AI-powered drones in Russia learn that there is no smoke without fire
Drones equipped with standard cameras can now identify fire hotspots from smoke using artificial intelligence.
In Russia, engineers have developed an AI-based system that can detect fires at the earliest stage by analyzing smoke. The technology can be installed on off-the-shelf drones equipped with standard video cameras and used for rapid monitoring of large areas.
Developed at MAI
The system was created by engineers at the Moscow Aviation Institute (MAI). Its core is a computer vision algorithm that analyzes video footage captured by a drone in flight. When signs of smoke appear, the system attempts to distinguish it from steam or fog. Once smoke is confirmed, the software calculates the coordinates of the fire source and automatically transmits them to emergency services. According to MAI, the system can detect a fire within a radius of up to five kilometers and relay the information within five to seven seconds.
Practical value
Project leader Nikita Laletin emphasizes the practical significance of the development.
In the future, the developers plan to add the ability to collect and transmit data on the type of fire and the speed at which it is spreading.
No smoke without fire
Unlike most existing systems, which rely on infrared cameras and react to open flames, the MAI solution works with a conventional video stream. To achieve this, the neural network was trained to analyze optical data. According to the developers, the training process used video materials provided by emergency services from training grounds.
To reduce false positives, the system uses multiple layers of verification. It first analyzes sequences of frames and then evaluates the broader scene context. The neural network ignores clouds and unrelated objects in the sky. At a later stage, the team plans to add a second model that will analyze object textures and further improve the system’s ability to distinguish smoke from steam.
Similar algorithms have previously been used in experimental forest monitoring systems in Russia. What sets the MAI development apart is that it is designed for mass-produced drones that are not equipped with specialized thermal imaging cameras.
A working prototype has already been completed and has passed bench testing. The next stage will involve field trials using a real drone. In 2026, the team plans to present a demonstration model and offer it for testing by emergency services and commercial operators.
