Teaching



Fall 2023 - Vision Algorithms for Mobile Robotics

UZH-DINF2039 / ETH-151-0632-00L



Registration for ETH students


This course is a core course of the ETH Master in Robotics, Systems and Control. Registration for ETH students is done through UZH. To do so, you will first need to register to get a UZH account. Afterwards, you can login and book the course. All the instructions can be found here.

ETH students should mind the following deadlines: the deadline to register for a UZH account is September 15 at 23:59hrs Zurich time. Once you receive the login credentials (it may take a few days), you can finally book the course until October 10 at 23:59. NB. If you fail to book the course, you won't be able to take the exam..



Exam cancelation


At UZH, registration for a course or cancelation of an examination is carried out directly in the module booking tool. Booking a module also automatically registers the student for the corresponding examination. Once the module booking deadline (October 10 23:59) has passed, absolutely no further registrations or cancelations can be processed. Cancelations at a later date due to illness are only possible if a petition is submitted within five business days from the exam, accompanied by a properly dated original doctor's attestation. More info here.



Goal of the Course


For a robot to be autonomous, it has to perceive and understand the world around it. This course introduces you to the key computer vision algorithms used in mobile robotics, such as image formation, filtering, feature extraction, multiple view geometry, dense reconstruction, tracking, image retrieval, event-based vision, visual-inertial odometry, Simultaneous Localization And Mapping (SLAM), and some basics of deep learning. Specifically, we will see the algorithms behind Apple ARKit, Google Visual Positioning Service, Microsoft Hololens, Magic Leap, Oculus Quest and Oculus Insight, the Mars Perseverance rover, and the Mars helicopter. Basic knowledge of algebra, geometry, and matrix calculus (product, inversion, eigenvalue decomposition) is required.




Time and location


All lectures and exercises will be taking place on-site in the UZH City campus (next to UZH and ETH main buildings) in SOC-F-106 (located on floor F) and will also be video recorded. If the front door of the building is locked, please use the side door (located on the right side of the building).


Lectures: every Thursday from 8:00 to 9:45 am. Please notice that the lecture starts at 8 o'clock and not 8:15!


Exercises: every Thursday from 12:15 to 13:45.


Please check out the course agenda below for the exact schedule.


Instructors


Lecture Instructor: Prof. Dr. Davide Scaramuzza.

Exercise Instructors: Leonard Bauersfeld and Jiaxu Xing.

Office hours of Prof. Scaramuzza: Every Thursday from 16:00 to 18:00 on ZOOM or in person (please book your slot by email by using "VAMR Office Hourse" as the email subject).




Video Recordings and Student Forum


All video recordings of the lectures and exercises will be made available within the course online platform OLAT. Please use the Forum therein to ask questions to the instructors.




Course Program, Slides, and Additional Reading Material


Please notice that this is a tentative schedule and that the effective content of the lecture may change from week to week.


Date

Lecture and Exercise description

Slides, exercises, and additional reading material

21.09.2023 Lecture 01 - Introduction to Computer Vision and Visual Odometry
Exercise: Camera Notation Tutorial.
Lecture Slides
Exercise Lecture Notes
Additional reading on Visual Odometry
Linear Algebra Primer from Stanford University
Linear Algebra interactive tool
Camera notation tutorial
28.09.2023 Lecture 02 - Image Formation: perspective projection and camera models
Exercise 01- Augmented reality wireframe cube
Lecture Slides
Exercise 01 vectorization code
Matlab primer
Matlab primer code
Python Tutorial
Exercise 01
Solutions 01 (Matlab/Python/Numerical)
05.10.2023 Lecture 03 - Camera Calibration
Exercise 02 - PnP problem
Lecture Slides (updated 12.10.2023, see changes here)
Additional reading on P3P and PnP problems
Exercise 02
Notes on SVD for DLT
Solutions 02 (Matlab/Python/Numerical)
12.10.2023 Lecture 03 continued
Lecture 04 - Filtering & Edge detection
Exercise session replaced by continuation of Lecture 4.
Lecture Slides
Numerical Exercise Lecture 04
Numerical Exercise Lecture 04 Solution
19.10.2023 Lecture 05 - Point Feature Detectors, Part 1
Exercise 03 - Harris detector + descriptor + matching
Lecture Slides
Exercise 03
Solutions 03 (Matlab/Python/Numerical)
26.10.2023 Lecture 06 - Point Feature Detectors, Part 2
Exercise 04 - SIFT detector + descriptor + matching
Lecture Slides
Additional reading on feature detection
Exercise 04
Notes on Convolution 04
Solutions 04 (Matlab/Python/Numerical)
02.11.2023 Lecture 07 - Multiple-view Geometry 1
Exercise 05 - Stereo vision: rectification, epipolar matching, disparity, triangulation
Lecture Slides
Additional reading on stereo image rectification
Exercise 05
Solutions 05
09.11.2023 Lecture 08 - Multiple-view Geometry 2
Exercise 06 - Eight-Point Algorithm
Lecture Slides
Additional reading on 2-view geometry
Exercise 06
Solutions 06
16.11.2023 Lecture 09 - Multiple-view Geometry 3
Exercise 07 - P3P algorithm and RANSAC
Lecture Slides
Additional reading on RANSAC
Exercise 07
Solutions 07
23.11.2023 Lecture 10 - Multiple-view Geometry 4
Continuation of Lecture 10 + Exercise session on Intermediate VO Integration
Lecture Slides
Additional reading on open-source VO algorithms
Numerical Exercise 10
Numerical Exercise 10 Solution
30.11.2023 Lecture 11
1st hour: seminar by Dr. Jeff Delaune from NASA-JPL: "Vision-Based Navigation for Mars Helicopters."
2nd hour: Optical Flow and KLT Tracking
Exercise 08 - Lucas-Kanade tracker
Lecture Slides
Additional reading on KLT
Exercise 08
07.12.2023 Lecture 12a (1st hour) - Place Recognition
Lecture 12b (2nd hour) - Dense 3D Reconstruction
Lecture 12c (3rd and 4th hour, replaces exercise) - Deep Learning Tutorial
Optional Exercise on Place Recogntion
14.12.2023 Lecture 13 - Visual inertial fusion
Exercise 09 - Bundle Adjustment
21.12.2023 Lecture 14 - Event-based vision
Exercise session: Final VO Integration
11.01.2024 Written exam: from 08:00 to 10:00 am. More details will be given during the semester.



