کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

Dynamic Gait Analysis Using Hybrid Neural Networks: Online classification and Prediction of Locomotion Modes

نویسندگان :

Samira Dehghanian¹ Hamed Jalaly Bidgoly²

1- دانشگاه صنعتی ا 2- دانشگاه صنعتی اص

کلمات کلیدی :

CTN،Attention،Gait Classification،Gait Prediction

چکیده :

Dynamic gait analysis is a cornerstone in wearable robotics and rehabilitation, enabling advanced applications such as exoskeleton control and personalized healthcare. This paper proposes a hybrid neural network integrating convolutional layers, bidirectional LSTMs, and an attention mechanism to achieve robust gait classification and prediction across diverse locomotion modes, including level ground, stair ascent/descent, and ramp ascent/descent. The model processes time-series data from inertial measurement units (IMUs), effectively capturing local temporal patterns and long-term dependencies. A two-stage framework is introduced, starting with the prediction of gait parameters such as joint angles, followed by classification of locomotion modes. Experimental results demonstrate the model’s superior performance, achieving 98.92% accuracy in classifying gait conditions and reliably generalizing across subjects. This approach highlights the potential for seamless integration into wearable assistive devices, providing enhanced adaptability and precision in diverse real-world scenarios.