Human Activity Recognition from Body Sensor Data using Deep Learning.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Wearable Technology

Background:

• Human activity recognition (HAR) using wearable sensors is crucial for e-health applications, particularly for monitoring elderly and impaired individuals.
• Accurate HAR is challenging due to the complexity and variability of human movements.
• Existing methods often struggle with the precision required for effective rehabilitation and smart home monitoring.

Purpose of the Study:

• To develop an accurate and automated human activity recognition system using wearable body sensor data.
• To address the classification challenges in HAR by proposing a novel deep learning approach.
• To enhance the robustness and speed of activity recognition through advanced feature processing techniques.

Main Methods:

• Human activity recognition framed as a classification problem using wearable sensor data.
• Extraction of key features from raw sensor data.
• Application of Kernel Principal Component Analysis (KPCA) and Linear Discriminant Analysis (LDA) for feature enhancement.
• Training a Deep Belief Network (DBN) model with the processed features.

Main Results:

• The proposed Deep Belief Network (DBN) model demonstrated high accuracy in human activity recognition.
• Experimental results on a real-world dataset confirmed the effectiveness of the DBN approach.
• The DBN model outperformed other comparative algorithms in activity recognition performance.

Conclusions:

• The developed Deep Belief Network (DBN) model offers a promising solution for accurate and automated human activity recognition.
• Feature extraction and dimensionality reduction techniques (KPCA, LDA) significantly contribute to the model's robustness and efficiency.
• This research validates the potential of deep learning for advancing e-health applications through reliable human activity recognition.