Abstract:We present a novel high-quality vectorized modeling pipeline for large-scale scenes. With a reconstructed dense point cloud and its corresponding multi-view source images and camera parameters as input, our system can efficiently reconstruct a geometrically complete and detail-preserved vectorized model. Unlike most existing planar shape assembling methods which cannot handle large-scale vectorized modeling well due to the limitation of memory and computation, we can achieve complete high-quality vectorized modeling for complicated large-scale scenes in a time and memory efficient way. Our pipeline first carries out a 3D semantic segmentation on the dense point cloud, by performing 2D semantic labeling on the source images with a semantic segmentation network and fusing the 2D semantic labels into the point cloud. According to the fused dense 3D semantic labels, we then divide the scene into main structure including the grounds, walls and ceilings, and isolated objects that do not belong to the main structure. After the scene division completes, vectorized modeling is performed successively on the main structure and isolated objects to extract their polygonal models respectively instead of vectorizing the whole scene, to improve both time and memory efficiencies. Additionally, the previously vectorized main polygonal structures are used as priors to refine the segmentation and guide the vectorization of the objects to ensure the geometrical completeness and topological consistency of the entire vectorized model. Especially, during the vectorization procedure, a well designed binary space partition tree is designed to better slice the space so that high-quality polygonal mesh with more geometric details can be reconstructed with both time and memory efficiencies. Experiments with quantitative and qualitative evaluations on large-scale scenes demonstrate the accuracy and efficiency of the proposed vectorization pipeline. We also compare our method with state-of-the-art planar shape reconstruction methods to show its effectiveness in reconstructing large-scale vectorized models.

3D visualization method of large-scale vector data for operation

A Novel Pre-Process Method of Raw Point Data for Real-Time Complex Scenes Rendering

A Spatial Data Generalization Method for Online Vector Data 3d Visualization

Enhanced Real-time 2D Flow Visualization Based on Flow Lines Enhancement

Visualization Methods of Vector Data on a Digital Earth System

3D Integration Quick-speed Formation of DEM and Contours

A Processing and Scheduling Method for Vector Rendering in Global 3D GIS

A Quick and Feature based Visualization Algorithm for Large-scale Flow Data.

Efficient High-Quality Vectorized Modeling of Large-Scale Scenes

Importance Driven Texture-Based Vector Field Visualization

HiVision: Rapid visualization of large-scale spatial vector data

A Grid-Aided and STR-Tree-based Algorithm for Partitioning Vector Data

A method of the dense-sparse vector tile generation accounting for the spatial distribution of features

Parallel Visualization for Large-Scale Datasets

Accelerating the Computation of Multi-Scale Visual Curvature for Simplifying A Large Set of Polylines with Hadoop

Surface Carving-Based Automatic Volume Data Reduction

A novel data decomposition method for rapid parallel processing of vector polygon rasterization

Geometry-based Mapping and Rendering of Vector Data over LOD Phototextured 3D Terrain Models

Efficient and Dynamic Simplification of Line Drawings.

Parallel scanline algorithm for rapid rasterization of vector geographic data