Adaptive decoy optimization in the long-wave

This exploratory work addresses privacy concerns arising from long-wave infrared (LWIR) surveillance systems through an adaptive thermal signature generation framework. By integrating adversarial optimization with physics-based constraints, the method creates real-time thermal perturbations that disrupt object detection while ensuring compatibility with actuator capabilities and environmental dynamics. The system formulates perturbation design as a constrained maximization problem balancing detector deception (<i>adversarial loss</i>) against smoothness (<i>L</i>₂ Laplacian penalties) and temporal coherence (Kalman filtering constraints). Validated on the Teledyne ADAS FLIR dataset, this approach achieves a 45% increase in mean detection error rates compared to baseline YOLOv4 models. The framework overcomes limitations of static material-based countermeasures and unconstrained adversarial methods by dynamically adapting to environmental conditions such as motion or temperature fluctuations.