[Paper Review] OxIOD: The Dataset for Deep Inertial Odometry

Changhao Chen, Peijun Zhao+4 more|arXiv (Cornell University)|Sep 20, 2018

Indoor and Outdoor Localization Technologies24 references59 citations

TL;DR

OxIOD releases a large, diverse inertial odometry dataset with ground-truth labels for 158 sequences totaling 42.587 km, enabling learning-based inertial navigation research.

ABSTRACT

Advances in micro-electro-mechanical (MEMS) techniques enable inertial measurements units (IMUs) to be small, cheap, energy efficient, and widely used in smartphones, robots, and drones. Exploiting inertial data for accurate and reliable navigation and localization has attracted significant research and industrial interest, as IMU measurements are completely ego-centric and generally environment agnostic. Recent studies have shown that the notorious issue of drift can be significantly alleviated by using deep neural networks (DNNs), e.g. IONet. However, the lack of sufficient labelled data for training and testing various architectures limits the proliferation of adopting DNNs in IMU-based tasks. In this paper, we propose and release the Oxford Inertial Odometry Dataset (OxIOD), a first-of-its-kind data collection for inertial-odometry research, with all sequences having ground-truth labels. Our dataset contains 158 sequences totalling more than 42 km in total distance, much larger than previous inertial datasets. Another notable feature of this dataset lies in its diversity, which can reflect the complex motions of phone-based IMUs in various everyday usage. The measurements were collected with four different attachments (handheld, in the pocket, in the handbag and on the trolley), four motion modes (halting, walking slowly, walking normally, and running), five different users, four types of off-the-shelf consumer phones, and large-scale localization from office buildings. Deep inertial tracking experiments were conducted to show the effectiveness of our dataset in training deep neural network models and evaluate learning-based and model-based algorithms. The OxIOD Dataset is available at: http://deepio.cs.ox.ac.uk

Motivation & Objective

• Provide a large, diverse, labelled inertial odometry dataset to advance learning-based inertial navigation.
• Capture everyday pedestrian motions with varying device placements to reflect real-world usage.
• Enable training and evaluation of deep learning models for inertial odometry and pedestrian dead reckoning.
• Offer precise ground-truth trajectories to support objective benchmarking of models and baselines.

Proposed method

• Collect 158 inertial sequences totaling 42.587 km using off-the-shelf smartphones placed in four attachments (handheld, pocket, handbag, trolley).
• Record high-precision ground truth with a Vicon motion-capture system (0.5 mm accuracy) for most sequences; use Google Tango as pseudo ground truth for large-scale office-floor data.
• Vary motions (halting, slow walking, normal walking, running) and devices (iPhone 7 Plus, 6, 5, Nexus 5) across five users.
• Train and evaluate learning-based inertial odometry models (based on IONet) and simple PDR baselines on OxIOD data.
• Demonstrate velocity and heading regression and end-to-end 2D trajectory reconstruction from raw inertial data using recurrent neural networks.

Experimental results

Research questions

• RQ1 Can deep learning models trained on OxIOD accurately reconstruct pedestrian trajectories from raw inertial data across diverse device placements and motion modes?
• RQ2 How does model-based inertial odometry compare with data-driven approaches when evaluated on OxIOD ground-truth trajectories?
• RQ3 What is the impact of attachment type, motion mode, and device on inertial odometry performance?
• RQ4 Is OxIOD suitable for training models to generalize to new users and devices in real-world scenarios?

Key findings

• A DeepIO (IONet-based) model can reconstruct 2D trajectories from raw inertial data and demonstrate velocity and heading regression across attachments.
• Ground-truth data from Vicon provides 0.5 mm accuracy for 132 sequences (excluding the large-scale subset), enabling precise evaluation.
• OxIOD comprises 158 sequences totaling 42.587 km and 14.72 hours of recordings, exceeding most prior inertial odometry datasets in size and diversity.
• Large-scale localization data using Google Tango provides pseudo ground truth on two office floors for evaluating long trajectories.
• PDR baselines fail on non-step-like placements (e.g., trolley), while learning-based methods show effectiveness across varied real-world placements.