Neuromorphic cameras (NCs) provide high dynamic range and microsecond-level temporal resolution, making them well-suited for fast robotic perception. However, at high speeds, they produce large event rates that challenge real-time processing on lightweight embedded platforms. This work introduces an efficient optical flow (OF) approach for NCs, designed for low-power hardware (Raspberry Pi 5). Our method optimizes a Lucas–Kanade-based algorithm through parallelized processing (25% speedup) and gyroscopic derotation (
99% accuracy) and incorporates an adaptive slicing technique (splitting the continuous event stream for further processing) to improve robustness to rapid scene changes. The method is validated on a fixed-wing uncrewed aerial vehicle (UAV) flying at 15 m/s, generating up to 21 million events per second. Despite high event rate, our approach enables real-time altitude estimation and autonomous landing in both daylight and low-light conditions. NC-based altitude estimation matches the performance of a conventional camera in standard lighting (100 000 lux) and remains effective in low-light conditions (30 lux) where the conventional camera fails. These results demonstrate the feasibility of fast, event-based perception on lightweight UAVs using low-power embedded systems.