Abstract
We describe a methodology for modeling backgrounds sub- ject to significant variability over time-scales ranging from days to years, where the events of interest exhibit subtle variability relative to the nor- mal mode. The motivating application is fire monitoring from remote stations, where illumination changes spanning the day and the season, meteorological phenomena resembling smoke, and the absence of suffi- cient training data for the two classes make out-of-the-box classification algorithms ineffective. We exploit low-level descriptors, incorporate ex- plicit modeling of nuisance variability, and learn the residual normal- model variability. Our algorithm achieves state-of-the-art performance not only compared to other anomaly detection schemes, but also com- pared to human performance, both for untrained and trained operators.