Aerial Vehicle Detection at Night: A Synthesized Dataset and Performance Evaluation of State-of-the-Art Object Detection Models

Vehicle detection from aerial images is crucial for effective traffic management and safety, especially during night-time when visibility is low and the risk of accidents is higher. However, detecting vehicles at night is challenging due to the lack of adequate training data under low-light conditions, making it difficult to develop models that are both accurate and reliable. In this paper, we address the challenge of vehicle detection in nighttime traffic environments captured by drones. Utilizing the VisDrone-DET2019 dataset, we first transform daytime images to nighttime images using CycleGAN, thereby generating a comprehensive dataset that simulates night conditions. Subsequently, we apply advanced object detection algorithms, including YOLOv8, YOLOvlO, and RTMDet, to identify vehicles within the gen-erated nighttime dataset. Our experimental results demonstrate the effectiveness and robustness of these detection models in low-light conditions, providing significant in-sights into the development of reliable traffic monitoring systems for nighttime scenarios.