Multi-ratio fusion change detection

In this paper three ratio-based change detection algorithms, dual ratio (DR), multi-ratio (MR) and multi-ratio fusion (MRF), are tested with full motion video data collected from an unmanned aerial vehicle (UAV) platform. The dataset suffers from several practical issues that generally hinder change detection utility and performance: including image registration error, changes in perspective and significant illumination changes. The ratio-based approaches are compared to change detection methods from literature and are found to be more robust to these practical issues. MRF is found to be the top performing method exhibiting a 10% average performance advantage over the next best performing method across all false alarm regions. MRF also outperforms the next best performing method by 22% at low false alarms rates that are critical in many applications.