Stanford CS223B Computer Vision, Winter 2006 Lecture 1 Intro and Image Formation Professor Sebastian Thrun CAs: Dan Maynes-Aminzade and Mitul Saha Guest lectures: Rick Szeliski (Microsoft Research) and Gary Bradski (Intel Research and Stanford)
Today’s Goals Learn about CS223b Get Excited about Computer Vision Learn about Image Formation (Part 1)
Administrativa Time and Location Web site Mon/Wed 11:00-12:20, McCullough 115 SCPD Televised Web site http://cs223b.stanford.edu
People Involved You: 90 students signed up Me: Sebastian Thrun Office hours Wed 3-4 with appointment, Gates 154 CA: Mitul Saha office hours Tue, Thu 3-5pm, Clark S264 CA: Dan Maynes-Aminzade office hours Mon, Wed 3-5pm, Gates 386 Guest lectureres Gary Bradski, Intel Research and Stanford Rick Szeliski, Microsoft Research
Guest Lecturers
Goals To familiarize you with basic the techniques and jargon in the field To enable you to solve computer vision problems To let you experience (and appreciate!) the difficulties of real-world computer vision To get you excited!
Course Requirements + Criteria You have to Turn in all assignments (30% of final grade) Pass the midterm (30%) Pass the competition (40%) Late policy Six late days Teaming: Assignments/competition: up to three students
Course Overview Basics 3D Reconstruction Motion Image Formation and Camera Calibration Image Features Calibration 3D Reconstruction Stereo Image Mosaics, Stiching Motion Optical Flow Structure From Motion Tracking Object detection and recognition Grouping Detection Segmentation Classification
Course Overview
The Text
The Class Competition No Major Project Instead: A competition
The Competition: Motivation
Implications Why not run a competition in CS223b?
The Competition March 13, 11-11:30am: The Competition Given a stream of images acquired by a vehicle on a highway Predict a classification of moving/non moving objects 5 seconds ahead Same data used in all programming assignments HW1: Feature/object detection (due Jan 23) HW2: Camera calibration (due Jan 30) HW3: Visual odometry (due Feb 13)
The Competition, Example This is not the real data. We’ll collect the data with Stanley
Today’s Goals Learn about CS223b Get Excited about Computer Vision Learn about image formation (Part 1)
Computer Graphics Output Image Model Synthetic Camera (slides courtesy of Michael Cohen)
Computer Vision Output Model Real Scene Real Cameras (slides courtesy of Michael Cohen)
Combined Output Image Model Real Scene Synthetic Camera Real Cameras (slides courtesy of Michael Cohen)
Example 1:Stereo See http://schwehr.org/photoRealVR/example.html
Example 2: Structure From Motion http://medic.rad.jhmi.edu/pbazin/perso/Research/SfMvideo.html
Example 3: 3D Modeling http://www.photogrammetry.ethz.ch/research/cause/3dreconstruction3.html
Example 4: 3D Modeling Drago Anguelov
Example 4: 3D Modeling
Example 4: 3D Modeling
Example 5: Segmentation http://elib.cs.berkeley.edu/photos/classify/
Example 6: Classification
Example 6: Classification
Example 7: Optical Flow Demo: Dirt Road Andrew Lookingbill, David Lieb, CS223b Winter 2004
Example 8: Detection David Stavens, Andrew Lookingbill, David Lieb, CS223b Winter 2004
Example 9: Tracking http://www.seeingmachines.com/facelab.htm
Example 10: Human Vision
Example 9: Human Vision
Excited Yet?
Today’s Goals Learn about CS223b Get Excited about Computer Vision Learn about image formation (Part 1)
Topics Pinhole Camera Orthographic Projection Perspective Camera Model Weak-Perspective Camera Model
Pinhole Camera -- Brunelleschi, XVth Century “shards of diamond” – those little pieces of beautiful truth scattered about. -- Brunelleschi, XVth Century *many slides in this lecture from Marc Pollefeys comp256, Lect 2
Perspective Projection Derive the perspective equations on notes page 1.1 A “similar triangle’s” approach to vision. Marc Pollefeys
Perspective Projection X x -x f Z f
Consequences: Parallel lines meet There exist vanishing points Marc Pollefeys
The Effect of Perspective
Vanishing points Different directions correspond VPL VPR VP1 VP2 Different directions correspond to different vanishing points VP3 Marc Pollefeys
Implications For Perception* Same size things get smaller, we hardly notice… Parallel lines meet at a point… * A Cartoon Epistemology: http://cns-alumni.bu.edu/~slehar/cartoonepist/cartoonepist.html
Perspective Projection X -x Z f
Weak Perspective Projection Z O -x Z Z f
Generalization of Orthographic Projection When the camera is at a (roughly constant) distance from the scene, take m=1. Marc Pollefeys
Pictorial Comparison Weak perspective Perspective Marc Pollefeys
Summary: Perspective Laws Weak perspective Orthographic
Limits for pinhole cameras