3D Building Model Generation Based On Multi-source 3D Data Fusion

With the development of city intelligence,3D city models are widely used in all aspects of urban management.The 3D building models account for a high proportion of the whole city model and have a wide range of applications.Obtaining high-precision and large-scale 3D building models has also become the pursuit and research focus of the industry.In recent years,3D model generation methods continue to emerge,but the existing methods cannot take into account the local accuracy,overall integrity,and the data scale of the 3D building models at the same time.Due to the shortcomings of existing 3D model generation methods,this thesis focuses on characteristics of 3D data from different sources and proposes a 3D building model generation method based on multisource 3D data fusion.The method proposed in this thesis uses the mesh obtained by oblique photography modeling and the point cloud obtained by vehicle LiDAR scanning in the corresponding region as source data,which is mainly divided into three steps:registration of different source point clouds and creation of multi-source 3D data sets,3D model quality evaluation based on deep neural network and 3D model generation based on multi-source data fusion.The main work of this thesis is as follows:1.To address the problem of the great difference in position and pose of different source data in the same coordinate system,this thesis uses the method of rough registration based on feature matching and precise registration of ICP based on scale optimization to realize the registration of point clouds from different sources.2.Aiming at the problem of quality evaluation of the 3D model,this thesis introduces the method of deep learning and proposes a quality evaluation network for multi-source 3D data(Multi-Source 3D Data Quality Evaluation Network,MS3DQE-Net).The evaluation results guide the generation of the model.3.To address the problem that the model based on single-source data generation cannot consider the local accuracy,overall integrity,and data scale at the same time,this thesis proposes a model generation method based on multi-source data fusion,which adopts an adaptive fusion strategy.According to the results of quality evaluation,the data fusion scheme is adaptively adjusted in different regions,to realize the generation of large-scale and high-precision 3D building models.In this thesis,the proposed method is applied to the real scene data for experimental verification,and the results are analyzed on the corresponding indicators.In the 3D data quality evaluation task,the MS3DQE-Net proposed in this thesis is compared with the representative methods in related fields,and the results show that the MS3DQE-Net proposed in this thesis has a better effect.