A Multi-Cluster Tracking Algorithm with an Event Camera

Abstract

Robotic perception continues to be one of the main challenges in autonomous robotics. Accurate and real-time perception is very important as it constitutes the basis of important tasks such as decision making and control. The latency of the sensing pipeline is a major limiting factor for the agility of a robot. A novel sensor called an event camera has been recently developed to overcome the limitations of traditional frame-based ones. Event cameras mimic the human perception system as they measure the per-pixel intensity change rather than the actual intensity level. This paper presents our initial investigation of this novel sensing modality by building an event-based dynamic multi-cluster tracker. This can constitute a building-block for higher-level event-based multi-object trackers. The clustering algorithm will be experimentally evaluated in different scenarios including its use for mapping.