The UMA-VI dataset: Visual–inertial odometry in low-textured and dynamic illumination environments

Abstract

<jats:p> This article presents a visual–inertial dataset gathered in indoor and outdoor scenarios with a handheld custom sensor rig, for over 80 min in total. The dataset contains hardware-synchronized data from a commercial stereo camera (Bumblebee®2), a custom stereo rig, and an inertial measurement unit. The most distinctive feature of this dataset is the strong presence of low-textured environments and scenes with dynamic illumination, which are recurrent corner cases of visual odometry and simultaneous localization and mapping (SLAM) methods. The dataset comprises 32 sequences and is provided with ground-truth poses at the beginning and the end of each of the sequences, thus allowing the accumulated drift to be measured in each case. We provide a trial evaluation of five existing state-of-the-art visual and visual–inertial methods on a subset of the dataset. We also make available open-source tools for evaluation purposes, as well as the intrinsic and extrinsic calibration parameters of all sensors in the rig. The dataset is available for download at http://mapir.uma.es/work/uma-visual-inertial-dataset </jats:p>

Run Your Visual-Inertial Odometry on NVIDIA Jetson: Benchmark Tests on a Micro Aerial Vehicle

“…It is crucial to prove that VO/VIO work well in a real environment with several harsh cases. Zuñiga-Noël et al [34] proposed an in/outdoor dataset in which low-texture scenes or scenes with dynamic illumination are included. These conditions are difficult cases of vision-based odometry.…”

Benchmarking Pedestrian Odometry: The Brown Pedestrian Odometry Dataset (BPOD)

“…Following the development of structure from motion (SfM) techniques, several datasets have used SfM to generate ground-truth camera trajectories. For example, [32] and [47] use COLMAP and Pix4D respectively to generate their ground truths. However, unlike these carefully constructed hand-held and MAV datasets, our dataset contains rapid blur and feature-less segments that make SfMbased ground truth generation infeasible.…”

“…Similarly, a dataset whose images have not been photometrically calibrated (i.e., where exposure times, the camera response function, and lens vignetting have been measured) disfavors direct methods. These and other nuances have led to an abundance of benchmarks with varying targets [30], [31], [32], [33], [34], [35], [36]. Benchmarks are inherently taskoriented and must be constructed carefully to satisfy the requirement of the application at hand.…”

“…Pedestrian tracking from head-mounted cameras is an application for which VO datasets are not available to the best of our knowledge. Existing datasets are commonly [34] Hand-held Indoors Global Shutter Monocular/Auto Exposure Motion Capture ICL-NIUM [37] Synthetic Synthetic-Indoors Synthetic Monocular/Synthetic Synthetic MineNav [38] Synthetic Synthetic-Outdoors Synthetic Monocular/Synthetic Synthetic UMA-VI [32] Hand-held Indoor/Outdoors Global Shutter Stereo / Auto Exposure SfM PennCOSY [33] Hand vehicle-mounted [13], [39], [40], Segway-mounted [35], MAV-mounted [41], [36], or hand-held [42], [43], [44], [29].…”

Graph-based fusion of magnetometer and learned-based inertial odometry

“…Camera-based SLAM include visual odometry [4] and visualinertial odometry [5] [6] based built-in camera and inertial sensor of the smartphone is a low-cost and high-percision positioning solution with the potential to be widely used. Unfortunately, VIO still cannot work well in low-textured and dynamic illumination environments [7]. Wireless sensorbased methods (e.g., Ultra-wideband, 5G) rely on pre-arranged signal base stations.…”