Research On Multi-source Data Fusion In Building 3D Reconstruction Technology

Three-dimensional modeling technology is of great significance in the fields of cultural relics protection,disaster assessment,and smart cities.The current 3D model construction technology mainly includes 2D image combined with geometric modeling,3D laser scanning,UAV photogrammetry,etc.Each specific modeling technology has certain limitations.When the 3D laser scanning technology scans large structures,there will be dead corners in the scan,that is,the top information cannot be obtained;the drone performs tilt photogrammetry,which can quickly obtain the space of the target surface Information,but the bottom information cannot be obtained,and the constructed model will be seriously distorted.Data fusion has become a trend,which can solve the problems of holes and garnishes in the model construction of a single data source,but there are still some problems,especially the problems of fusion evaluation accuracy and fusion efficiency.In the process of multi-source data fusion algorithm research,there are data The inefficiency of the algorithm caused by the large quantity,which in turn leads to the low fusion efficiency.The single modeling accuracy evaluation usually adopts the pointto-point error method,but for fusion algorithm modeling,the fusion error will be generated during the algorithm fusion process,so the point is The medium error is not suitable for multi-data source fusion modeling accuracy assessment.Aiming at the abovementioned problems,this paper conducts the following research:(1)The working principle,process and scanner characteristics of terrestrial 3D laser scanning technology are described,and the preprocessing of point cloud data is analyzed.Secondly,the drone tilt photogrammetry technology is introduced,including tilt photogrammetry principle,drone Classification,etc.,focused on the key technologies related to photogrammetric 3D modeling,including point cloud data acquisition,aerial triangulation and image dense matching,etc.Finally,the feasibility analysis of the registration and fusion of point cloud data and image data was analyzed.(2)For heterogeneous point cloud data,the traditional point cloud sampling consistency(Sample Consensus Initial Alignment,SAC-IA)algorithm is inefficient and the Iterative Closest Point(ICP)algorithm is slow to converge.Based on the above algorithm,this paper proposes an optimization method of SAC-IA combined with ICP algorithm.First,the original point cloud data is down-sampled to improve the registration efficiency.The sampling method is based on the voxel grid center of gravity algorithm,and then the next The key points of the sampled data are extracted,and the extracted key points are optimized.The purpose of the optimization processing is to reduce the matching error caused by the key points of edge noise.The processed point cloud data calculates the fast histogram of its points.Point Feature Histogram,FPFH)feature,and use the SAC-IA algorithm to complete the initial registration,and finally through the improved ICP algorithm for fine registration.On the basis of the algorithm in this paper,and compared with the other two literature algorithms,the experimental results show that the algorithm in this paper increases the accuracy of the measured data by 1.3%and 74.7%,and the fusion time is increased by 97.1%and 97.5%.It provides certain reference value for engineering fusion modeling.(3)In the stage of model construction and accuracy analysis,use Geomagic Wrap software to model and combine with Geomagic Control for accuracy analysis.Generally,3D max and Sketch Up modeling software are commonly used for point cloud model modeling,but they have problems such as complex operation,weak modeling automation,and distortion of texture mapping.Therefore,this article uses Geomagic Wrap software for 3D model construction;accuracy analysis stage In this paper,the feature points are used for fusion,and the control points are ignored in the data collection stage,so the overall analysis method is used to evaluate the progress.Use Geomagic Wrap software to complete the automatic modeling of the fusion data,complete the model construction according to the three stages of "point processing-polygon processing-shape processing",and finally the texture mapping stage to obtain a three-dimensional model close to the entity.In the accuracy evaluation stage,the three point cloud models and the threedimensional model are used for comparative analysis,and the scanner point cloud comparison is used as a reference.The comparative analysis method adopts 3D and 2D deviation analysis methods to conduct overall analysis and extract XZ and YZ plane analysis respectively.The experimental results show that the Geomagic Wrap software has high modeling automation efficiency,and the constructed three-dimensional model is close to the real scene.In terms of accuracy analysis,the accuracy of the fusion model is close to the accuracy of the scanner model by using the deviation analysis method.Figure[53]table[18]reference[77]...