Classification denoising and 3D reconstruction of roadway point clouds

Objective Noise removal and 3D reconstruction of the roadway point cloud data are the key links to realize the digital modelling and analysis of roadways, but at present, the traditional single filtering algorithm is difficult to effectively remove the noise of different scales from the roadway point cloud, and the existing 3D reconstruction algorithms have problems such as low modelling accuracy and easy to be distorted, so it is necessary to research on the method of obtaining high-quality point cloud data and the technology of constructing high-accuracy three-dimensional model of the roadway. Methods A deep learning implicit surface reconstruction method based on signed distance functions (SDF) is designed through an adaptive denoising algorithm based on R, I_min, σ_c, σ_s parameters for classified roadway point clouds. Integrate the mean value method, the improved density-based spatial clustering of applications with noise (DBSCAN) algorithm and the improved bilateral filtering algorithm to construct the framework of classification processing technology, and the integrated algorithm automatically analyzes the type of noise in the roadway point cloud data and removes different scales of noise efficiently through an adaptive mechanism, and the integrated algorithm automatically analyzes the type of noise in the roadway point cloud data and removes different scales of noise efficiently through an adaptive mechanism. The integrated algorithm automatically analyzes the noise types in the point cloud data of the roadway. It efficiently removes the noise of different scales through an adaptive mechanism, realizing the in-depth purification of the main point cloud data. PointNet++ is used to extract the local area features of the point cloud of the roadway, import the neural implicit network to learn the local context information, generate the global model SDF, and combine it with the moving cube algorithm to construct the refined three-dimensional model of the roadway. Results and Conclusion Taking the 1:1 simulated roadway of Zhangji coal mine in Anhui Province as the experimental scene, we carry out the research of denoising and 3D reconstruction of roadway point clouds in multi-dimensional space. The results show that: (1) the integration algorithm can dynamically adjust the denoising strategy according to the roadway scene and noise categories, with adaptive optimization performance, resulting in Class I and Class II errors of 1.54% and 5.37%, respectively, and effectively removing large, small-scale and repetitive points of three types of noise while retaining the main point cloud features. (2) The reconstruction algorithm can effectively reduce the holes while maintaining the accuracy and smoothness of the roadway model and accurately portray the structural details of the complex location, and the average, standard, and root-mean-square errors of the reconstructed roadway are 0.037, 0.040, and 0.041 m, which can meet the high-precision requirements of the construction of intelligent mines, and provide high-quality three-dimensional data support for the digital transformation and upgrading of mines, as well as for the intelligent and precise mining.