This paper makes the following contributions. We
present and evaluate a system for large-scale 3D reconstruction
of outdoor scenes that runs at interactive frame rates
on modern mobile devices. Our system uses monochrome
fisheye images and thus does not come with the limitations
of systems using active depth cameras, such as a restricted
depth range and sensitivity to background illumination. We
demonstrate the importance of filtering outliers in the depth
maps and propose multiple filtering steps, which despite
their computational simplicity significantly improve the reconstruction
quality without affecting the real-time performance.
An extensive experimental evaluation on both synthetic
and real data shows that our system compares favorably
against active sensor based approaches.