Estimating the number of refugees and internally displaced persons is important for planning and managing an efficient relief operation following disasters and conflicts. Accurate estimates of refugee numbers can be inferred from the number of tents. Extracting tents from high-resolution satellite imagery has recently been suggested. However, it is still a significant challenge to extract tents automatically and reliably from remote sensing imagery. This paper describes a novel automated method, which is based on mathematical morphology, to generate a camp map to estimate the refugee numbers by counting tents on the camp map. The method is especially useful in detecting objects with a clear shape, size, and significant spectral contrast with their surroundings. Results for two study sites with different satellite sensors and different spatial resolutions demonstrate that the method achieves good performance in detecting tents. The overall accuracy can be up to 81% in this study. Further improvements should be possible if over-identified isolated single pixel objects can be filtered. The performance of the method is impacted by spectral characteristics of satellite sensors and image scenes, such as the extent of area of interest and the spatial arrangement of tents. It is expected that the image scene would have a much higher influence on the performance of the method than the sensor characteristics. (C) 2014 Elsevier B.V. All rights reserved.
|Number of pages||7|
|Journal||International Journal of Applied Earth Observation and Geoinformation|
|Early online date||5 Dec 2014|
|Publication status||Published - Apr 2015|
This research was partly supported by the European Commission under FP7 (Seventh Framework Programme): “SENSUM: Framework to Intergrade Space-based and in-situ sENSing for dynamic vUlnerability and recovery Monitoring” (312972). We gratefully acknowledge the helpful comments from Michael Ramage, Dilkushi de Alwis Pitts, and the anonymous referees.
- remote sensing
- relief operation
- remotely-sensed data