Embodied AI Glossary中文

Stereo Camera

双目相机Essential

Two side-by-side cameras that shoot the same scene at once, computing distance from how much the two images differ.

A stereo camera mounts two cameras side by side at a fixed separation, called the baseline, mimicking human eyes. The same physical point appears at a slightly different horizontal position in the left and right images — the disparity — and after calibration and epipolar rectification, stereo matching finds corresponding points and computes depth as Z = f·B/d, where f is focal length, B is the baseline, and d is the disparity: a larger disparity means closer, and a longer baseline gives more accurate readings at range. Stereo that relies only on ambient light is called passive stereo, and it struggles on textureless surfaces such as white walls or smooth tabletops where matching fails; active stereo cameras such as the RealSense D435 and D455 add an infrared speckle projector to work around this. Recent work also uses deep networks for the matching step, such as FoundationStereo. Stereo cameras commonly serve as a robot's head- or wrist-mounted depth sensor.

ExampleA stereo camera with a 700-pixel focal length and a 0.12-meter baseline observes a 42-pixel disparity at some point, giving a depth of Z = 700 × 0.12 ÷ 42 = 2 meters. Stereolabs' ZED 2i has a baseline of 120 mm.

Also called
Stereo Vision, Passive Stereo
Related
Disparity · Stereo Matching · Stereo Baseline · Active Stereo · Depth Camera · FoundationStereo
Sources
Wikipedia: Computer stereo vision
Stereolabs ZED 2i product page

See it in the full glossary →