Multiview angle UAV infrared image simulation with segmented model and object detection for traffic surveillance

With the rapid development of infrared (IR) imaging UAV technology, infrared aerial image processing technology has been applied in different fields. But it is not very convenient to obtain real aerial images in some cases because of flight limitations, acquisition costs and other factors. So, it is necessary to simulate UAV infrared images by computer. This paper proposed an improved infrared aerial image simulation method based on open source AirSim. By improving the original AirSim infrared image simulation method, the simulation quality of the infrared image is improved via 3-dimensional segmented model processing. The infrared aerial images of the traffic scene with different viewing angles are simulated via the proposed method in this paper and we constructed infrared traffic scene simulation dataset (IR-TSS) containing seven types of objects. We propose the efficient EfficientNCSP-Net net for the IR-TSS dataset and use popular methods for comparative experiments. The experimental results show that the proposed EfficientNCSP-Net has an mAP₅₀ greater than 96% for object detection on IR-TSS dataset, which is better than those of the existing methods. This paper not only contributes to research on infrared image simulations of traffic scenes, but also has referential significance in other aerial image simulation fields.