Comparative Analysis of Various Hybrid Neural Network Models to Determine Human Activities using Inertial Measurement Units
Year: 2024
Type: article
Abstract: Human Activity Recognition (HAR) holds a pivotal role in a diverse range of applications that impact various aspects of human life. Advancements in sensor technology and the integration of IoT have expanded the scope of research in HAR through the utilization of deep learning algorithms. End-to-end learning is provided by the advanced deep learning paradigm from complex and amorphous data. Smartphones and IoT wearables are now widely employed in Ambience Assisted Living, e-health monitoring, fitness tracking, biometrics, smart cities, IIoT and other applications. Wearables and Smartphones employ Inertial measurement units (IMU) for the detection of human activities. This research proposes different hybrid neural network model built using GRU, bidirectional GRU, LSTM and bidirectional LSTM with CNN. WISDM, USCHAD, and MHEALTH activity recognition datasets are used to test the method. The hybrid model outperforms the other activity recognition algorithms in terms of accuracy.
Authors: Sowmiya, S., Menaka, D.
Institution: Noorul Islam University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7217459318
