Research On Cavity Completion And Incomplete Stalactite Point Cloud Repair Of Cave Wall Point Cloud Based On Deep Learning

In the development of karst caves concerning the tourism,obtaining complete three-dimensional spatial information inside karst caves is a prerequisite for tourism development design.The detailed three-dimensional spatial information inside the karst cave cannot be obtained and expressed through traditional measurement methods.With the continuous development of surveying and mapping technology,the use of 3D laser scanning technology can quickly obtain 3D point cloud data on the surface of the measured object,and photogrammetry technology can also obtain 3D point cloud data of the measured object by processing the acquired image data.This provides an effective method for the acquisition and expression of three-dimensional spatial information in the cave.However,on the one hand,neither 3D laser scanning technology nor photogrammetry technology can obtain the 3D spatial information of the occluded area,and the complex spatial structure in the cave is not easy to be presented,resulting in partial missing of the acquired data,which is disadvantageous to the engineering design.On the other hand,some stalactites are often stolen due to lack of protection in the caves which have not been developed.In some developed karst caves,the structures in the cave are damaged due to improper construction.The pirated stalactite mining and improper construction will affect the landscape of the cave,making it unable to meet the sensory needs of tourists.Based on the above two reasons,it is necessary to repair and complete the obtained three-dimensional cavern data.In the current engineering practice,in order to realize the restoration and completion of the 3D spatial information of the karst cave,the method usually adopted is to first establish the incomplete 3D mesh model based on the preprocessed residual fault cloud data,and then use the model post-processing software to repair and complete the model.But this is often time-consuming and laborious,and the repair results are seriously distorted.This paper takes the point cloud data as the research object,and introduces how to repair and complete the incomplete stalactite point cloud and the point cloud cavity of the cave wall by deep learning method.The main research content and results are as follows:(1)Construct stalactite and cave wall point cloud data sets for point cloud completion network trainingAt present,there is no point cloud data set that can be used to train stalactites and cave walls for point cloud completion networks,and the construction of the data set is crucial to the completion effect of the network.Therefore,this article uses a variety of data collection methods to collect point cloud data from Jiuxiang Triangle Cave,Bamei Taoyuan Cave and Tangna Cave,and extracts stalactites and cave wall point clouds from the point cloud data of the three caves.The stalactites and cave wall point clouds are preprocessed to establish a data set of stalactites and cave wall point clouds.(2)Research and construction of point cloud completion networkBy studying the deep learning technology and the current main point cloud repair and completion network,a deep learning network that can be used for the repair and completion of incomplete stalactite point clouds and cave wall point clouds is constructed;the network is based on the architecture of generating a confrontation network.First,a multi-resolution encoder structure is used,and then make a feature extraction of incomplete cloud data of different resolutions by combining multi-layer perceptrons and fusion of the features of different resolutions,effectively extracting the local and global features of the point cloud,which contains the high-level and low-level semantic information;secondly,the output of the multi-level decoder structure is corrected hierarchically by using the chamfering distance loss function;finally,the discriminator structure is used to make the network output point cloud closer to the real point cloud.(3)Incomplete stalactite point cloud and cave wall point cloud hole repair and completionBy using self-built stalactite point cloud data set and cave wall point cloud data set to train the network,finally the repair and completion of incomplete stalactite point cloud and cave wall point cloud voids are realized.By comparing with the results of some current repair networks and software,it shows that the method proposed in this study to repair the incomplete stalactite point cloud and cave wall point cloud voids has a better result than other methods,including Folding Net and PCN network,Geomagic Studio and Maptek I-site studio in terms of repair effect and repair accuracy.