On benchmarking camera calibration and multi-view stereo for high resolution imagery

TLDR

The study focuses on two key challenges for outdoor 3‑D reconstruction: accurate camera calibration and dense multi‑view stereo. The authors aim to evaluate whether image‑based 3‑D modelling can replace LIDAR systems for outdoor 3‑D data acquisition. They collected outdoor scenes with both LIDAR and cameras, used LIDAR data as ground truth, and created evaluation sets to assess pose estimation, camera calibration, and multi‑view stereo.

Abstract

In this paper we want to start the discussion on whether image based 3-D modelling techniques can possibly be used to replace LIDAR systems for outdoor 3D data acquisition. Two main issues have to be addressed in this context: (i) camera calibration (internal and external) and (ii) dense multi-view stereo. To investigate both, we have acquired test data from outdoor scenes both with LIDAR and cameras. Using the LIDAR data as reference we estimated the ground-truth for several scenes. Evaluation sets are prepared to evaluate different aspects of 3D model building. These are: (i) pose estimation and multi-view stereo with known internal camera parameters; (ii) camera calibration and multi-view stereo with the raw images as the only input and (iii) multi-view stereo.