基于Django技术和MVS网络的无人机遥感数据三维重建可视化平台设计

Design of a 3D reconstruction visualization platform for unmanned aerial vehicle remote sensing data based on Django technology and MVS network

摘要:

为了提升无人机遥感图像数据可视化效果,设计基于Django框架技术的无人机遥感数据三维重建可视化平台。基于Django技术,搭建由浏览层、业务逻辑层与数据层构成的无人机遥感图像数据可视化平台框架。平台硬件设计上,设计WEB服务器搭载业务层的view.py、GeoDjango地理框架、URL分析器以及JBrowse,在服务器中创新性地加设评估板用于监测主控制器芯片的工作程序,保证数据可视化处理的运行效率,并加装寄存器组作为外置存储器,降低GPU内存占用量；设计由GPU、S显卡、显示硬件构成的三维渲染引擎,在GPU芯片顶点着色器上加装像素处理器,利用帧缓冲和帧渲染提升三维渲染质量。平台软件设计上,利用SQLite数据库为数据编码；利用最近邻域重采样算法对无人机遥感图像数据进行畸变改正；基于摄影测量解析提取无人机遥感图像目标几何信息。设计基于分组相关性与多尺度特征的MVS网络,通过深度图估计,实施无人机遥感图像的三维重建,完成无人机遥感图像数据可视化平台的设计。测试结果表明,该平台在三个区域的GPU内存占用量均低于5000MB,整体深度图平滑度较高,整体建模时间为8245ms,渲染时间为5067ms,具有较好的可视化效果。

Abstract:

In order to improve the visualization effect of drone remote sensing image data, a 3D reconstruction visualization platform for drone remote sensing data based on Django framework technology is designed. Building a unmanned aerial vehicle remote sensing image data visualization platform framework based on Django technology, consisting of a browsing layer, a business logic layer, and a data layer. In terms of platform hardware design, a web server is designed to be equipped with the business layer"s view. py, GeoDjango geographic framework, URL analyzer, and JBrowse. An innovative evaluation board is added to the server to monitor the working program of the main controller chip, ensuring the efficiency of data visualization processing. Register groups are also installed as external memory to reduce GPU memory usage; Design a 3D rendering engine consisting of GPU, S-graphics card, and display hardware, install pixel processors on GPU chip vertex shaders, and improve 3D rendering quality through frame buffering and frame rendering. In terms of platform software design, SQLite database is used for data encoding; Using nearest neighbor resampling algorithm to correct distortion in unmanned aerial vehicle remote sensing image data; Extracting target geometric information from unmanned aerial vehicle remote sensing images based on photogrammetric analysis. Design an MVS network based on group correlation and multi-scale features, implement 3D reconstruction of unmanned aerial vehicle remote sensing images through depth map estimation, and complete the design of a unmanned aerial vehicle remote sensing image data visualization platform. The test results show that the GPU memory usage of the platform in all three regions is less than 5000MB, and the overall depth map smoothness is high. The overall modeling time is 8245ms, and the rendering time is 5067ms, which has a good visualization effect.