AusSmoke: smoke image database and detection algorithm pilot study

The Challenge

Early fire detection and suppression are a critical part of managing bushfire ignitions that have the potential to develop into catastrophic bushfires. Like many bushfire effected organisations and emergency response services, WaterNSW’s early detection capability relies largely upon manually-staffed fire towers. This leaves significant gaps in surveillance.

Developing a reliable automated camera based fire detection system would help address these surveillance gaps and allow response crews to be dispatched earlier, reducing the risk of new ignitions burning out of control. Lack of image data for Australian conditions currently restricts the quality of automated detection.

The Solution

The AusSmoke project developed a free-to-access Australian Bushfire Dataset, evaluation benchmark. Algorithms trained on AusSmoke were shown to improve Australian bushfire detection performance, in this trial with WaterNSW. Future iterations of AusSmoke will make the resource available to innovators working on bushfire related image processing capabilities in academia and industry. These capabilities can then be applied to imagery from watchtowers, drones, satellites and other high altitude sensing platforms.

Industry Partner