In this study, we propose a method for contactless anomaly detection using event cameras, which are well -suited for industrial environments due to their high temporal resolution and robustness to noise. Our approach reconstructs continuous mechanical vibrations from sparse event data and applies CNN-based classification to spectrogram representations. Experiments on DC motors confirmed that (1) periodic and frequency structures of vibrations can be clearly reconstructed, and (2) anomalies can be accurately detected and classified. This method demonstrates the potential of event-based vision for low-cost, real-time condition monitoring.
Ishikawa et al. (Wed,) studied this question.