A start-up that approaches firefighters. XXII, a software publisher using artificial intelligence techniques and, more particularly, computer vision, approached Meuse’s departmental fire and rescue service (SDIS). The start-up founded in 2015 in Paris has developed a real-time video analytics solution and wants to offer a new fire detection feature starting in September. Ultimately, the idea is to connect the software to a city’s cameras so it can quickly analyze, detect and alert in the event of a fire.
Learning in real conditions
XXII is still in the learning and validation phase and wants to ensure that its systems recognize lights, like car lights. The start-up has already carried out fire experiments, but in other contexts, such as in recycling companies. “But fire is different, you have to have other situations”says Damien Mulhem, Chief Data Officer and co-founder of XXII.
For now, as it is very rare to find cars on fire, the start-up has worked with a lot of learning. But it is preferable to learn and understand vehicle fire starts directly on the ground. “Our algorithms must be fed with knowledge”simply explains Damien Woman.
XXII, therefore, naturally approached the firefighters to join the training exercises carried out by SDIS de la Meuse on the night of the 27th to the 28th of June and during the 28th of June. This approximation was easily made because the XXII teams have a volunteer firefighter who made the connection between the two. SDIS, through the voice of Lieutenant Colonel David Hantzo of SDIS de la Meuse, evokes a “partnership” and ensure that there is “no economic interest” and “This approach is in the collective interest.”
Explanations of the different parameters
During these exercises, “fires in unpolluted cars are lit in different configurations so new firefighters can train”, summarizes David Hantzo. For XXII, this experiment should allow you to validate your algorithms, ensure that they detect fire starts under real conditions, and teach your neural networks about other fire start circumstances. “The goal is to film a fire at various distances and different heights so that our systems are independent of distance and viewing angles”explains Damien Mulhem.
XXII wants to better understand the sources of fire in vehicles and, above all, teach its neural networks the different types of fire and how to analyze them. Here, specialists who are firefighters can explain live to the XXII teams the parameters to be taken into account (flame colors, height, smoke, etc.) and their influence on the evolution of the fire, both to better understand the evolution of firing and convey the correct information. Then the software will be able to automatically detect these fires, analyze them and warn firefighters, informing them at what stage the fire has reached and assessing its severity.
The many uses of AI
The software developed by XXII can be used by anyone, be it a community or a private company and eventually individuals. Damien Woman evokes a “agnostic platform that can be used in different ways”. In addition to fire detection, XXII would also like to develop a drowning detection solution. But data on this subject is scarce, so it must be simulated.
Another issue that interests the start-up is the detection of special vehicles (firefighters, ambulances, police, SAMU, vehicle transport agencies, etc.). Once these vehicles are detected and the software XXII is connected with smart cities, it is possible to control the traffic lights to facilitate the circulation of these vehicles.
For firefighters, artificial intelligence can have many uses, but the cost of this technology currently limits its use. David Hantzo evokes the use of artificial intelligence coupled with sensors in the nozzle holders to very quickly alert firefighters when the fire is about to degenerate when various elements are combined (temperature, light, smoke, heat). There is also personal protective equipment with probes that recreate an operational situation (heat, smoke, etc.) using artificial intelligence. Uses that can multiply in the coming years.