HelmetPoser: A Helmet-Mounted IMU Dataset for Data-Driven Estimation of Human Head Motion in Diverse Conditions

TL;DR

HelmetPoser tackles reliable head-motion localization under challenging conditions by providing a publicly available helmet-mounted IMU dataset with millimeter-level VICON ground-truth. It shows that data-driven bias correction via LSTM and Transformer networks, trained on pre-integrated IMU data, can markedly improve IMU-based pose accuracy and reduce drift. The work systematically analyzes the impact of IMU data window length, motion style, and sensor type on bias-prediction performance, demonstrating strong cross-motion and cross-sensor generalization with substantial Δα improvements. This dataset and baseline models offer a practical foundation for advancing helmet-based localization in industrial, construction, and emergency-rescue applications.

Abstract

Helmet-mounted wearable positioning systems are crucial for enhancing safety and facilitating coordination in industrial, construction, and emergency rescue environments. These systems, including LiDAR-Inertial Odometry (LIO) and Visual-Inertial Odometry (VIO), often face challenges in localization due to adverse environmental conditions such as dust, smoke, and limited visual features. To address these limitations, we propose a novel head-mounted Inertial Measurement Unit (IMU) dataset with ground truth, aimed at advancing data-driven IMU pose estimation. Our dataset captures human head motion patterns using a helmet-mounted system, with data from ten participants performing various activities. We explore the application of neural networks, specifically Long Short-Term Memory (LSTM) and Transformer networks, to correct IMU biases and improve localization accuracy. Additionally, we evaluate the performance of these methods across different IMU data window dimensions, motion patterns, and sensor types. We release a publicly available dataset, demonstrate the feasibility of advanced neural network approaches for helmet-based localization, and provide evaluation metrics to establish a baseline for future studies in this field. Data and code can be found at https://lqiutong.github.io/HelmetPoser.github.io/.

Figures (5)

• Figure 1: HelmetPoser dataset for data-driven pose estimation of human head motion using IMU, especially for emergency conditions. The dataset is collected using the helmet platform with multiple IMUs with diverse motions and diverse persons. The millimeter-level ground truth is obtained using the VICON system.
• Figure 2: Sensor setup for the helmet system. (a) Side view; (b) Top view of the helmet including two IMUs at different levels.
• Figure 3: LSTM architecture. An IMU data input $\mathcal{I}_{mj}$ of size $w$ (with $m = j - w$) and the previous bias $b_i$ are passed to the LSTM. The hidden state is preserved for the next inference step and the output is passed through a fully connected layer to predict a bias.
• Figure 4: Transformer architecture. A sequence of IMU windows $I_{mj}$(with history $l$) and biases $b_i$ are combined and integrated with positional encoding before being input into the Transformer.
• Figure 5: Dataset Information: The dataset was recorded with ten participants, labeled A through J. The first five participants are male, and the remaining five are female. The diagram illustrates three actions performed by each participant, shown from left to right: walking, running, and stair climbing.