Flood areas detection based on UAV surveillance system

Abstract

In this paper we propose a methodology for detection, localization, segmentation and size evaluation of flood areas from aerial images which are taken with drones. The approach is based on sliding box method and texture features analyses. The process of feature selection takes into account a performance degree obtained from false positive and false negative cases. We combined different properties of the images like color, texture and fractal types. A class of flood and one of non-flood were established based on clustering properties of some features and a criterion of similarity is used to segment the flood zones. Finally, the evaluation of the flood size is proposed. The method was tested on 10 images of flood zones and a rate of accuracy of 98.87% was obtained.