Gait & posture · 2026
Classify head-turning in walking: Multi-segment IMU sensing with recurrent neural networks.
СтатьяПоддержка исследования: вне правительства СШАНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Hsu Sheng-Ming, Liang Jing Nong, Lee Yun-Ju
Аннотация
BACKGROUND: Head movement is a fundamental component of gait control and environmental orientation during locomotion. Wearable inertial measurement units (IMUs) enable real-time inference of upcoming movement, and early detection of impending head turns during walking is critical for enhancing interactive systems.
RESEARCH QUESTIONS: The study aimed to investigate whether multi-segment IMU data reliably detect and classify upcoming head turns and their direction before and during head turns while walking, and to hypothesize that neural networks effectively capture the temporal integration patterns of early pre-movement and active head turning in gait.
METHODS: Fifty healthy participants performed four head-turning directions (up, down, left, right) while walking, instrumented with eight IMUs placed on the head, chest, left wrist, lower back, and bilateral heels and toes. IMU signals were segmented around detected head turns and processed using a sliding-window strategy across two temporal windows: pre-turn and during-turn phases. Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks, including sliding-window variants, were trained for four-class classification.
RESULTS: Head-mounted sensors achieved maximum accuracy of 98.91% (pre-turn) and 98.73% (during-turn). Distal segment performance was strong: the SW-LSTM at the right heel achieved 96.93% accuracy and 97.05% precision, while the SW-GRU at the left toe achieved 95.85% accuracy and 95.89% precision.
SIGNIFICANCE: Successfully highlighted the optimized sliding-window sizes for predicting pre-head-turning kinematics during walking from wearable IMU data. The crucial role of head and chest sensing for recognizing pre-head-turning kinematics and direction shows promise for gait applications. Real-world validation demonstrates robust performance across head-turning directions, with practical applications in gait-adaptive systems.
RESEARCH QUESTIONS: The study aimed to investigate whether multi-segment IMU data reliably detect and classify upcoming head turns and their direction before and during head turns while walking, and to hypothesize that neural networks effectively capture the temporal integration patterns of early pre-movement and active head turning in gait.
METHODS: Fifty healthy participants performed four head-turning directions (up, down, left, right) while walking, instrumented with eight IMUs placed on the head, chest, left wrist, lower back, and bilateral heels and toes. IMU signals were segmented around detected head turns and processed using a sliding-window strategy across two temporal windows: pre-turn and during-turn phases. Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks, including sliding-window variants, were trained for four-class classification.
RESULTS: Head-mounted sensors achieved maximum accuracy of 98.91% (pre-turn) and 98.73% (during-turn). Distal segment performance was strong: the SW-LSTM at the right heel achieved 96.93% accuracy and 97.05% precision, while the SW-GRU at the left toe achieved 95.85% accuracy and 95.89% precision.
SIGNIFICANCE: Successfully highlighted the optimized sliding-window sizes for predicting pre-head-turning kinematics during walking from wearable IMU data. The crucial role of head and chest sensing for recognizing pre-head-turning kinematics and direction shows promise for gait applications. Real-world validation demonstrates robust performance across head-turning directions, with practical applications in gait-adaptive systems.