What Is Stereo Vision? How Two Cameras Measure Distance.

What Is Stereo Vision?

Stereo vision is a computer vision method that estimates depth by comparing images captured from two different viewpoints.

The approach resembles the way human binocular vision provides information about distance.

In robotics, stereo cameras are used for navigation, obstacle detection, three-dimensional reconstruction, and object manipulation.

For example, a mobile robot can compare two camera images to estimate how far nearby boxes and other objects are from its cameras.

Key Takeaways

• Stereo vision estimates depth using images from separated viewpoints.
• It relies on identifying corresponding features in both images.
• Accuracy depends on camera calibration, separation, image quality, and the ability to match visual details reliably.

How Does Stereo Vision Work?

Imagine two cameras mounted side by side on a robot.

Both cameras capture the same scene, but from slightly different positions.

A nearby object appears at different horizontal locations in the two images.

A distant object generally shows a smaller positional difference.

Software identifies corresponding points in the images.

It measures their displacement, called disparity.

Using the cameras' known geometry, the system estimates depth.

This allows the robot to construct a spatial representation of visible objects.

What Is Disparity?

Disparity describes the difference between the image locations of a corresponding scene point in the two camera views.

For a simplified calibrated, rectified stereo system:

Depth = Focal Length × Baseline ÷ Disparity

The baseline is the distance between the camera centers.

The focal length and disparity must use compatible units, commonly pixels for the image measurements.

For example, suppose:

• Focal length: 800 pixels
• Baseline: 0.1 meters
• Disparity: 40 pixels

The estimated depth is:

800 × 0.1 ÷ 40 = 2 meters

This is an idealized calculation. Real measurements also depend on calibration and image correspondence accuracy.

Stereo Vision vs. Monocular Vision

Monocular vision uses a single camera viewpoint.

Stereo vision uses two viewpoints.

A monocular image can identify objects and provide useful visual information.

However, geometric depth from one ordinary image is generally ambiguous without additional information.

Stereo vision uses the known distance between cameras to estimate depth through triangulation.

The additional camera and calibration requirements introduce hardware and computing costs.

Stereo Vision vs. LiDAR

Stereo vision estimates distance through image relationships.

LiDAR measures distances using light-based ranging methods.

Stereo systems can provide rich visual information and dense depth estimates under suitable conditions.

LiDAR can provide direct spatial measurements.

Both technologies have limitations involving range, environmental conditions, and difficult surfaces.

Some robots use both.

Why Does Stereo Vision Struggle With Certain Objects?

Stereo matching requires useful visual features.

A plain white wall may have little texture for identifying corresponding points.

Reflective or transparent objects can also create difficulties.

Low lighting, motion blur, and poor calibration reduce measurement quality.

Stereo depth uncertainty generally increases with distance under a fixed camera configuration.

Frequently Asked Questions

Does stereo vision require artificial intelligence?

No. Classical stereo-matching algorithms can estimate depth using geometric and image-processing methods.

Can stereo cameras work in darkness?

Passive stereo cameras need usable scene illumination. Some systems use infrared illumination or projected patterns to improve measurements.

The Bottom Line

Stereo vision uses two viewpoints to estimate the depth of objects in a scene. It provides robots with useful three-dimensional information using cameras, although accurate results depend on calibration and reliable image matching.