GeoEngine - Geomatics Engineering (Master of Science)

Computer Vision

3. Semester GeoEngine

Computer Vision (3. Semester)

Studiengang: GeoEngine (Master of Science)
Semester: 3. Semester  
Module: 77790 Computer Vision and Pattern Recognition
Instructure: Prof. Dr.-Ing. Norbert Haala
Dr.-Ing. Michael Cramer
Lecture: Thursday / weekly / 15:45-17:15
Raum: M24.12
Exercise: Wednesday / biweekly / 15:45-17:15
Raum: M24.01
Course Contents
Computer Vision for Automatic Photogrammetric 3D Data Collection, SIFT Feature Extraction and Matching Projective Geometry for efficient solution of photogrammetric tasks, Efficient modelling of camera parameters: Perspective Projection Matrix, Structure-from-Motion, Dense Multi-View-Stereo Image Matching for 3D Surface Reconstruction
Pre-requisites
Elementary knowledge of mathematics, linear algebra, signal processing, statistical inference, photogrammetry and aerial data acquisition.
Recommended textbooks
  • Hartley, R., Zisserman, A. (2013): Multiple View Geometry in Computer Vision, Cambridge University Press, 655p.
  • Szeliski, R. (2010) Computer Vision: Algorithms and Applications, Springer Verlag
  • Gonzales, R., Woods, R. (2007) Digital Image Processing
Exams
 

 

Lecture Notes
  1. Introduction to Image based 3D Data Capture
    1. Structure-from-Motion: The big picture
  2. Automatic Tiepoint Generation
    1. Introduction to Image Matching
    2. Recap: Correlation and Feature Matching
    3. SIFT Feature Extraction and Matching
    4. RANSAC for robust parameter estimation: Elimination of outliers by geometric validation
    5. Example: Automatic image alignment by Projective Transformation Estimation
  3. Projective Geometry for efficient solution of photogrammetric tasks
    1. Efficient modelling of camera parameters: Perspective Projection Matrix
    2. Providing approximate values for Bundle Block Adjustment
    3. Application and Assignment: Spatial Resection by DLT
  4. Relative orientation for geometric validation
    1. Structure-from-Motion and Connectivity Matrix
    2. Essential and Fundamental matrix representation
  5. Dense Stereo Image Matching
    1. Generation of epipolar image pairs
    2. Pixel correspondences by Semi-Global-Matching
    3. Multi-Image-Matching and point triangulation
  6. Surface Reconstruction: Dense Multi-View-Stereo Image Matching
    1. Multi-Stereo-Triangulation – add-on for potential optimization
    2. Representation of 3D shape: From range images to 3D point clouds, meshes and voxel-space
    3. Volumetric labeling for multi-view-stereo 3D surface extraction
    4. Integration of stereo-image-matching and LiDAR measurements
To the top of the page