Three techniques dominate. Stereo cameras compare two lenses like human eyes. Structured light projects a known dot pattern and reads its distortion. Time-of-flight sensors measure the travel time of infrared light per pixel.
The output is an RGB-D image: a normal colour picture plus a depth value per pixel, which software converts into a point cloud. From there a robot can segment a box from the pallet, measure the height of a stack or plan where to put the fingers of a gripper.
Limits are familiar to anyone who has used one: black, shiny or transparent surfaces return poor depth, strong sunlight drowns infrared projectors, and accuracy falls with distance. Widely used models include the Intel RealSense family in research and bin-picking cells.
