Wildfires are detected accidentally

30 Lis 2020

For many years firefighters, foresters, non-profit organizations, insurance companies, and many other institutions actively educate society on wildfire prevention. From primary education, through Smokey Bear to the national Firewise program – we learn how to prevent wildfires and minimize the potential consequences. Thanks to longstanding education 95% of adults and 77% of children recognize SmokeyThe Bear’s admonition, ‘Only you can prevent forest fires’. Almost 1400 local communities in the United States apply the principles of reducing wildfire risks and home protection drawn up in Firewise programs. Undoubtedly, the effects of education, are a major success.

9 out of 10 fires are detected by chance

Thanks to education and worsening fire seasons, the sensitivity to the matters of wildfire in American society is very high. More than 90% of reports are made by random people and only 10% by specialized services. Modern, very expensive, government-financed satellite technologies (MODIS, VIIRS, Landsat-8/OLI, GOES-16, even Australian geostationary satellite Himawari-8) are detecting only a minority of fires. Regardless of the method of detection, time is of utmost importance while fighting a fire. The quicker the fire is detected, the faster the firefighters reach it. It goes without saying: the quicker the fire is under control, the lesser the consequences.

Firefighters do not like to waste time and are ready for work as soon as they receive a report. But how much time passes since the fire started and is in fact detected and reported? Sometimes it takes minutes, sometimes hours – in 90% of cases, it happens by chance.

Human sensitivity and observance is certain, but what about when no one is there to see the fire? How to help local communities?

Cameras instead of eyes

For years, specially designed cameras and software have been used for fire surveillance. The cost of installation is high, and the creation of the complicated and necessary infrastructure makes it almost impossible to be applied on a big scale. Foresters and firefighters are familiarized with IT tools that analyze and detect smoke in images. Unfortunately, those devices do not function at their best. The software takes two pictures and analyzes differences between them, this method’s side effect is the high percentage of false alarms. Additionally, such an approach eliminates the potential input of the local communities.

What if an inexpensive AI-equipped camera, capable of independently detecting fire, was used? Such a perfect surveillance device could be installed on homes, buildings, poles, chimneys, etc. It could automatically notify the residents and firefighters about the detected fire. The smokeD intelligent detector is exactly such a device. It uses its software to detect fire, works 24/7, and is available now. This will help communities be aware and help with early fire detection warnings.

Artificial Intelligence

Traditional methods of image analysis are based on comparing two pictures made at different times. The algorithms search for differences in the images and if any are found, an alarm is sent. It is difficult for an algorithm to find smoke, as it lacks distinct boundaries and is frequently similar to clouds, fog, smog, or dust. Traditional algorithms work very well when there are no clouds in the sky. Unfortunately, their efficiency drops when images include clouds or even their shadows.

The newest research proves that those algorithms supported by AI analyze the images more accurately. The methodology used perceives smoke not as an object, but rather as a noise, an interference disrupting the image. Moreover, the AI is capable of learning about its surroundings – the longer it works, the smarter it gets. Thanks to all those characteristics, the SmokeD camera is capable of detecting smoke in one picture, without the necessity of comparing it to another, which considerably shortens the time of fire detection.

10 minutes and a phone notification

In 2017 SmokeD camera’s software was able to detect fires, on average, in 10 minutes or less from the beginning (meaning the time when smoke appeared above vegetation). The fastest detection was 45 seconds. Some false alarms happened during that year, but they were all a part of the learning process for the AI. The SmokeD system, which consists of a camera and software, using millions of images to learn.

The fire notifications were sent to thousands of users of SmokeD Alerts app. In the newest version of the app, apart from being sent notifications, users can report a fire themselves by taking a picture and sending it to the fire department. This is a perfect tool for firefighters, wildfire zones residents as well as tourists. SmokeD Alerts app is available for free both on the AppStore and Google Play.

According to the report by the National Interagency Fire Center only last year 71,499 fires affecting over 10 million acres were detected. Extinguishing the fires cost almost 3 billion dollars.

42 death toll in California only.

What would be different if all of those fires were detected and reported in up to 10 minutes?