Real-Time Activity Recognition Using Minimal Biomechanical Features: A Lightweight IMU-Based Classifier for Older Adults

Real-time implementation of human activity recognition systems is often challenging due to high computational costs and the need for extensive datasets. Existing models rarely account for older adults, particularly those with mobility limitations. This study proposes a lightweight classifier for redhuman activity recognition that utilizes minimal biomechanical features based on thigh and foot angles measured by two IMUs to classify activities, including walking, stair ascent and descent, standing, and sitting. Data from 10 young adults were used to train the model, which was subsequently tested on separate cohorts of 5 young and 10 older adults. A KNN classifier was implemented in a smartphone application for real-time use. To optimize its performance, an adaptive segmentation method was developed to accurately detect toe-off in different activities and reduce computational overhead. The model was trained to recognize mobility limitations in older adults, non-alternating leg stair negotiation, where individuals rely on one leg. redRather than using deep learning architectures, the proposed physics-based framework uses simple and interpretable biomechanical features to enable real-time implementation on resource-constrained devices. The system achieved overall 99% detection accuracy for young and old adults with normal movement patterns and achieved 93% accuracy in distinguishing non-alternating leg stair negotiation from normal stair negotiation patterns in older adults. The results demonstrate the feasibility of a computationally efficient model with minimal features for real-time applications. By addressing the movement characteristics of older adults, this study contributes to the development of an accessible health monitoring system for real-world activity detection in aging populations.