Stereo Vision & Disparity

Two eyes, one depth

Where f is focal length (in pixels), B is the baseline (distance between cameras), and d is the disparity (how far the same point shifts between left and right images).

• Closer objects → larger disparity → can resolve depth precisely
• Far objects → tiny disparity → quantization kills depth precision
• Disparity = 0 → object is at infinity (sky, far buildings)

Visualizer · stereo pair

Move the object slider to bring it closer/farther. The two camera views show the same scene from slightly different angles. The disparity map (right) shows the per-pixel shift — bright = close, dark = far.

Why baseline matters

A wider baseline (cameras spread further apart) gives you more disparity per unit depth — better for far objects. But it also shrinks the overlap region between left and right views. Stereo rigs are a compromise:

• 2 cm baseline — phone cameras, near-field AR (10 cm – 1 m)
• 6–12 cm baseline — robotics indoor (0.5–5 m)
• 50 cm+ — automotive, drones (10–100 m)

Active stereo (Intel RealSense, Microsoft Kinect) projects an IR speckle pattern so even blank walls become matchable. ToF (time-of-flight) skips the matching problem entirely by measuring how long the light takes to come back — but it's noisier and lower resolution.