It is almost impossible for any video analytics solution provider to promise exact false alarm numbers or rates. This happens because the monitored scene conditions can never be fully known or predicted. This is the case for Meleagros, too. Here is a brief list of what we do to improve system accuracy and reduce false alarms:
- Scene characterization: the user can easily characterize what a camera sees in specific classes, such as forest, water, sky, built environment, road, etc.
- Zone exclusion: the user can exclude zones where fire is not expected and may produce false alarms (e.g., chimneys).
- Spatiotemporal self-learning algorithms: when the user explicitly indicates false alarms, the system searches for any spatiotemporal patterns in these (e.g., smoke in a specific tile every afternoon) in order to self-adjust its sensitivity.
- Custom adjustment of specific detection parameters by Meleagros support staff, if required
