Training and Testing Database for IMU Based Gait Analysis Methods

Healthy Subjects or Volunteers

Gait analysis, Gait events, Spatiotemporal gait parameters (SGP), Inertial measurement units (IMU), Wearable sensors, Motion capture (Vicon), Deep learning, Machine learning, Healthy adults

The goal of this study is to establish a high-quality, synchronised dataset of gait events (GE) by simultaneously collecting inertial measurement unit (IMU) data and validated ground truth detections using a Vicon motion capture system. The primary objective is to address existing limitations in GE detection - such as poor generalisability, limited data diversity, and lack of precise synchronisation - through a rigorous protocol that ensures accuracy and transparency. The experiment is structured in three phases. First, Vicon-derived GE will be validated and refined using complementary modalities (force plates and video recordings). Next, deep learning (DL) algorithms will be developed and evaluated for GE detection directly from IMU data, with Vicon annotations serving as ground truth. Finally, the impact of differences in GE timing on spatiotemporal gait parameters (SGP) will be analysed to assess the feasibility of using IMU-only systems for reliable gait analysis. By achieving these objectives, the study aims to improve the accuracy of GE detection from wearable sensors and enable more accessible, scalable, and reliable gait analysis outside the laboratory environment.

This project investigates the development of accurate and reliable gait event (GE) detection methods using wearable inertial measurement units (IMUs), validated against goldstandard motion capture data (Vicon). Gait analysis plays a central role in understanding human locomotion and has important clinical applications in rehabilitation, neurology, orthopaedics, and fall-risk assessment. However, current IMU-based approaches are limited by synchronisation issues, small or homogeneous datasets, and insufficient validation against ground truth. This study addresses these gaps by systematically collecting and validating gait data in healthy participants. Data collection will be performed at the Brubotics Rehabilitation Research Center (BRRC) motion analysis laboratory. Participants will complete walking trials at different speeds (slow, self-selected, and fast speeds) along a standardised 10 m pathway. Reflective markers will be placed on anatomical landmarks, and a sacrum-mounted IMU will capture inertial signals. GE will be simultaneously recorded with the Vicon system, complemented by video and force plate data for validation. The study is organised into three phases. Phase 1 validates and refines Vicon-detected GE using complementary modalities. Phase 2 develops and evaluates deep learning algorithms for IMU-based detection, including the exploration of self-supervised learning. Phase 3 examines how differences in GE timing influence spatiotemporal gait parameters (e.g., step time, cadence, asymmetry), with the goal of establishing whether IMU-only systems can serve as reliable alternatives to motion capture. Ultimately, this project will deliver a robust dataset and algorithmic framework that improve the precision and generalisability of IMU-based GE detection.

No related papers found yet.

None listed

Belgium