EROAM: Event-based Camera Rotational Odometry and Mapping in Real-time

TL;DR

EROAM tackles real-time rotational odometry and mapping for event cameras by introducing a spherical event representation and the ES-ICP optimization, operating directly in continuous $SO(3)$ on the unit sphere. The system achieves robust 3DoF rotation estimation and high-quality panoramic reconstructions while maintaining bounded computation through incremental spherical-map maintenance with ikd-Tree and Regional Density Management. Extensive synthetic and real-world experiments demonstrate superior accuracy, robustness to high angular velocities, and significant runtime advantages over state-of-the-art CM- and EGM-based methods, all while using CPU resources. The work enables reliable, high-resolution panoramas and demonstrates the practical viability of purely rotational event-camera pipelines for challenging motion patterns.

Abstract

This paper presents EROAM, a novel event-based rotational odometry and mapping system that achieves real-time, accurate camera rotation estimation. Unlike existing approaches that rely on event generation models or contrast maximization, EROAM employs a spherical event representation by projecting events onto a unit sphere and introduces Event Spherical Iterative Closest Point (ES-ICP), a novel geometric optimization framework designed specifically for event camera data. The spherical representation simplifies rotational motion formulation while operating in a continuous spherical domain, enabling enhanced spatial resolution. Our system features an efficient map management approach using incremental k-d tree structures and intelligent regional density control, ensuring optimal computational performance during long-term operation. Combined with parallel point-to-line optimization, EROAM achieves efficient computation without compromising accuracy. Extensive experiments on both synthetic and real-world datasets show that EROAM significantly outperforms state-of-the-art methods in terms of accuracy, robustness, and computational efficiency. Our method maintains consistent performance under challenging conditions, including high angular velocities and extended sequences, where other methods often fail or show significant drift. Additionally, EROAM produces high-quality panoramic reconstructions with preserved fine structural details.

Figures (21)

• Figure 1: Event-based 3DoF camera pose estimation and panoramic reconstruction. (a) Our method accurately estimates the 3DoF pose of an event camera from continuous event streams. (b) Events are projected onto a unit sphere and aligned using our novel Event Spherical Iterative Closest Point (ES-ICP) algorithm. (c) The aligned event sphere enables the reconstruction of a clear panoramic image with preserved fine details.
• Figure 2: System Overview: The proposed event-based rotational motion estimation system consists of two main modules. The tracking module processes event streams to estimate $SO(3)$ pose using spherical representation and ES-ICP. The mapping module maintains and updates the spherical event map, which supports both tracking and panoramic image generation.
• Figure 3: Spherical projection process: The 3D point $\mathbf{P}_i$ is first projected onto the image plane as $\mathbf{x}^I_i$. After undistortion and normalization, it becomes $\mathbf{p}^n_i$ on the normalized image plane. Finally, $\mathbf{p}^n_i$ is projected onto the unit sphere, resulting in the spherical point $\mathbf{p}^s_i$.
• Figure 4: Illustration of motion sensitivity differences between spherical and pixel representations. When the event camera undergoes a small rotation, the 3D point position relative to the camera changes by $\Delta\mathbf{P}_i$. This results in a measurable change $\Delta\mathbf{p}^s_i$ on the spherical surface. However, in the pixel coordinate system, the motion may not cause a change in pixel location, resulting in $\Delta\mathbf{x}^I_i = 0$. This demonstrates the higher sensitivity and continuous nature of the spherical representation compared to the discrete pixel representation.
• Figure 5: Event spherical frame formation process: Each downward arrow represents a triggered event $e_i$. Colored backgrounds indicate time segments of duration $1/f$. The brace underneath shows the selection of the first $n$ events within each time segment to form an event spherical frame.