A fast and robust approach for human detection in thermal imagery for surveillance using UAVs

The use of Unmanned Aerial Vehicles (UAV)s has spread in various fields such as surveillance and search and rescue purposes. This leads to many research efforts that are focusing on the detection of people using aerial images. However, these platforms have limited resources of power and bandwidth which cause many restrictions and challenges. The use of the thermal sensors offers the possibility to work day and night and the detection of the human bodies because of its distinguished thermal signature. In this paper, we propose a fast and efficient method for the detection of humans in outdoor scenes using thermal images taken from aerial platforms. We start by extracting the bright blobs based on a conjunction between a saliency map and a contrast enhancement techniques. Then, we use the Discrete Chebyshev Moments as a shape descriptor and finally, we classify the blobs into humans and non-humans. The proposed framework is first tested using a well-known thermal database that covers a wide range of lighting and weather conditions and further and then compared to an also well-known blob extractor which is the Maximally Stable Extremal Regions detector (MSER). The results highlight the effectiveness and even the superiority of the proposed method in terms of true positives, false alarms and processing time.