Attentional Event-RGB Sensor Fusion for Fast Drone Detection

This paper presents an embedded multi-modal vision system for drone detection, combining an event-based camera, an IMU, and an RGB sensor. The method leverages an attentional mechanism on the event stream and is robust to rotations along all three axes (roll, pitch, and yaw) of a rotating platform. The event-based sensor enables localization of fast moving objects, while the RGB camera provides classification, with the entire system optimized for embedded computational constraints. Performance analysis, with and without attention mechanisms and across various algorithmic variants, assesses the trade-off between computational cost and detection accuracy. The study identifies optimal operating situation for each configuration, validated on an outdoor test data samples.