Epipolar geometry is the math relating two camera views of the same scene — it constrains where a point in one image can appear in the other, the foundation of stereo, structure-from-motion, and visual SLAM.
Epipolar geometry is the rule that, if you see a point in one camera image, it must lie along a specific line in a second camera's image. That constraint hugely simplifies matching points between two views and recovering 3D.
🎯 Quick challenge
Epipolar geometry says a point seen in one image must lie, in the other image, on…
When two cameras (or one camera at two moments) view the same scene, there's a beautiful, strict relationship between the two images. Epipolar geometry captures it — and it's the mathematical backbone of stereo, 3D reconstruction, and visual SLAM.
The core constraint
Here's the key insight: if you see a point in one image, its match in the other image can't be just anywhere — it must lie along a specific line, called the epipolar line. This is because the point, and the two camera centers, all lie in one plane (the epipolar plane), which slices each image in a line.
The practical payoff is huge: searching for a match drops from 2D (the whole image) to 1D (along one line) — faster and far less error-prone.
One point, one line
The geometry of two views forces a point's match onto a single line, turning a hard 2D search into an easy 1D one and rejecting impossible matches.
How it's encoded
The relationship between the two views is captured in a matrix:
The fundamental matrix (F) relates the two images in pixel coordinates (uncalibrated cameras).
The essential matrix (E) does so for calibrated cameras, and can be decomposed into the relative rotation and translation between the two views.
That last point is the magic: from matched features and the essential matrix, you recover how the camera moved — the heart of visual odometry and structure from motion.
Where it's used
Stereo vision. Rectifying the two images aligns epipolar lines to image rows, so matching is a simple horizontal search — the basis of disparity/depth.
Feature matching verification. RANSAC fits the epipolar geometry to reject wrong matches (outliers) — the standard robustness step.
Motion estimation. Recovering camera pose from two views in SLAM and SfM.
Why it matters
Epipolar geometry is one of the pillars of multi-view computer vision. It provides the constraint that makes matching tractable and the mathematics that recovers 3D structure and camera motion from images. Nearly every camera-based robotic 3D capability — stereo depth, visual odometry, reconstruction — rests on it.
