Human Activity Recognition via Hybrid Deep Learning Based Model.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning
• Human-Computer Interaction

Background:

• Human Activity Recognition (HAR) is crucial for health and human-machine interaction.
• Existing AI models struggle with spatial-temporal feature extraction, limiting real-world HAR performance.
• Limited public datasets with few activities hinder research in physical activity recognition.

Purpose of the Study:

• To develop an advanced HAR model addressing limitations in spatial-temporal feature extraction.
• To introduce a new, challenging dataset for physical activity recognition.
• To evaluate the proposed model against traditional and deep learning approaches.

Main Methods:

• A hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model was developed.
• CNN was employed for spatial feature extraction, while LSTM handled temporal information learning.
• A new dataset was created from 20 participants, capturing 12 distinct physical activities using Kinect V2.

Main Results:

• The hybrid CNN-LSTM model achieved a high accuracy of 90.89% on the HAR task.
• An extensive ablation study confirmed the model's superiority over other methods.
• The model demonstrated suitability for practical HAR applications.

Conclusions:

• The proposed CNN-LSTM model effectively addresses the challenges of spatial-temporal feature extraction in HAR.
• The newly generated dataset provides a valuable resource for advancing physical activity recognition research.
• The model's high accuracy indicates its potential for robust and reliable HAR systems.