Measurement of Three-Dimensional Environment with a Fish-Eye Camera Based on Structure from Motion - Error Analysis
Authors: Kenji Terabayashi & Co
Year: 2009
- Use a polynomial fish eye camera model rf = k1 theta + k3 theta^3 + k4 theta^5
- Feature points tracked with KLT or SIFT
- ray vector p = [sin t cos p, sin t sin p, cos t]
- E (essential matrix) computed with 8 points algorithms and RANSAC
- Translation and rotation obtained with essential matrix
- skew lines = lines that neither intersect neither are parallel
- Triangulation with mid point method
- Inaccurate point removal with partial derivative
- Bundle adjsutement in a RANSAC fashion for less computation
Super papier de structure from motion, big up!