Approaching near real-time mapping solutions for geointelligence

The ability to derive intelligence from airborne or satellite-based sensors in an automated or semi-automated way offers numerous advantages for government and relief agencies responding to global disasters and conflicts. It is also a key element in an early response system designed to guide decision making during the critical phases of a disaster.

eCognition’s advanced approach to feature extraction, known as object-based image analysis, supports a broad range of intelligence and rapid-response applications. It is highly scalable, thrives in multi-sensor production environments and, in the hands of a trained expert, provides a tangible tactical advantage for disaster response teams.

Case in Point: Haiti’s 7.0 Earthquake

Haiti’s magnitude 7.0 earthquake on January 12th resulted in a “worst-case” humanitarian scenario. On January 14th, the G-MOSAIC Rapid Geospatial Reporting service was activated at the request of the United Nations Division of Field Support (UN/DFS) and the Spanish Red Cross. Using eCognition software, G-MOSAIC partners delivered damage indicator maps within 12 hours of receipt of the ortho-rectified satellite images.

In the G-MOSAIC case, the automated approach using eCognition was able to achieve up to 80 percent agreement with manual analysis conducted on the same data sets despite serious limitations in the data including:
  • down-sampling of the spatial resolution in the panchromatic band,
  • geometric inaccuracies in image pre-processing, such as geometric shift and radiometric differences, 
  • variations in the pre- and post-earthquake images such as season and angle-of-image capture, and
  • the use of images only, without aerial laser scanner data or an elevation model.

Workflow for automated damage indication
The approach was based on changes in shadows cast by buildings before and after the earthquake. To avoid false positives from vegetation shadows, a vegetation mask was created for both the pre and post-earthquake images. Because metadata was not available on the processing levels, a fast image-to-image or image-to-map registration was hampered. To resolve the registration problem while avoiding a time-consuming additional (co-) registration of the images, an object-linking approach was implemented. After the damage indications were extracted during the change analysis, spatial queries were used to link similarly shaped objects found in the pre- and post-earthquake images. If a shadow object identified in the pre-event image overlapped a shadow object identified in the post-event image, the size and form of the two objects was automatically compared and, in case of similarity, used as an indication of a still existing structure (or vice versa). This effectively created an object-by-object comparison between images despite the geometric shift.

Scaling Human Interpretation
The rapid mapping delivered in Haiti provided essential geointelligence to help guide decision making. In cases of this sort, time is of critical importance and analysts inevitably face information overload. Clearing this bottleneck is essential to achieving a tactical advantage in the early hours of an emergency response. eCognition provides that advantage by empowering a human analyst to reduce the time required to screen imagery, queue information for examination and, in some cases, automate analysis processes.

eCognition allows expert knowledge to be scaled using computing power. This is accomplished using a development environment for the codification of analysis logic within a rule-set-based application. These applications can be executed across vast amounts of data using a robust service-oriented architecture (SOA), GRID infrastructure and 64-bit processing. These capabilities, together with sophisticated data fusion, allow eCognition to effectively manage feature extraction within multi-sensor environments for production rapid response, geointelligence or civil applications.
Analysis completed by ZGIS: Dirk Tiede, Petra Füreder, Daniel Hölbling & Stefan Lang and Trimble: Christian Hoffmann

» Back to our Aerial Mapping 2011 Issue