Depth Map

An image whose pixel values store distance to the camera instead of color.

A depth map is a single-channel image where each pixel's value represents the distance from the corresponding scene point to the camera — generally measured along the camera's optical axis (the Z axis), not the straight-line distance to the lens. It can come directly from a depth camera's measurement, or be predicted from an ordinary color image by a monocular depth estimation model such as Depth Anything, which sometimes outputs only relative distance and sometimes metric depth in meters. Depth is commonly stored as a 16-bit integer that must be multiplied by a depth unit to get meters. A depth map combined with the camera's intrinsics can be back-projected, pixel by pixel, into a point cloud. Common problems include holes where nothing was measured (value 0), ‘flying pixels’ at object edges, and a mismatch between the depth map's and the color image's viewpoints, which needs to be corrected by alignment before the two are used together.

ExampleIf the depth unit is 1 millimeter, a pixel value of 850 in the depth map means that point is 0.85 meters from the camera along the optical axis; a value of 0 usually means nothing was measured there.

Also called

Depth Image, Range Image

Related

Depth Camera · Point Cloud · Projection / Back-Projection · Depth Holes · Monocular Depth Estimation · Depth-to-Color Alignment

Sources

Wikipedia: Depth map
RealSense SDK 文档：Depth from Stereo（视差转深度与深度单位） (Chinese)