What Is a Stereo Camera?

A stereo camera captures two images of the same scene at the same moment, from two sensors mounted side by side. Software matches each point in the left image to the same point in the right image and turns the difference into distance, so the camera outputs depth for the whole scene, not just a flat picture.

It works the same way human vision does. Each eye sees the world from a slightly different position, and the brain uses the difference to judge distance. That is why stereo vision cameras are a common choice for robots, autonomous vehicles and mapping systems that need to judge how far away things are while they move. A single stereo camera gives them a color image, a depth map and a 3D point cloud from one device.

The distance between the two sensors is called the baseline. A nearby object appears in quite different positions in the two images, while a distant object barely moves. That shift is the disparity, and the camera triangulates depth from it: depth equals the focal length times the baseline, divided by the disparity.

Two consequences follow. First, a longer baseline resolves depth further away, and a shorter baseline works closer in, so the baseline is chosen for the working range. Second, depth error grows roughly with the square of the distance, so a stereo camera is very precise up close and less precise at the far end of its range.

The Stereolabs range shows the trade-off clearly. The wide-baseline ZED X measures out to around 35 m with its narrow lenses, while the short-baseline ZED X Nano measures from 3 cm to 2 m on a robot wrist. Modern stereo cameras also use neural networks to fill in surfaces with little texture, such as plain walls, where simple pixel matching struggles.

Stereo is one of several ways to measure depth, and each suits different conditions.

Stereo’s main advantages are that it is passive, so sunlight helps rather than hurts, and that depth and color come from the same sensors, already aligned. Its main limits are darkness and featureless surfaces.

Stereo depth suits machines that move through changing, often outdoor, environments, where an active sensor may struggle in sunlight and a single camera cannot measure distance.

We pair depth cameras with NVIDIA Jetson computers for mobile robots, robot arms, agricultural machines and outdoor inspection, choosing the baseline, lens and interface to suit the distances the machine has to measure. Our stereo cameras include the Stereolabs ZED range in USB and GMSL2 versions.

Choosing a stereo camera comes down to four questions. What is the nearest and furthest distance the machine needs to measure, which sets the baseline? How wide a view does it need, which sets the lens? Is the camera or the scene moving fast, in which case global shutter sensors avoid distorted depth? And how far is the camera from the computer, which decides between USB and GMSL2? Depth processing also takes GPU time, so the Jetson module has to be sized for the number of cameras and the AI models running alongside.

For machines in the field we usually specify GMSL2 stereo cameras, which carry video and power over one locking coax cable. Our guide to the ZED X GMSL camera interface explains the hardware side, and our guide to setting up a ZED X robotics stack with ROS 2 covers the software, from the camera driver to the depth and point cloud topics your robot reads.

We supply stereo cameras with Jetson computers as standard, with the drivers, SDK and ROS 2 packages installed and the depth output validated before the system ships.