Glossary · Machine vision
Stereo vision
Also known as: Stereoscopic vision, Binocular vision
German: Stereovision
In 3D vision, stereo vision computes depth by comparing images from two or more cameras viewing the scene from different positions; the displacement (disparity) of corresponding points between the images gives their distance.
- Machine vision
In one sentence
Stereo vision computes depth from the disparity between images of two or more cameras viewing the scene from different positions.
Example
A stereo camera on the mobile robot estimates the distance to pallets so it can align its forks before lifting.
How it applies
- Engineering: Passive stereo needs surface texture to find corresponding points; active stereo adds a projected pattern for textureless surfaces. Depth accuracy decreases with distance and increases with camera baseline. The computation of disparity maps is processing-intensive, so many stereo cameras compute depth on board.
- Commissioning: The cameras must be calibrated to each other and mounted rigidly; any shift degrades depth measurement.
- Documentation: The documentation team should document calibration, the measuring range and accuracy validated for the application, and any laser or projector class if active illumination is used.
Stereo vision vs. structured light
Stereo vision relies on two viewpoints and scene texture. Structured light uses a projector and usually one camera, creating its own texture, which works better on uniform surfaces.