Detecting Human Gait Phases and Conditions with Deep Learning

Al Shareeda S. Y. A., Al-Jadahsa J.

2023 International Symposium on Networks, Computers and Communications, ISNCC 2023, Doha, Qatar, 23 - 26 October 2023 identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/isncc58260.2023.10323976
  • City: Doha
  • Country: Qatar
  • Keywords: classification, deep learning, gait patterns, recurrent neural networks
  • Istanbul Technical University Affiliated: Yes


This study presents an innovative, cost-effective approach for estimating human gait patterns utilizing a Deep Learning (DL) model. In particular, a Recurrent Neural Network (RNN) model is used to accurately predict the four phases of gait by acquiring data from a single IMU sensor placed above the participant's ankle. By conducting experiments on two separate databases and analyzing normal and abnormal gait signals, the model achieves high accuracy in detecting gait phases while highlighting differences in phase shapes between normal and abnormal movements. These results indicate that the proposed model provides a promising and affordable substitute for traditional physical or high-tech examinations when evaluating gait stability. Future research could include incorporating additional sensors and assessing more abnormal gait behaviors to enhance the model's predictive power.