RANSAC and Robust Verification
Minimal samples, residual thresholds, and adaptive iteration counts for fitting models among outliers.
Key Terms
Minimal SampleInlier ThresholdConsensus SetAdaptive IterationsConfidenceMSAC / PROSAC
Python Tools
OpenCVNumPyMatplotlib
Textbook Reference
Szeliski Ch.8Hartley & Zisserman Ch.4
Topics Covered
- 1Outlier-contaminated correspondences and the breakdown of least squares
- 2Minimal sample sets for lines, homographies, and fundamental matrices
- 3Residual definitions and choosing an inlier threshold from noise
- 4Probability of success and the adaptive iteration-count formula
- 5Refinement on the consensus set and variants such as MSAC and PROSAC
Learning Outcomes
- Derive the RANSAC iteration count from inlier ratio and target confidence
- Choose residuals and thresholds appropriate to a geometric model
- Combine RANSAC with a final least-squares refit on the inliers
Robust estimation lab
RANSAC & Robust Geometric Verification
Observe sampling, model fitting, reprojection scoring, consensus growth, and final inlier-only refinement.
Randomly select the minimum number of matches.
weak
● detected match◇ ambiguous / sampled━ verified inlier┄ × rejected outlier
Live statistics
Iteration
1 / 60
Sample size
4
Adaptive N
60
Current inliers
4 (10%)
Best inliers
4 (10%)
Degenerate
0
Current mean e
0.000
Best mean e
0.000
Success prob.
0%
Homography H
eᵢ < τ → inlier-0.821-0.0610.431-1.1400.0300.566-1.815-0.1601.000
p=0.990 · w=0.10 · s=4 · N≈55974
Lecture PDFs
RANSAC and Robust Verification
The embedded PDF preview is disabled to reduce Vercel bandwidth usage. Open the lecture file directly in Google Drive for viewing, downloading, or presenting.