Progressive Rendering Of Oblique Photography Scene Based On Point Cloud

Oblique photography breaks through the limitations of traditional orthophotography,which can only be taken at a vertical angle,and collects images from a vertical angle and multiple oblique angles at the same time,and after positioning,fusion,modeling,and other processing can generate photo-realistic 3D scene models,becoming an important technology in the visualization of urban scenes in areas such as smart cities,geographic information systems,and disaster detection.As oblique photography is often used for 3D reconstruction of large-scale scenes,the 3D models obtained can reach the level of GB or even TB in data volume.In practice,C/S architecture where the model data is deployed on the server-side,and the rendering utility is deployed on the client,is widely applied to facilitate the unified management of model data and the protection of intellectual property rights.Therefore,for large-scale oblique photography scenes,the study of its real-time rendering technology under the C/S architecture is of great significance to the popularization of oblique photography in different fields.This thesis takes web-based real-time rendering of oblique photography scenes as the research object and analyzes the existing LOD-related technologies and solutions in detail.On the basis of a full study of existing continuous LOD algorithms,this thesis proposes to use point clouds as basic primitives to construct the LOD hierarchy structure of oblique photography scenes and proposes a progressive rendering algorithm based on point cloud LOD.The core idea of the algorithm in this thesis is that in a C/S-structure-based web application,the preprocessed point clouds are sorted in order of visibility before being streamed to the client,while the client submits the transmitted data to the rendering pipeline to enhance the visual effect.To address the problems in the above process,this thesis further proposes a multi-LOD hybrid rendering algorithm based on the Splatting algorithm.Compared with the existing algorithms,the algorithm in this paper solves the problem of the lagging and popping of the LOD level switching of the scene terrain block in the unstable network environment with limited bandwidth.Specifically,the main work of this thesis includes:1.Adopt point clouds as basic primitives to build LOD hierarchy structure.The vertices in the point cloud are independent and complete,allowing us to perform more fine-grained processing of the LOD leaf nodes in the transmission and rendering stages,which is an important basis for other improvements in this thesis.2.Propose a multi-LOD hybrid rendering algorithm.To address the problem that high-precision data that has only partially completed transmission cannot contribute to the visual effect,this thesis proposes a multi-LOD hybrid rendering algorithm that mixes high-precision data that has only partially completed transmission with complete lowprecision data,so that the features in partial high-precision data can be expressed in a timely and effective manner.3.Propose a chunking preprocessing and prioritization strategy for point clouds.To address the problem that the vertices that reach the client first may lie heavily behind the current viewpoint during point cloud data transfer,this thesis performs normal vector-based preprocessing chunking of point cloud terrain blocks then prioritizes and transfers point cloud chunks according to the current viewpoint at runtime.The implementation of this strategy ensures that prioritized-transfer data will not be massively rejected during the rendering phase.The experimental results show that allowing high precision data that has only partially completed transmission to participate in rendering can make the point cloud layer switching timely and smoothly,and the multi-LOD hybrid rendering algorithm further proposed in this thesis can make the features contained in high precision data better expressed in hybrid rendering;in terms of data transmission,the point cloud chunking pre-processing and coarse-grained visibility calculation at runtime adopted in this thesis on the server side can eliminate some of the vertices on the back side of the current viewpoint and reduce the total transmission volume.At the same time,the runtime prioritization strategy for point cloud chunking allows for better visible vertex rates and visual effects before the point cloud has been transferred.