruction of 3D building facade models from mobile LiDAR data based on repeated tile structures. Due to characteristics of mobile LiDAR data, such as massive number of points, uneven point density, significant amounts of occlusion, and complicated landscapes, it remains difficulties to reconstruct building models, while the detection of facade structures remains in the exploratory stage. This dissertation proposes a "point cloud pretreatment -building facade points unfolding-repeated facade structures detection-building facade model construction" technical strategy, which reconstruct 3D building facade models with high integrity. First, a point cloud pretreatment method is proposed to rapidly and efficiently extract the building facade points, and preliminary split the complex building structures. Second, an unfolding method whereby a building facade is flattened onto a 2D plane is introduced, to reduce the difficulty of building reconstruction for multiple facades. Third, a two-direction detection strategy, with facade structure detection performed in the vertical and horizontal directions successively, is used to determine repeated patterns. Fourth, a restoration method is developed to consolidate the imperfect data according to the similarity of facade structures, and the high integrity 3D building facade models are reconstructed by using the restored facade points and the repeated facade structures. The experimental results shows that the proposed method can effectively deal with missing areas caused by occlusion, viewpoint limitation, and uneven point density, as well as realizing the highly complete 3D reconstruction of a building facade, provide a new solution to reconstruct building facade models.