Long-Term Gait-Balance Monitoring Artificial Intelligence System for Various Terrain Types

Abstract

A long-term gait-balance monitoring system for various terrain types was developed using an inertial measurement unit (IMU) and deep-learning model. The system aims to identify unstable gait caused by lower-limb degeneration to prevent fall-related injuries. Unlike previous studies that have only focused on gait stability in flat terrain walking, the proposed system is also capable of analyzing stability on stairs and slopes. A lightweight, nine-axis IMU was used for data collection, and a combined convolutional neural network with gated recurrent unit model was implemented on the portable Raspberry Pi Zero 2 W for predicting Berg balance scale (BBS) scores. The BBS scores and gait data were then wirelessly transmitted to a cloud provider for long-term data storage. The system is as small and lightweight as a baseball and can monitor users for extended periods. The system can identify abnormal balance scores to provides physicians with long-term gait information, assisting their analysis and decision-making. This prevents falling and the corresponding consumption in healthcare resources that comes with fall-related injuries.

Keywords

• Deep learning
• gait balance
• long-term monitoring
• edge computing
• Berg balance scale (BBS)