ADVIO

行人视觉惯性里程计缺乏现实和开放的基准数据集，这使得很难确定已发布方法的差异。现有数据集要么缺乏完整的六自由度地面实况，要么仅限于具有光学跟踪系统的小空间。我们利用纯惯性导航的进步，为视觉惯性里程计开发了一套通用且具有挑战性的真实世界计算机视觉基准集。为此，我们构建了一个配备 iPhone、Google Pixel Android 手机和 Google Tango 设备的测试台。我们提供范围广泛的原始传感器数据，几乎可以在任何现代智能手机上访问，并提供高质量的地面实况跟踪。我们还将 Google Tango、ARCore 和 Apple ARKit 产生的视觉惯性轨迹与学术论坛上最近发布的两种方法进行了比较。数据集涵盖室内和室外案例，包括楼梯、自动扶梯、电梯、办公环境、购物中心和地铁站。Pedestrian visual-inertial odometry lacks realistic and open benchmark datasets, making it difficult to quantify the performance differences between published methods. Existing datasets either lack complete six-degree-of-freedom (6DoF) ground truth, or are restricted to small spaces equipped with optical tracking systems. We leverage advancements in pure inertial navigation to develop a general, challenging real-world computer vision benchmark suite for visual-inertial odometry. To this end, we built a testbed equipped with iPhones, Google Pixel Android smartphones, and Google Tango devices. We provide a wide range of raw sensor data accessible on nearly any modern smartphone, alongside high-quality ground truth tracking. We also compare the visual-inertial trajectories generated by Google Tango, ARCore, and Apple ARKit against two recently published methods from academic communities. The dataset covers both indoor and outdoor scenarios, including staircases, escalators, elevators, office environments, shopping malls, and subway stations.

搜集汇总

背景与挑战

背景概述

ADVIO是一个用于视觉惯性里程计的开放基准数据集，旨在解决该领域缺乏真实世界数据的问题。它提供了来自智能手机的原始传感器数据和高精度地面实况跟踪，覆盖多种室内外环境，并比较了不同方法的性能。