Embodied AI Glossary中文

Depth Camera

深度相机RGB-DEssential

A camera that outputs, for every pixel, both a color and the distance from that point to the camera.

A depth camera, also called an RGB-D camera, outputs a depth map (D) alongside an ordinary color image (RGB). Three ranging principles dominate: stereo vision, where two lenses view the same point and triangulate distance from the disparity between them, often paired with an infrared projector casting a random texture onto the scene to help matching on textureless surfaces like white walls (called active stereo); structured light, which projects a known pattern and computes depth from how it deforms; and time-of-flight (ToF), which measures how long light takes to travel out and back. Depth cameras are cheaper and smaller than lidar and more precise at close range, but error grows with distance, and transparent or mirror-like objects are often measured incorrectly. They're commonly mounted on a robot's head or wrist to produce point clouds and support grasping and obstacle avoidance; common brands include RealSense, Orbbec, and ZED.

ExampleThe RealSense D435i is an active-stereo camera officially rated for an ideal working range of 0.3–3 meters, with error under 2% at 2 meters, and is widely used in tabletop manipulation and mobile-robot research.

Also called
RGB-D Camera, 3D Camera, Range Camera
Related
Depth Map · Active Stereo · Structured Light · Time of Flight · Point Cloud · RealSense Depth Camera (D435i / D405)
Sources
Wikipedia: Range imaging(深度相机的几类原理) (Chinese)
RealSense 白皮书:Projectors for D400 series(主动双目与红外投射器) (Chinese)
RealSense D435i 产品页 (Chinese)

See it in the full glossary →