Event-Based Camera Image Processing System Using Topological Mapping

Abstract

In this paper, we focus on image sensor-based line detection systems, which are necessary to handle real-world events in the digital world, and propose a monitoring system using an Event-based camera, which is a sensor with features not found in conventional cameras. To address the main challenges of noise, feature extraction, and object recognition in Event-based cameras, we introduce three novel image processing algorithms: noise removal using Spiking Neural Network (SNN), feature extraction based on topological mapping using Growing Neural Gas (GNG), and object recognition using Graph Convolution Network (GCN). Furthermore, we develop a pan-tilt mechanism using two Eventbased cameras, enabling us to switch between recognition and tracking roles and facilitating effective target tracking. Three main evaluation criteria were used for each experiment: control of the pan-tilt mechanism with two event cameras, object recognition using GCN, and whether the tracking target could be captured at the center of the field of view. Through experiments, we confirmed that the proposed method can perform real-time target tracking using event cameras and that the proposed method is effective for fast-moving objects that cannot be measured with conventional cameras.