Vision-based UAV Detection in Complex Backgrounds and Rainy Conditions

Studies aimed at detecting UAVs in real time using processer vision and deep learning are in their infancy. Although there are many possible advantages to using unmanned aerial vehicles (UAVs), some people are concerned about the risks and misuse of these machines. All potential threats to safety, privacy, and security fall under this category. It is common practice to combine hardware components, such as cameras, with software components when developing vision-based detection systems. Under difficult circumstances such as changing experiences, varying UAV sizes, complicated related settings, and light to heavy rain, this research evaluates the performance of popular and cutting-edge vision-based object-identification algorithms in detecting unmanned aerial vehicles (UAVs). Two datasets were created; one utilized the sky and the other had a more complex background. The study's overarching goal was to evaluate how various methods fared in this particular context. In YOLOv5, YOLOv8, Retina Net, and Faster-RCNN, we examine and evaluate indicators with one stage and two stages. The results of this research could provide light on the performance of the chosen models in difficult settings and lead to the creation of more trustworthy UAV detection systems.