Image Reconstruction from Event Cameras for Autonomous Driving

Daniel Dauner

Abstract: Event cameras are novel sensors that output a stream of asynchronous per-pixel brightness changes called 'events' rather than capturing brightness images. They offer outstanding performance in capturing high-speed motion and high dynamic range scenarios where traditional cameras are prone to fail. In particular, autonomous systems can benefit from event cameras by acquiring more robust visual information. Although the events theoretically encode a complete visual signal, event streams are incompatible with conventional computer vision techniques. Recent work has demonstrated the qualitative reconstruction of intensity images from event streams. This approach acts as a bridge between event-based vision and conventional computer vision. This report aims to introduce the field of event vision, present state-of-the-art image reconstruction techniques and examine their application for autonomous driving.

Track: Research Insight

Type: PDF