Efficient Event-based Intrusion Monitoring using Probabilistic Distributions

Abstract

Autonomous intrusion monitoring in unstructured complex scenarios using aerial robots requires perception systems capable to deal with problems such as motion blur or changing lighting conditions, among others. Event cameras are neuromorphic sensors that capture per-pixel illumination changes, providing low latency and high dynamic range. This paper presents an efficient event-based processing scheme for intrusion detection and tracking onboard strict resource-constrained robots. The method tracks moving objects using a probabilistic distribution that is updated event by event, but the processing of each event involves few low-cost operations, enabling online execution on resource-constrained onboard computers. The method has been experimentally validated in several real scenarios under different lighting conditions, evidencing its accurate performance.