RealityCapture produces incorrect geometry when using images extracted from 360° panoramas.

General Feedback & Requests
1

Hello,

I tried to reconstruct a street using approximately 305 screenshots extracted from Google Earth 360 panoramas (Street View).

Instead of generating a proper 3D model, RealityCapture creates mostly flat wall-like geometry and projects the images onto those surfaces.

I tested multiple locations, not just a single street, and obtained similar results each time. The images have significant overlap and were captured following standard photogrammetry principles as closely as possible.

I am using RealityCapture 2.1.1.

Is this a known limitation when using images extracted from 360 panoramas, or could there be a workflow that improves the reconstruction quality?

Thank you.

2

Hi, I’m doing something similar and I was getting the big flat walls. Were you using FFmpeg to break up your images? Recently this new script / pipeline came out with the intention of making cube maps out of panos for use in RealityScan, so you might give it a shot. I’ve tried it and am at least getting fewer of the flat walls. But I think the variation in lighting in StreetView shots (even when taken on the same day) and the distance between panos along the street are a major obstacle. I have placed MANY control points and am still only getting mixed results. Let me know if this helps!