| Defence date |
2008-09-15 |
|
Mathématiques et leurs applications
Sciences et technologies de l'information et de la communication |
| Library |
Mines ParisTech |
| Keywords |
Détection de dégâts – Détection de changements – Image satellite – Télédétection – Très haute résolution spatiale – Bâtiments – Milieu urbain – évaluation de la qualité – Recalage géométrique – Classification – Traitement de données |
| English title |
Damage assessment on buildings due to major disasters using very high resolution satellite multimodal images |
| English abstract |
In a disaster aftermath, it is required to know rapidly the severity of the damage on buildings. Presently, this damage assessment is manually conducted through a visual comparison of satellite images. Automatic methods are immature; their performances being seldom quantified, they are not used by operational. We propose a standard protocol to quantify the performances of the damage assessment methods. It is based on reference databases obtained from five various disaster cases. The protocol allows to quantify the performances of a method and to compare its results to other ones. Having this assessment protocol, we propose a damage assessment method from a pair of panchromatic very high spatial resolution satellite images and a set of objects of interest defined in the reference image. The developed method has to lead to satisfying and reproducible results using images acquired with different modalities, and to be automated as much as possible. The damage on buildings are quantified from the amplitude of the changes on their roof. To compare the latter, they have to be registered. The geometric registration of very high resolution (VHR) images is an unsolved difficult problem; a new method that is adapted to our problem is developed and assessed. It generally leads to satisfying results for our application. Then change features are extracted. Two correlation coefficients and some textural features obtained by filtering are extracted, and a damage degree is attributed to each building through a supervised classification method. The impact of the differences in image modality on the performances of our method is assessed. The proposed method is fast, can be applied mostly generally, and is robust along with the use of VHR images with different spatial resolution or acquired with different sensors; the influential parameter is the B/H of the images pair. |
| English keyword |
Damage detection – Change detection – Satellite image – Teledetection – Very high spatial resolution – Buildings – Urban area – Quality assessment – Geometric registration – Classification – Data processing |
