Enhanced Thermal Human Detection with Fast Filtering for UAV Images

Real-time human detection systems can facilitate search and rescue (SAR) operations by expediting the identification of individuals in disaster-stricken areas. Unmanned aerial vehicles (UAVs), equipped with thermal imaging, can help locate missing or injured people, especially at night or in adverse weather conditions. In this study, a preprocessing filter for increasing the detection accuracy of YOLOv8 in thermal images, has been developed. Combining Contrast Limited Adaptive Histogram Equalization (CLAHE) with color value remapping filters, the proposed method creates a 3-channel RGB image from a grayscale thermal image. Experimental results show that the proposed filtering method is successful in decreasing both the false positive and the false negative rates. In addition, having minimal computational and memory requirements, it provides an alternative technique for future studies involving real-time detection in edge computing devices.