The project builds on the Department of Measurement's long-term research in acoustic gunshot detection. The students worked with a specialized development platform designed to run neural networks with extremely low power consumption. Within three months, they trained their own model for gunshot recognition and successfully validated it during practical testing. The system correctly detected and classified all indoor test shots, demonstrating the potential of energy-efficient embedded artificial intelligence for public safety and security applications.
"The main objective of today's IoT devices is to become as energy self-sufficient as possible. If a sensor is installed on a streetlight pole or on the roof of a building, the ideal solution is for it to operate without a connection to the power grid. That is exactly the direction the students pursued, and the resulting power consumption is truly impressive. Powered by a standard lithium battery, such a device could operate for several years," says Prof. Jan Holub, Head of the Department of Measurement at CTU FEE.
The project was carried out within the Cybernetics and Robotics study programme as part of the compulsory Team Project course, which aims to teach students how to collaborate on real engineering challenges. This year's team brought together students from Europe and Asia, who jointly designed, trained, and tested a fully functional solution over the course of the semester.
"What makes me happiest is not that the device works, but that the students genuinely enjoyed working on it. It wasn't a commercial contract or a grant-funded assignment—it was part of their education. They experienced the entire development process, from the initial idea to successful real-world validation," adds Prof. Holub.
The student project is part of the Department of Measurement's ongoing research into acoustic gunshot detection, which has been underway since 2016. The department's system is capable of detecting a gunshot, determining the direction of the incoming sound, and subsequently localizing the shooter. The research has recently advanced even further with a new paper published in the prestigious journal Measurement, where the research team introduced a novel acoustic signal processing method. The new approach significantly improves the accuracy of estimating the direction of an incoming gunshot while maintaining the same hardware requirements and computational complexity, representing another important step towards the practical deployment of the technology in security applications.
