Title: A Survey of 3D Reconstruction with Event Cameras

Abstract

Event cameras are gaining traction as formidable vision sensors for 3D reconstruction, distinguished by their ability to asynchronously detect per-pixel brightness variations. Unlike conventional frame-based cameras, event cameras generate data streams that are sparse in space but dense in time, facilitating precise and resilient 3D reconstruction even in difficult environments characterized by rapid movement, poor lighting, or extreme dynamic ranges. These advantages hold great potential for revolutionizing sectors such as robotics, autonomous driving, aerial navigation, and immersive virtual reality.

This paper presents the inaugural comprehensive review focused solely on event-based 3D reconstruction. We systematically organize existing methods by input modality—dividing them into stereo, monocular, and multimodal systems—and further classify them by reconstruction technique, encompassing geometry-based methods, deep learning strategies, and neural rendering approaches like Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). Within each group, techniques are arranged chronologically to illustrate the progression of core ideas and technological advancements. Additionally, we offer an extensive overview of publicly accessible datasets tailored for event-based reconstruction. The survey concludes by addressing critical open issues, including dataset scarcity, the need for standardized evaluation metrics, efficient data representation, and the complexities of reconstructing dynamic scenes, thereby suggesting promising avenues for future investigation. This work is intended to act as a vital resource and a clear, inspiring guide for progressing the field of event-driven 3D reconstruction.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC