Infrared Object Detection

Infrared object detection finds objects from heat patterns, helping perception when visible-light images are dark, obscured, or low contrast, and supporting tasks such as driving, aerial surveillance, and industrial monitoring. Methods include multiscale and edge-aware networks for small targets, CNN and transformer detectors, and visible–infrared fusion using alignment, reliability gating, frequency-aware features, or cross-modal distillation. Benchmarks such as KAIST, FLIR, LLVIP, M3FD, and DroneVehicle evaluate performance across pedestrians, vehicles, and varied conditions; synthetic thermal data and RGB-to-IR translation are also used to address limited annotations. A central challenge is balancing the complementary information from multiple sensors against misalignment, domain shifts, unreliable inputs, and the cost of real-time deployment.