Structure-from-motion (sfm) sequencing using non-visual information

Abstract

Disclosed are techniques for image processing. For example, a computing device can obtain a first video of a scene (including a plurality of first image frames), first non-visual data along a first route, a second video of the scene (including a plurality of second image frames), and second non-visual data along a second route. The computing device can compare the first non-visual data with the second non-visual data and determine first image frame(s) that are similar to second image frame(s). The computing device can determine pairs of image frames from the first image frame(s) and the second image frame(s) based on image data of the scene in the first and second image frame(s). The computing device can process the pairs of image frames to reconstruct the scene with a reconstruction of an overlap of the first route and the second route.