BDSR:A Best Discrepancy Sequence-Based Point Cloud Resampling Framework For Geometric Digital Twinning

Digital construction models are more and more widely used in various fields of research.In the field of civil engineering,digital models can be used to better understand and control the operating cycle of buildings.To better construct digital models,accurate and efficient data must be used.The research object in the field of civil engineering is no longer simply one-dimensional or two-dimensional data,but three-dimensional or even higher-dimensional data.Therefore,a different processing method from low-dimensional data is required.The research object of this paper is mainly point cloud,which is generally obtained by three-dimensional instruments such as LiDAR,three-dimensional scanner,etc.The data obtained by such instruments often cause data errors due to the position of the instrument and human operation,and the obtained point cloud data may contain many messy irrelevant points,such as noises and outliers.At the same time,buildings in the civil engineering field generally contain a huge amount of data.Some building point cloud files may contain tens of millions of points,which will greatly affect the accuracy and efficiency of subsequent digital model construction.This paper combines the two inherent defects of the original point cloud to propose a new point cloud pre-processing and resampling technology,which is called BDSR.Put the pre-processed data into the model construction to get a better digital model,which is beneficial to the subsequent series of operation procedures.The main work and contributions of this paper are as follows:1.This paper first downsamples the point cloud based on the BDS(best discrepancy sequence)sampling method to achieve the preliminary effect of simplifying point clouds.The BDS algorithm is superior to other methods such as random sampling commonly used in statistics.The extracted data can well represent the characteristics of the population,laying a foundation for the representativeness of the resampling results to the overall point cloud.2.Pay attention to the key information of the point cloud,distinguish and filter the key parts of the point cloud through normals and curvatures,so as to realize the different processing of different regions and improve the efficiency of the algorithm.Compared with other feature,normal and curvature are easier to be calculated in the point cloud.Thus,the algorithm is efficient,concise,and easy to operate.3,Innovatively uses voxels to retrieve key parts of the point cloud for upsampling to ensure that important information is retained to the greatest extent and avoid losing too much key information during the BDS downsampling process.This step is of great significance to the entire algorithm.The overall number of point clouds has decreased,but the information on key parts is sufficient,and the simplified points come from non-key parts as much as possible.4.Because the original point cloud obtained by the instrument will inevitably have sparse areas,this paper combines the overall density of the point cloud to determine the sparse area and add fake points,which can more realistically enrich and restore the lack of key parts caused by external factors;5.This paper proves the theoretical advantages of BDS sampling compared to other sampling methods,and uses multiple judgment indicators such as VFH descriptor,Cloud-to-Cloud distance,Hausdorff distance,Area rate,and 2D entropy to compare BDS-based sampling techniques and other existing algorithms to ensure the reliability and accuracy of the experiments.Besides,this paper applies statistical analysis to verify experiment results.In general,the BDSR makes full use of the normals and curvatures of the point cloud to distinguish and filter key parts,and uses the voxel to capture the key information eliminated to ensure that the key information of the point cloud tends to be complete.Compared with other algorithms,BDSR proposed in this paper effectively simplifies the sample and enriches the information,and the sampling results perform more stable,which can be beneficial to the later operation.