HelmetPoser: a helmet-mounted IMU dataset for data-driven estimation of human head motion in diverse conditions

Abstract

Helmet-mounted wearable positioning systems are essential for enhancing safety and facilitating efficient coordination in industrial, construction, and emergency response environments. Such systems, incorporating technologies such as LiDAR-Inertial Odometry (LIO) and Visual-Inertial Odometry (VIO), frequently face difficulties in localization within challenging environments, marked by conditions like smoke, dust, and a scarcity of distinct visual cues. To address these issues, we propose a new dataset featuring head-mounted Inertial Measurement Unit (IMU) data paired with ground truth, aimed at improving data-driven pose estimation techniques. The dataset captures human head motion patterns using a helmet-mounted system, with data gathered from ten participants performing various activities. We explore the application of neural networks, including Long Short-Term Memory (LSTM) and Transformer architectures, to reduce IMU biases and improve localization accuracy. Additionally, we evaluate the effectiveness of these methods across different IMU data window dimensions, motion patterns, and sensor setups. To promote further advancements, we have made our dataset publicly available. This highlights the capability of advanced neural network models to enhance helmet-mounted localization systems. Moreover, we offer benchmark metrics to serve as a foundation for future investigations in this field. All related data and code are available at https://lqiutong.github.io/HelmetPoser.github.io.