Railway scenes point cloud model denoising based on improved score-based denoised network
Reverse modeling based on point cloud data is a pivotal step in the digital advancement of railway engineering,serving as the foundation for various extended applications such as digital twins of railways and structural health monitoring.Traditional point cloud models constructed directly from unmanned aerial vehicle(UAV)survey data often suffer from low accuracy and poor reliability,particularly in complex railway environments,leading to significant deviations between the models and actual structures.To address these issues,this study introduced an Improved Score-based Denoised Network(ISDNet),which constructed a new set of encoders and decoders on the foundation of traditional score-based denoising networks.This network not only enhances the precision of three-dimensional point cloud models but also effectively reduces computational complexity and costs.The algorithm decreases discrete noise caused by environmental disturbances during the UAV-based reverse 3D point cloud modeling process and performs well in handling structural details of railway infrastructure.The reconstructed point cloud model's coordinate points were compared and validated against coordinates measured by a total station.Experimental results indicate that the error in the 3D models of railway scenes denoised with the ISDNet network was reduced to half of that of the original models,with average errors of 5.67 mm in the X direction,6.67 mm in the Y direction,and 6.83 mm in the Z direction.The denoised 3D point cloud model effectively minimizes the loss of detail data in elements such as the catenary system and rails and exhibits enhanced shape representativeness for clearer visibility.The method proposed in this paper can realize the efficient denoising of the point cloud model of the railway infrastructure and improve the accuracy of the point cloud model,which effectively reduces the loss of the details of each component in the point cloud model,and can provide reliable technical support for the related extended applications.
UAV 3D point cloud modelrailway engineeringdeep learningpoint cloud denoisingmacro modeling
万方数据
维普
