High quality synthetic particle data: simulating dust, smoke, and fog in Unreal Engine

Dust, smoke, and fog significantly impact human life, affecting traffic safety, public health, and climate dynamics. Monitoring these phenomena is essential but challenging due to the expense and dangers of replicating realworld conditions. This paper proposes using high-quality smoke plumes in Unreal Engine 5 to generate synthetic training data for deep learning models, addressing the shortage of datasets in this specialized area. The utilization of heterogeneous volumes facilitates the accurate calibration of visual effects through adjustable parameters like size, lifetime, color, and transparency. They also integrate external data interfaces for enhanced detail and control. The study explores the challenges of creating a pipeline with high-quality synthetic plumes, emphasizing computational demands and physical realism, and investigates the optimal ratio of real to synthetic data to improve model detection rates. The evaluations of the results demonstrate that the generated datasets enable competitive model performance, underscoring the potential of high-quality synthetic plumes to substantially improve monitoring and detection capabilities in low-visibility environments.