EAPTON: Event-based Antinoise Powerlines Tracking with ON/OFF Enhancement

Abstract

Event-based powerlines detection and tracking is a thriving research topic with significance in autonomous powerlines inspection, especially for payload-limited UAVs. Currently, event-based line detection methods are developed to address line detection in dynamic scenes, including UAV-based powerlines tracking. However, due to the motion sensitivity of event cameras, it is challenging to eliminate false line detection caused by clustered events from a complex background. Taking inspiration from animals’ motion vision, we propose the EAPTON algorithm to address the vulnerable robustness against complex backgrounds. In our method, the brightness increments (ON events) and decrements (OFF events) are separately processed in parallel pathways, which is prevalent in animals’ motion vision. To increase selective sensibility to powerlines, we utilize the feature that twin-born ON and OFF events will simultaneously arise on the bilateral side of the powerlines. We verified our algorithm in a simulated powerlines inspection scenario with background noise, it outperforms the SOTA algorithm in robustness. Surprisingly, we found that the relative motion type between UAV and the target (e.g., powerlines) is strongly correlated with the statistical asymmetry of the ON/OFF events, which provides a modal of visual cues for motion estimation in other robotic tasks and may shed a light on the explanation of why ON/OFF neural mechanism exists in numerous animals motion vision systems.