Variable-Baseline Depth Sensing

What happens to depth estimation when the cameras are very far apart?

Role
Hardware prototyping, enclosure, lens calibration, disparity-mapping algorithm, demonstration and report
Areas
Computer Vision, Hardware, Research, Optics

Most commercial depth sensors use some form of disparity mapping. They compare two viewpoints, either two cameras or a pattern emitter and a camera, usually a few millimeters or centimeters apart.

A two-column comparison. A typical depth sensor compares two viewpoints, two cameras or an emitter and a camera, millimeters to centimeters apart. This rig used three cameras, two with the same field of view and one wider, ultrawide and macro lenses, and extremely wide spacing between sensors.
The study moved every variable outside the usual comfort zone, above all the spacing between sensors.

This exploration measured what changes outside that comfort zone:

  • Three cameras: two with the same field of view and one wider.
  • Ultrawide and macro lenses.
  • Above all, extremely wide spacing between the sensors.

We built the hardware and enclosure, the lens calibration stack and a custom disparity-mapping algorithm, and delivered a demonstration and a report.

A five-step flow: the three-camera hardware prototype, a custom enclosure, a lens calibration stack, a custom disparity-mapping algorithm, and a demonstration and report.
The full stack built for the study, from hardware to report.

Part of Enhanced Presence.