Title: Motion-Aware Event Suppression for Event Cameras

Abstract: This study presents the pioneering framework for Motion-aware Event Suppression, designed to real-time filter events generated by independently moving objects (IMOs) and ego-motion. By simultaneously segmenting IMOs within the current event stream and forecasting their subsequent trajectories, the model enables the preemptive suppression of dynamic events prior to their occurrence. The proposed lightweight architecture delivers an inference speed of 173 Hz on consumer-grade GPUs, utilizing under 1 GB of memory. It surpasses existing state-of-the-art techniques on the rigorous EVIMO benchmark, achieving a 67% improvement in segmentation accuracy and operating at a 53% faster inference rate. Furthermore, the method yields substantial advantages for downstream tasks: it enhances Vision Transformer efficiency by 83% through token pruning and boosts event-based visual odometry performance, lowering the Absolute Trajectory Error (ATE) by 13%.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC