イベント駆動型パーティクルフィルタによる物体追跡技術の提案

抄録

A method is proposed for tracking pulsating objects by integrating a high-speed, low-latency event-driven camera and a newly developed event-driven particle filter. The event-driven particle filter is an efficient algorithm that computes the input from an event-driven camera as an event representation. It updates the particle state by a state transition model and the particle weight by a likelihood calculation for each event occurrence. The proposed method was evaluated by presenting an event-driven camera with a test video of a target in pulsating motion on a fast-refresh-rate display. The experimental results confirm that the method can estimate the next state in a 1-ms period of the target.