Neuromorphic Event Sensor-Based Optical Flow Estimation with Occlusion Handling for UAV Perception in Industry 4.0 Inspection Systems

Unmanned Aerial Vehicles (UAVs) require accurate and robust motion estimation to enable autonomous navigation in complex environments. Event cameras sensor, a bio-inspired sensors capturing high-temporal-resolution brightness changes, provide an attractive vision modality under challenging conditions such as high dynamic range and rapid motion. However, event-based sensor optical flow estimation remains challenging due to noise, sparsity, and occlusions inherent in event data, limiting UAV performance. In this work, we present a novel event-based sensor optical flow estimation framework that integrates a lightweight OcclusionNet to pre-emptively mask occluded and noisy events, enhancing the downstream Contrast Maximization (CM) optimization. Our approach extends classical CM by incorporating occlusion-aware data filtering, reducing overfitting, improving convergence, and preserving the sharpness of the image of warped events (IWE). Quantitative evaluation demonstrates our method achieves competitive end-point errors (EPE ≈ 1.7 pixels) and flow angular errors, closely matching ground-truth flow warping loss metrics, while eliminating large error outliers. The model was evelauted on a 1th Gen Intel Core i7-13700 with 32GB RAM, achieving approximately 0.09 seconds per frame on the MVSEC indoor_flying sequences. Qualitatively, our occlusion-aware flow yields sharp IWEs and robust performance under complex motions and lighting, showing strong potential for UAV tasks. This work highlights the benefits of integrating occlusion reasoning into event-based flow estimation, paving the way for reliable and efficient sensory for UAV perception in industry 4.0 inspection system.