Stereo Vision Egon Elbre Hans Mesalu general stuff about this 3D - PowerPoint PPT Presentation

Stereo Vision Egon Elbre Hans Mäesalu

general stuff about this 3D thing

why?

applications games movies simulators robotics product presentations architecture visualization virtual television studios virtual presence for video communications general virtual reality applications

Single-View Geometry

orthographic projection

perspective projection

IRL

simple camera projection

just think of it as magic

extrinsic/intrinsic camera calibration matrix

finding 3D point with triangulation assuming we know where the cameras are

finding a depth map

image rectification

rectification gives us finding matching points simpler

step 1 - get the pictures

step 2 - find some interesting points

step 3 - guess similar points

step 4 - remove outliers (RANSAC)

RANSAC RANdom SAmpling Consensus

rectify

disparity map

demo video

Introduction to Epipolar Geometry

terminology

F(undamental)-matrix

E(ssential)-matrix

and finally...

reconstruction from two views only note! can't be done uniquely due to some ambiguity 1. identify a number (at least 8) of point correspondences 2. form linear equations based on x' T Fx=0 formula 3. find the solution F for those equations 4. compute P, P' camera matrices from F 5. given to cameras P, P' and corresponding point pairs triangulate the 3D point X we know how to do 1 and 5 we won't discuss 2, 3 as it's about solving some linear equations and no one will remember it after the lecture anyway about 4 - well that's complicated

further reading "Multiple View Geometry in Computer Vision" , Richard Hartley and Andrew Zisserman "Uncertain Projective Geometry: Statistical Reasoning for Polyhedral Object Reconstruction" , Stephan Heuel "Computer Vision: Algorithms and Applications", Richard Szelinski "Learning OpenCV" , Gary Bradski and Adrian Kaehler