Damage assessment on buildings using multisensor multimodal very high resolution images and ancillary data

CD-ROM; PosterInternational audienceMajor disasters across the world are more and more reported because their consequences in terms of human and economical losses are increasing. This increase is explained by the growing population and by its migration in areas that are prone to disasters like seacoasts. Remote sensing has proved is usefulness for the crisis mitigation through situation report and damage assessment, as acknowledged by creation of initiatives like the International Charter Space and Major Disaster or UNOSAT. In this operational scope, the required information is manually extracted from images acquired by satellites. Usually a reference image acquired before the disaster and a crisis image acquired after the disaster are compared to retrieve damage. Concerning damage assessment on buildings, Very High Resolution (VHR) images are usually used because it allows a more reliable visual analysis at this scale. The production of information has to be as short as possible, hence the need of automation to speed it up. The images are to be made comparable, that is to say registered, and this requirement is by fare acute when considering an automatic image analysis method. On one hand, the crisis image has to be acquired as soon as possible following the disaster, regardless to the sensor type and the acquisition parameters; on the other hand, the reference image has to be as recent as possible, to avoid additional major changes that aren't related to damage. Hence there is little chance for this reference image to be acquired in the same conditions (acquisition angles for example), or even with the same sensor, than the crisis image. Moreover, the multitemporal analysis of VHR images exhibits more natural changes that aren't related to damage. This is for examples changes due to human activities, or shadow changes due to different illumination conditions. These natural changes have to be corrected or filtered out. All these differences represent a challenge for automatic images analysis methods that is seldom addressed in the literature. Object-oriented methods allow to focus the analysis on the objects of interest, thus to partly avoid false alarms due to natural changes. We focus on the buildings, more precisely on their roofs because they are most often visible by means of remote sensing. For this purpose, we use ancillary data that consist of the buildings roofs outlines in the reference image only. This data can be obtained by segmentation of the reference image, or from a Geographical Information System. Using this approach, the main remaining difficulty for automatic analysis is the registration of images acquired with different parameters, or even with different sensors and different resolutions. Using the ancillary data in agreement with the reference image, we propose a method that automatically searches for the buildings roofs outlines in the crisis image. Then it attributes change coefficients to each building by assessing the amount of change on their roof. From these features, the buildings are individually classified to quantify the damage on them. A supervised classification based on SVM is chosen. It allows to reach good classification performance with a small training set. The chosen area of study is Beirut, in Lebanon. It is particularly adapted to our study because several images are available, before, during and after the bombing in summer 2006. We use images acquired with very different conditions, and with two VHR sensors, Ikonos and QuickBird. The studied urban area, Haret Hreik, includes high buildings that generate large shifts of the roofs from one image to the other, and also some occlusion areas. The images set is composed of two reference images and three crisis images concerning the QuickBird sensor; two reference images and two crisis images concerning the Ikonos sensor. Considering monosensor and multisensor image pairs, it represents 20 different reference/crisis pairs. As ancillary data, the outlines of the buildings roofs have been manually extracted from one QuickBird reference image and registered in the other reference images. Our methodology is applied to the 20 different pairs of images. The performance of the method is evaluated for each pair as a function of the associated base to height ratio (B/H). This B/H lies between 0.42 and 1.4, which is equivalent to a difference angle between 24o and 80o. The performance of the proposed damage assessment method decreases along with the B/H. However, we show that it remains robust against extreme difference in acquisition angles. It distinguishes undamaged from damaged buildings with an accuracy of 84% using a pair of images with a difference angle of 80o. It reaches a performance of damage detection equal to 93% when the angle difference is 24o. The second conclusion of this study is that the proposed method is as efficient with QuickBird or Ikonos data, regardless to the difference of spatial resolution and sensor characteristics. Finally, we show the robustness of our method considering multisensor pairs of images. These are promising results for a future operational application