3D Reconstruction and Mapping

in Computer Vision, Robotics, and Augmented Reality

Fall 2012 , T-Th 9.30-11.00

Professor: Frank Dellaert Office Hours TBA

“TA”s: Jason Antico and Paritosh Mohan

Prerequisites

Linear algebra, esp. the Singular Value Decomposition (SVD). Familiarity with graphical models is a plus but not a must.

Emailing about the class:

Please use “CS8803” in the subject line, automatic by clicking the links:

Email Frank: dellaert@cc.gatech.edu

Class Goals

The desired learning outcomes for the students are:

1. -know what SLAM, SFM are
   

2. -know graphical model inference
   

3. -know its implementation using linear algebra
   

- understand the practical issues regarding multi-platform reconstruction

Organization

The course is organized into five main sections:

1. -Reasoning about images and scenes (feature detection, descriptors, clustering, etc.)
   

2. -Data association, and basic multi-view geometry
   

3. -Visual odometry, Structure from Motion, and Bundle Adjustment
   

4. -Dense multi-view reconstruction
   

5. -Student Projects
   

Sensors and Platforms:

1. -Kinect and other range sensors
   

2. -Robots, such as UAVs, UGVs, and AUVs
   

3. -Community Photo Collections
   

Text

I will not be using a required textbook. Instead I will be handing out notes and we will be readings some papers. Good books to have as reference are:

- Multiple View Geometry in Computer Vision, by Richard Hartley and Andrew Zisserman. Cambridge University Press.

Materials

I will be using the blackboard a lot, rather than powerpoint. Students are expected to take notes and will be asked to actively participate as scribes.

Assignments

There will be a series of programming assignments to provide familiarity with the techniques we learn in class. Some of these will be done in teams, others individually. In addition, there will be a team-based final project where the goal is to demonstrate multi-platform 3D reconstruction and/or mapping. Each team will be asked to present this project in class.

In addition to these, there will be a couple of small exercises to prime your thoughts about the topic of the next week, or quizzes to assess your understanding. They are graded at about 1% each and are not expected to be much work at all.

Finally, each student will be asked to read and present research papers in class when we are discussing state of the art where lecture style is less appropriate.

Collaboration Policy: Collaboration on assignments is encouraged at the "white board" level. That is, share ideas and technical conversation, but write your own code. Students are expected to abide by the Georgia Tech Honor Code. Honest and ethical behavior is expected at all times. All incidents of suspected dishonesty will be reported to and handled by the Dean of Students.

Late Policy: 1 day late: 50% of the grade, 2 days late: 25% of the grade, later than that: 0% of the grade. It is always to ask prior approval to hand in an assignment late because of special reasons.

Grading

Project: 35%
Assignments: 25%

Midterm & Final Exam: 10% each

Exercises/Quizzes: 10%

Paper Presentation: 5%
Class Attendance & Participation: 5%