EventStream: Spatiotemporal Event Clustering for Feature Detection and Tracking
Yehia A. Helwa, Louis B. Kratchman, Robert H. Bishop · 2024
The event camera or neuromorphic camera is a biologically inspired imaging sensor with high temporal resolution, low power requirements, and a high dynamic range compared to traditional cameras. A scene representation from an event camera is sparse due to the camera's asynchronous pixels that only detect brightness changes known as "events". The sparse scene representation of event cameras requires less data to represent a moving scene than conventional cameras. However, most feature detection algorithms for event data have relied on conventional frame-based vision processing techniques, collecting events into frames rather than capitalizing on the inherent sparse scene representation. This paper leverages characteristics of event data flow to perform efficient feature identification and tracking by adapting the DenStream algorithm for an event camera. The performance of Eventstream was evaluated compared to existing approaches and showed improvement in feature tracking accuracy.