Exam


The written exam will take place on January 11th, 2024 from 08:00 to 10:00 am. More details will be given during the semester.




Exercises


You will be required to bring your own laptop to the exercise session. This year, for the second time, students will have the chance to solve the exercises in Python or Matlab. However, the exercise statements and material presented during the exercise sessions will only cover the implementations in Matlab. Additionally to the Matlab solutions, we will upload Python scripts as template solutions for the exercise statements. You will need to have Matlab or Python already pre-installed on your machine for the exercises. To install Matlab, follow the links below according to your affiliation.

  • ETH students: Download from here
  • UZH students: Download from here; Info on how to setup the license can be found there.
Please install all the toolboxes included in the license.




Optional Mini Project and Grading


Students taking the final exam have the opportunity to submit a mini-project to increase their final grade. Notice that the mini project is optional, that is, not mandatory. Depending on the result of the mini-project, the student will be rewarded with a grade increase of up to 0.5 on the top of the grade of the written exam. However, bear in mind that the mini-project can be quite time-consuming. Students who want to submit a mini project should work in a group of minimum 2 and max 4 persons. The deadline for the mini-project is Sunday, 07.01.2024, 23:59:59, and it can be submitted via e-mail to the assistants. Mini project specifications and files can be found in the table below.


DescriptionLink (size)
Project Specification vo_project_statement.pdf
FAQ Frequently Asked Questions
Parking garage dataset (easy) parking.zip (208.3 MB)
KITTI 05 dataset (hard) kitti05.zip (1.4 GB)
Malaga 07 dataset (hard) malaga-urban-dataset-extract-07.zip (2.4 GB)
Matlab script to load datasets main.m


Recommended Textbooks

(All available in the Swisscovery catalogue)

Unfortunately, a single textbook that covers all the material seen in class does not exist yet. However, the topics covered in class can be found in the books listed below. At the end of each lecture's slide deck, you will find a list of chapters and books where the topics can be found.

  • Computer Vision: Algorithms and Applications, by Richard Szeliski, Springer, 2nd Edition, 2021. The PDF of the book can be freely downloaded from the author's webpage. Although the author is working on a 2nd edition, this is still under progress. Thus, in the lecture nodes we will reference chapters of the 1st edition.
  • Chapter 4 of "Autonomous Mobile Robots", by R. Siegwart, I.R. Nourbakhsh, D. Scaramuzza. PDF
  • Robotics, Vision and Control: Fundamental Algorithms, 2nd Ed., by Peter Corke 2017. The PDF of the book can be freely downloaded (only with ETH VPN) from the author's webpage.
  • An Invitation to 3D Vision, by Y. Ma, S. Soatto, J. Kosecka, S.S. Sastry.
  • Multiple view Geometry, by R. Hartley and A. Zisserman.




Spring 2007-2015 - Autonomous Mobile Robots


From 2007 to 2015 I was co-lecturer of the course Autonomous Mobile Robots together with Prof. Roland Siegwart. The old Power Point slides and videos can be found here. The lectures of the course inspired the second edition of the omonym book Autonomous Mobile Robots below.

The course continues to be taught by Prof. Siegwart at the following link.

The course is also available as an MOOC (Massive Open Online Course) on the popular online platform edX.



Reference book

R. Siegwart, I.R. Nourbakhsh, and D. Scaramuzza

Introduction to autonomous mobile robots 2nd Edition (hardback)

A Bradford Book, The MIT Press, ISBN: 978-0-262-01535-6, February, 2011

The book can be bought during the first lecture or on Amazon.

MIT Website Buy