Stereo Vision

Stereo Vision

Principles

• Mimicking human vision.
• Two or more cameras.
• Triangulation. Find corresponding points in each camera.
• Disparity is proportional to depth.

Key considerations

• Passive, no active lighting required.
• Long range and outdoor usage.
• Sensitive to flat texture-less scenes.
• Less suitable for dense point cloud representations.
• Region based correspondence, block averaging.
• Look out for correspondence problem.

Active Stereo Vision

Principles

• Same as stereo vision.
• Additional random pattern projector (RPP).
• Mimicking human vision.

Key considerations

• Same as stereo vision, but active source.
• Shorter range
• Sensitive to ambient / sunlight.
• No texture on surfaces required.
• Solves correspondence problem.

Compare stereo and structured light vision point clouds.

Stereo- and active stereo cameras are often used by AI algorithms in bin-picking applications. Here are examples of a stereo camera point cloud and the same scene captured with a time-coded structured-light camera (like Zivid). 

Keep learning!

Read the whole article about active and passive 3D vision principles here, including stereo, active stereo, laser, ToF, and structured light machine vision.