A Hybrid Approach for Moving Object Detection and Tracking in Event-Based Cameras

Abstract

Event cameras, also called dynamic vision sensor-based cameras, capture visual information differently than frame-based cameras. These asynchronous event streams record brightness changes with great temporal resolution and low latency, making them perfect for difficult applications. Data preparation, noise removal, and object tracking are issues when using event cameras in computer vision. This work offers a hybrid clustering-tracking method to accurately locate and track moving objects in event camera data. This study introduces a hybrid technique for accurate moving object detection and tracking in event camera data. Our method uses the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) density-based algorithm to eliminate noise and cluster and track in two steps. The clustering process uses standard methods led by centroids from prior frames for accuracy. The tracking method predicts cluster positions in later frames using speed and direction information when clusters overlap. Our hybrid technique achieves 95.35% accuracy against ground truth labels, promising major improvements in event-based camera data-based computer vision and robotics applications.