A Novel Hybrid Deep Learning Model for Human Activity Recognition Based on Transitional Activities

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Human activity recognition (HAR) algorithms often overlook postural transitions due to their short duration.
• Accurate HAR requires the inclusion of both basic and transition activities for robust performance.

Purpose of the Study:

• To propose a hybrid multi-model deep learning approach for enhanced human activity recognition.
• To incorporate both basic and transition activities into a unified recognition framework.

Main Methods:

• Utilized multiple deep learning models simultaneously for activity recognition.
• Implemented a dynamic decision fusion module for final classification.
• Conducted experiments on publicly available datasets.

Main Results:

• Achieved 96.11% classification accuracy for transition activities.
• Achieved 98.38% classification accuracy for basic activities.
• Demonstrated superior performance compared to state-of-the-art methods in accuracy and precision.

Conclusions:

• The proposed hybrid approach effectively recognizes both basic and transition human activities.
• Integrating transition activities significantly enhances the overall performance of HAR systems.
• The dynamic decision fusion module contributes to the method's superior accuracy and precision.