LaMAR

本地化和映射是增强现实 (AR) 的基础技术，可实现现实世界中数字内容的共享和持久性。尽管已经取得了重大进展，但研究人员仍然主要由不代表现实世界AR场景的不切实际的基准驱动。这些基准通常基于具有低场景多样性的小规模数据集，这些数据集是从固定摄像机捕获的，并且缺乏其他传感器输入，例如惯性，无线电或深度数据。此外，它们的地面真相 (GT) 精度大多不足以满足AR要求。为了缩小这一差距，我们引入了LaMAR，这是一种具有全面捕获和GT流水线的新基准，可在大型，不受约束的场景中共同注册由异构AR设备捕获的现实轨迹和传感器流。为了建立准确的GT，我们的管道以全自动的方式将轨迹与激光扫描进行可靠的对齐。结果，我们发布了使用头戴式和手持式AR设备记录的各种大型场景的基准数据集。我们扩展了几种最先进的方法，以利用AR特定的设置并在我们的基准上对其进行评估。这些结果为当前的研究提供了新的见解，并为AR的定位和制图领域的未来工作提供了有希望的途径。Localization and mapping are fundamental technologies for augmented reality (AR), enabling the sharing and persistence of digital content in the real world. Despite significant progress, researchers are still largely driven by unrealistic benchmarks that fail to represent real-world AR scenarios. These benchmarks typically rely on small-scale datasets with low scene diversity, captured from fixed cameras, and lack other sensor inputs such as inertial, radio, or depth data. Furthermore, the accuracy of their ground truth (GT) is mostly insufficient to meet AR requirements. To bridge this gap, we introduce LaMAR, a novel benchmark with comprehensive capture and GT pipelines that co-register real-world trajectories and sensor streams captured by heterogeneous AR devices in large, unconstrained scenes. To establish accurate GT, our pipeline reliably aligns trajectories with LiDAR scans in a fully automated manner. As a result, we release benchmark datasets for various large-scale scenes recorded using head-worn and handheld AR devices. We adapt several state-of-the-art methods to leverage AR-specific settings and evaluate them on our benchmark. These results offer new insights into current research and provide promising avenues for future work in the field of AR localization and mapping.

搜集汇总

背景与挑战

背景概述

LaMAR是一个针对增强现实（AR）定位与映射的基准数据集，旨在弥补现有基准在场景多样性、传感器数据丰富性和地面真值精度方面的不足。它通过自动化流程整合了头戴式与手持式AR设备在大型场景中捕获的异构轨迹与传感器流，为AR技术研究提供了更贴近实际应用场景的评估基础。