CMPS 260 Home Page
Advanced Computer Graphics
Spring 2005

Instructor: Suresh Lodha
Office: E2-361
Phone (831) 459-3773
email: lodha@soe.ucsc.edu
Office Hours: Tu 2:00-4:00 and by appointment
Class Time: TuTh 10:00-11:45
Place: Porter Acad 250
Class Number: 54956

Topics:

The overall goal of this class is to learn how to construct image and/or model-based 3D geospatial environments using lidar (depth range data) and/or images. Particular focus will be on using aerial lidar data for terrain modeling, data classification, segmentation, urban modeling, simplification, and visualization. Machine learning technqiues including Expectation-Maximization (EM), AdaBoost, Support Vector Machines (SVM), and clustering will be discussed to solve these and related problems.

Prerequisites:

  • Ability to conduct literature search and read papers from
    graphics/visualization/vision/lidar conferences such as SIGGRAPH, IEEE Visualization, CVPR, and ISPRS
  • Ability to write programs in either MATLAB or OpenGL/FLTK with C or C++
  • Ability and initiative to define a feasible project on graphics/visualization/vision/lidar quickly and finish it within the quarter timeline
  • Prior experience and knowledge of graphics (CMPS 160), visualization (CMPS 161), Vision (CE 264), Image Processing (EE 264),
    or Machine Learning (CMPS 242) will be an added plus but not required.

    Schedule:

  • Topic 1: Overview, Expectations and Projects

    • March 29, Tuesday, March 31, Thursday, April 5, Tuesday

    3D Image-Based Virtual Environment using Images
    QuickTime Virtual Reality: An Image-Based Approach to Virtual Environment Navigation", SIGGRAPH, 1995 by S. E. Chen
    Contextualized Lab Tour of Delft University Labs

    3D Urban Modeling using Lidar
    Approaches to Large Scale Urban Modeling, IEEE CG&A, Vol. 23, No.6, Dec 2003 by Hu, You, and Neumann

    3D Urban Modeling using Images
    Sections 7.1 and 7.2 on 3D from Stereo Images, Trucco-Verri Book

    Machine Learning
    Machine Learning for Computer Graphics: A Manifesto and Tutorial by Aaron Hertzmann, 2003

  • Topic 2: Terrain Modeling using Aerial Lidar Data

    • April 7, Thursday, and April 14, Thursday
    Overview of Aerial Lidar Data

    Comparison of Filtering Algorithms
    Sithole and Vosselman, ISPRS Commission III, WG 3, 2003.

    Terrain Modeling Algorithms

    Topic 3: Empowering with MatLab

    April 12, Tuesday
    Matlab Primer

  • Topic 4: Machine Learning Techniques: AdaBoost

    • April 19 and April 21
    A Short Introduction to Boosting
    Aerial Lidar Data Classification using AdaBoost
    Improved Boosting Algorithms Using Confidence-Rated Predictions
    Face Recognition Using Boosted Local Features

  • Topic 5: Machine Learning Techniques: Clustering and EM

    • April 21 and April 26
    K-Means Clustering
    K-Means Clustering with Refined Seeding, 1998
    Expectation-Maximization (EM)

  • Topic 6: Aerial Lidar Data: Commercial Software

    • April 30
    TerraSolid
    Presentation by Arttu Soininen

  • Topic 7: 3D Urban Modeling Using Aerial Lidar Data

    • April 28, May 3, May 5
    References

  • Topic 8: Panoramic Mosaicing Using Images

    • May 10
    Pinhole Camera Model, Hartley and Zisserman, Section 5.1, pages 139-144
    Importance of Camera Center including Panormaic Mosaicing, ibid, Section 7.4, pages 192-196
    Homography Computation, ibid, Section 3.1, pages 71-73
    Automatic Homography Computation, ibid, Section 3.8, pages 107-110

  • Topic 9: 3D Reconstruction Using Images

    • May 12
    Epipolar Geometry, Fundamental and Essential Matrix, Camera Calibration
    Parts of Chapters 6 and 7 of Trucco and Verri
    Parts of Chapters 8, 9, and 10 from Hartley and Zisserman
    Photometric Stereo, Section 5.4 of Forsyth and Ponce

  • Topic 10: 3D Urban Modeling Using Images

    • May 17
    Automatic Description of Complex Buildings from Multiple Images
    by Zhao and Nevatia, CVIU, VOl. 96., No.1, Oct 2004

    Recognizing Panoramas Brown and Lowe, ICCV, 2003
    Baillard, Maitre, Werner, Zisserman, Luc Van Gool, 1998-2004

  • Topic 11: 3D Urban Modeling Applications
    Read on Your Own Papers
    Projects from South Carolina, Berkeley, CMU, MIT, Stanford, University of Tokyo, Georgia Tech, and USC


  • Topic 12: Student Presentations
    May 26 and May 31
    20 minute students' presentations

  • Topic 13: Forest Structure Determination using Aerial Lidar Data
    June 2
    Forest Papers

  • Topic 14: Machine Learning Techniques: SVM
    June 2
    Support Vector Machines (SVM)
    Associative Markov Networks


  • Topic 15: Additional Topic
    (possibly Automatic Vehicle Navigation in MARS, could change)
    could not be covered



  • Final Project Demonstration and Report
    June 7, Tuesday, 4-7pm


    • Final Project Suggestions

    Books

    References

    Expectations and Evaluation:

  • Homework (25%): 4 or 5 homeworks will be given
  • Class Presence and Participation (including initiative, interest, efffort, timeliness, etc.) (15%)
  • Class Presentation (15%)
  • Final Project Software, Report, and Demo (45% = 30%+10%+5%)

    I will encourage research-type or highly interesting programming projects leading to conference papers or compelling web presence. I may also allow literature seearch and survey type final project with prior consent. I am motivated to learn from you as much as you are motivated to learn from this class.

    Homework:

  • Homework 1:
    		• Virtual City Exploration: Target Date: April 7
  • Homework 2:
    		• Playing with Aerial Lidar Data Target Date: April 26
  • Homework 3:
    		• Project Proposal Target Date: May 5, Thursday
  • Class Presentation:
    		• Dates: May 26, Thursday and May 31, Tuesday
  • Homework 4:
    		• Machine Learning Target Date:
  • Homework 5:
    		• Modeling using Images Target Date:
  • Class Project:
    		• Target Date: June 7
    Maintained by Suresh Lodha