Real-time Reconstruction of Flooded Areas from Low Resolution UAV Imagery

This work focuses on the estimation of flooded areas from aerial images available from optical camera powered UAVs or others lighter than air platform (LTAP) sources which are deployed for large elevated surveillance during the natural calamities related to flood. It is observed that images in such cases are mostly low in resolution. Due to limited bandwidth during crisis, UAVs will have low bandwidth to send such information to the base station. Hence, relatively low complexity modules are required in real time for such scenario which can provide possible flooded area segmentation for any given time frame. This paper incorporates the technique for real-time segmentation flooded areas into UAVs, captured during UAV navigation in crisis regions. Here, the single UAV navigation has been considered for partitioning the surveillance regions depending on the minimum hop and allowable field of view (FoV). Then we have implemented Color quantization based K-Means clustering to segment such low resolution images accurately to reconstruct the flooded locations over a vast area. Moreover, this approach would be useful in urban as well as inaccessible terrains for real time flood area modelling using low resolution images.