Recurrent Transformation of Prior Knowledge Based Model for Human Motion Recognition.

Area of Science:

• Computer Science
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Human activity recognition (HAR) using wearable sensors is crucial for daily applications.
• Existing HAR methods often treat motion recognition as a standalone classification problem, neglecting physical properties and temporal dynamics.
• This approach leads to data dependency, the curse of dimensionality, overfitting, and models lacking intuitive understanding.

Purpose of the Study:

• To develop a novel human activity recognition model that leverages domain knowledge and temporal information.
• To address the limitations of traditional HAR methods, such as data inadequacy and model interpretability.
• To propose a Recurrent Transformation Prior Knowledge-based Decision Tree (RT-PKDT) model for enhanced motion recognition.

Main Methods:

• Analyzed the natural physical properties and temporal recurrent transformation possibilities of human motions.
• Developed the Recurrent Transformation Prior Knowledge-based Decision Tree (RT-PKDT) model.
• Utilized temporal information and a hierarchical classification approach, integrating sensor data with human knowledge.

Main Results:

• The proposed RT-PKDT model demonstrated superior performance compared to conventional methods like SVM, BP neural networks, and Bayesian Networks.
• Achieved a high accuracy rate of 96.68% in recognizing specific human motions.
• The model effectively compensates for data inadequacy by incorporating prior knowledge and temporal dynamics.

Conclusions:

• The RT-PKDT model offers a significant advancement in human activity recognition by integrating physical properties and temporal information.
• This approach enhances recognition accuracy and model robustness, particularly in scenarios with limited data.
• The findings suggest that incorporating domain knowledge into HAR models is a promising direction for future research.