Abstract - Structure from motion is a common approach to generate 3D models of objects and structures. The ease of the data acquisition and the wide selection of available algorithms makes the technique easily accessible. Previous benchmarks on the topic have been focused on scanning small structures, specially for archaeology, or have been limited to single or very few algorithms.
In this work different algorithms are benchmarked regarding accuracy and processing time for datasets acquired in urban environments, with the goal of analyzing the feasibility of utilizing this technique on search and rescue operations. Real-world rescue scenarios are demanding due to the presence of challenging surfaces and smoke.
For evaluation we recorded two datasets at the Bavarian Firefighter School in Würzburg, Germany.
Practice Hall | Burning House |
Visual SfM | ||||
Visual SfM and openMVS | ||||
Visual SfM and MVE | ||||
openMVG and openMVS | ||||
openMVG and MVE | ||||
MVE | ||||
COLMAP | ||||
AgiSoft PhotoScan |
Visual SfM | ||||
Visual SfM and openMVS | ||||
Visual SfM and MVE | ||||
openMVG and openMVS | ||||
openMVG and MVE | ||||
MVE | ||||
COLMAP | ||||
AgiSoft PhotoScan |