Event-based vision has recently emerged as a powerful paradigm for capturing visual information with high temporal precision and minimal redundancy. Unlike traditional frame-based cameras, event cameras asynchronously detect per-pixel brightness changes with microsecond resolution. This bio-inspired sensing approach offers substantial advantages for dynamic scenes, low-latency perception, and energy-efficient computation, making it highly relevant to applications in robotics, vision, and neuromorphic computing. A fundamental tool in analyzing visual and temporal signals is the Fast Fourier Transform (FFT), which efficiently computes the spectral representation and forms the foundation of modern signal processing, communication, and control systems. However, extending Fourier analysis to asynchronous, event-driven data poses unique computational and theoretical challenges. This paper presents an open-source implementation of the event-based Fourier Transform (eFFT), a novel algorithm for efficiently computing the exact 2D discrete Fourier transform of the spatial information in asynchronous events generated by an event camera. The proposed eFFT-C++ translates the theoretical method into a modular and optimized C++17 library, implemented as a header-only package with Eigen3 dependencies. It supports both event-by-event and packet-based processing modes, reusing intermediate computations to minimize overhead while allowing the current 2D spectrum to be queried after each update. Benchmarks against FFTW3 demonstrate exactness and efficiency, with per-event update times in the order of microsecond for common frame sizes. The library enables reproducible research, provides validated code to accompany the original eFFT publication, and offers a foundation for extending event-based frequency analysis to robotics, computer vision, and neuromorphic computing in applications where rapidly updating spectra is useful, such as denoising and filtering, pattern analysis, or tracking.
Tapia et al. (Thu,) studied this question.