Instructor: Prof. Norbert Haala, Institute for Photogrammetry, University of Stuttgart; e-mail: norbert.haala@ifp.uni-stuttgart.de

Target audience: PhD and Master students involved in geomatics and photogrammetry, staff from national mapping agencies, public authorities and third parties in charge of 3D databases like DTM and urban model.

 Course objective: The course will give the theoretical and practical background on the recent developments in photogrammetric image processing, which considerably improved the automatic image-based collection of high quality digital surface models and 3D point clouds. Course participants will become familiar with state-of-the-art techniques for pixel-wise stereo image matching like the Semi-Global Matching (SGM) approach. They will also learn how dense surface reconstruction profits from the redundancy of highly overlapping image flights using multi-stereo processing. The course will demonstrate how techniques and concepts from computer vision can be used to increase the efficiency of photogrammetric image processing.

Course outline: The introductory course during the kick-off seminar provides the participants with the theoretical framework of high density image matching and will give an overview on the relevant photogrammetric processing chain. The individual work and exercises as well as hands-on demonstrations during the following online study phase use real-world data e.g. from the EuroSDR benchmark on image matching for DSM computation. Potential applications are demonstrated, while additional sample calculations will exemplify the relevant processing steps and enable the required feedback.

Module 1. Structure-from-motion and image pre-processing

The module briefly reviews the standard photogrammetric processing chain including automatic image orientation, rectification of stereo pairs and point determination from estimated parallaxes. Main emphasis is put on the efficient computation of such photogrammetric tasks based on projective geometry as it is commonly used in computer vision within open source libraries and available software tools – 8 hours

Module 2. Pixel-based stereo matching for dense surface reconstruction

The module discusses state-of-the art algorithms and optimization criteria for dense stereo image matching. Main emphasis will be on the use of Semi-Global-Matching for the processing of aerial imagery. In addition to the core parallax estimation process, a general outline of multi-stereo image matching to efficiently evaluate highly overlapping image flights is given. For the hands-on demonstrations the multi-view stereo software solution SURE, developed at the University of Stuttgart will be made available to the participants. This software enables the derivation of dense point clouds from a given set of images and its orientations – 16 hours

Module 3. Quality analysis

Exemplary results from pixel-based image matching are used to introduce tools and approaches for visualization and analysis of high density 3D point clouds. By evaluating the quality of such data the participants will also be able to assess its potential for follow-up applications and further processing – 8 hours

Instructor: Dr. Clément Mallet, IGN France, e-mail: clement.mallet@ign.fr

Target audience: PhD and Master students involved in remote sensing, photogrammetry and digital cartography, staff from national mapping agencies, public authorities and third parties in charge of mapping using satellite imagery.

Course objective: The course will present state-of-the-art techniques for change detection (CD) in land-cover geospatial databases. The first part of the lecture will be dedicated to classification (feature extraction and labeling techniques) and the second part will focus on how to apply such techniques for CD issues. Both technical aspects and practical issues will be described during this course. The relevance of various remote sensing data will be discussed before providing solutions for 2D database change detection and update. In addition, the participants will learn how to benefit from various data sources and how to take advantage of the multiple spatial, spectral, and temporal resolutions of the input data.

Course outline:

Module 1. General assessments
The aim is to present the various remote sensing data sources as well as the various spatial and temporal scales of analysis for change detection. Questions like why to detect changes or which kinds of change can be detected are going to be answered. Land-Use/Land-Cover (LCLU) monitoring is presented as well as the various kinds of existing LCLU geodatabases.

• Remote Sensing data sources
• Spatial and Temporal scales of analysis
• Main applications
• Outlook on main change detection methods

Module 2. Image-to-Image 2D change detection

• First, focus will be made on the simple radiometric comparison of images (the so-called pre-classification). Additional lecture and papers are also provided.
• Secondly, a lecture on feature extraction from images is given since CD can be performed on other features that the raw channels of the images.
• Thirdly, classification methods can be carried out on the various images and can be compared (post-classification).
• Exercices will introduce you to both unsupervised and supervised methods on a single image, and two images and then to the post-processing techniques that should be carried out.

Module 3. Operational land-cover change detection

Techniques introduced in modules 1 and 2 will be presented in the operational frameworks of Spanish and German initiatives. In addition, practical solutions for improving CD results will be tested during exercices.

• Which features are the best ones for a given problem ?
• How to design efficient training sets for supervised classification ?

Module 4. 3D change detection
The relevance and the specificity of 3D change detection (i.e. Digital Surface Model vs. Digital Surface Model) will be introduced. Two main CD techniques will be detailed and followed by exercises enhancing which processing techniques should be adopted.

Module 5. Time-series Image Analysis

• Benefit of image sequences for LC/LU mapping and change detection.
• Outlook on existing methods for multitemporal change detection.

Exercices on how to benefit from multiple images to improve CD results.