stereo vision

Stereo vision is a way of measuring how far away things are by using two cameras side by side, the same trick your two eyes use. Hold up a finger and look at it with one eye closed, then the other: the finger seems to jump sideways against the background. Near things jump a lot; far things barely move. Stereo vision turns that jump into distance. Two cameras, mounted a fixed distance apart and pointing the same way, each take a picture of the scene at the same moment, and the system compares the two.

The amount each point seems to shift between the left and right image is called disparity, and it is the heart of the method. A point that lands far to one side in one image but well to the other side in the second has a large disparity, which means it is close to the cameras; a point that sits in almost the same place in both images has tiny disparity and is far away. Because the gap between the cameras (the baseline) and their focal length are known from calibration, simple geometry converts each disparity value directly into a real distance in meters. The hard part is matching: for every point in the left image, the system must find the very same point in the right image, which is where features and descriptors earn their keep.

Do this matching for many points across the whole image and you get a depth map — a picture where each pixel's value is not color but distance. That map lets a robot judge how far the wall, the table, or the person is, all from ordinary cameras with no laser or special lighting. The catch is that stereo needs visible texture to match on: a blank white wall gives nothing to line up, so depth there becomes unreliable, and accuracy fades for very distant objects whose disparity shrinks toward zero.