Outdoor 3D Reconstruction Using Monocular Depth Estimation and Photogrammetric Point Clouds

Augmented reality (AR) is a rapidly evolving field, yet research has predominantly focused on indoor applications, leaving outdoor environments relatively underexplored. Metric Depth Estimation (MDE) plays a pivotal role in AR, enabling essential functionalities such as object placement and occlusion handling by extracting depth and perspective information from single 2D images.  This repository contains data and code for evaluating 3D reconstruction methods using monocular depth estimation and photogrammetry. The data includes single outdoor images used for generating point clouds through monocular depth estimation models as well as a set of images with a resolution of 480x853 pixels, used to create reference point clouds via photogrammetry techniques. Scripts and tools for processing the images, generating point clouds are also included.

增强现实（Augmented reality, AR）是一门快速发展的领域，但当前研究大多聚焦于室内应用，室外环境的相关探索仍相对不足。度量深度估计（Metric Depth Estimation, MDE）在AR中发挥着关键作用，通过从单张二维图像中提取深度与透视信息，可实现物体摆放、遮挡处理等核心功能。本代码仓库包含了利用单目深度估计（monocular depth estimation）与摄影测量（photogrammetry）技术评估三维重建方法的数据集与代码。数据集包含两类内容：一是用于通过单目深度估计模型生成点云的室外单张图像，二是分辨率为480×853像素的图像集，后者可通过摄影测量技术生成参考点云。此外还附带了用于处理图像、生成点云的脚本与工具。