3D Computer Vision

 

Syllabus
MS Computer Science
3D Computer Vision
Spring, 2017
(3 credits)

Description

The main objective of this class is to introduce to the students the mathematical concepts and algorithms to reconstruct a scene from a collection of images. It will enable them to implement standard algorithms, evaluate the strengths and weakness of various approaches, and explore a topic of their own choosing in a course project. A further objective is to introduce the current research applications and that the student acquires a deeper understanding of the subject. This class emphasizes the development of 3D-vision systems via programming assignments and a final project.

Prerequisites

Proficiency in programming, and familiarity with matrix arithmetic and geometry. Most of the knowledge required should be part of the standard background in Computer Science, fundamentals of programming and undergraduate/graduate Mathematics and Geometry. Previous experience with computer vision, image processing and/or computer graphics will be helpful but it is not required. It is important to note that this is not an image processing course but a course on 3D vision which uses image processing as basic methodology but 3D geometry and mathematical methods to recover 3D information from single or multiple 2D images, thus being complementary to Image Processing courses.

Instructor

Dr. Furqan Ullah, furqanullah@ucp.edu.pk,

Office: R303, +92-42-35880007

Office hours: Tuesday, Wednesday 2:00–3:00 pm

Class website: http://cgav.ucp.edu.pk/teaching/3d-computer-vision/

Teaching Assistant

TBA, TBA@ucp.edu.pk,
Office: TBA
Office hours: TBA

Meetings

Monday, Wednesday, Thursday, 02:00–02:50 am, R303

Format

This course will be a mixture of lectures, discussions, and demonstrations. The student is expected to actively participate in all class activities.

Text and References

  • Richard Szeliski, Computer Vision: Algorithms and Applications, Springer, 2010 (Available free of charge online).
  • Computer Vision – A modern approach, by D. Forsyth and J. Ponce, Prentice Hall.

Reference Material

  • Additional material (scientific papers, chapters from other textbooks etc.) will be distributed during the course and made available via the course web-site.

Online Resources

Course webpage will be used for organizing additional reading and programming material. The web page will contain class notes, important announcements and assignments. Please check the class web page and your e-mail regularly.

Specific outcomes of instruction

Upon successful completion of the course, the students will be able to

  • Understand the fundamental problems of 3D computer vision
  • Understand the main concepts and techniques used to solve those
  • Evaluate the cutting edge algorithms in 3D computer vision and investigate and develop solutions for research problems in the area of 3D computer vision
  • To understand basic CV methodologies as used in industrial and entertainment systems.
  • Apply the geometry and reconstruction pipeline using calibrated/uncalibrated cameras and depth sensors (e.g., Kinect)
  • To enable participants to implement own solutions to applications of their choice.
  • Write code to implement standard algorithms (such as region analysis, edge detection, template matching, segmentation, stereo correspondence, perspective projection, epipolar geometry calculation, compression, 3D reconstruction)
  • Generate 3D model from images
  • Design and apply efficient and scalable computational 3D vision systems

Course Outline

  • Introduction
  • Image Formation and Image Models
  • Projective Geometry, Cameras
  • Camera distortions and artifacts
  • Camera calibration
  • The geometry of multiple views
  • Stereo Vision, epipolar constraints, disparity
  • Shape from stereo, correspondence
  • Image formation, homogeneous coordinates
  • Two-view geometry, homography estimation
  • Binocular stereo, matching criteria
  • Feature extraction, KLT tracking, Structure
  • Structure-from-Motion part II, the PnP problem and loop closing
  • Simultaneous Localization and Mapping, Kalman filtering
  • Multi-view stereo
  • Multi-view stereo part II and image-based rendering
  • Point cloud processing, normal estimation, Kinect sensors and data
  • Invariant descriptors for 3D data, k nearest neighbor classifier
  • Convex hulls, line intersection, mesh subdivision and simplification
  • Triangulation and partitioning, range search (k-d trees), mesh representation
  • Delaunay triangulations and Voronoi diagrams
  • Point cloud registration and compression: iterative closest point (ICP) & minimal scene representations.
  • The visual hull, silhouette-based modeling, occupancy grids, octrees

Projects

There will be a semester project for a group with no more than three members. Students are expected to investigate, develop, and implement a solution to a research problem in teams. They will present the motivation, design, and evaluation of their solution. As part of the homework assignments, students will present in teams of no more than three members a recent research paper.

Exams

Midterm, TBA
Final, TBA

Grading

40%      Assignment, Quiz, Project
25%      Midterm
35%      Final

Course Policies

  • All work must be your own, group work is CHEATING, and all group members will receive a zero.
  • Turn off cell phones and pagers before class.