The Swedish Defence Research Agency, FOI, has developed new AI-based methods for processing data from event cameras, a technology the agency is researching for applications including drone detection and tracking.
Event cameras, also called neuromorphic cameras, register changes in a scene pixel by pixel rather than capturing full frames. According to a 2024 FOI study, this makes them better than conventional cameras at capturing fast movement and working in poor lighting.
Among the new results is an AI method for predicting the motion of drones, intended to improve tracking of unmanned aircraft. One of the five papers, titled "Generalized CSTR: Multi-Dimensional Representations for Event-Based Vision," was written by FOI researchers Per Nyblom, David Gustafsson and Tomas Wilkinson. It presents new ways to represent event-camera data streams.
FOI said the advances bring event cameras "even closer to their big breakthrough in protection and surveillance." But techniques for processing this kind of data are "still at an early stage," the agency said.