An improved human activity recognition technique based on convolutional neural network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Convolutional Neural Networks (CNNs) are integral to artificial neural networks (ANNs), widely applied in computer vision and pattern recognition.
• Advancements in computing power and data availability have expanded CNN applications across diverse fields, including medical surveillance.
• Human Activity Recognition (HAR) using wearable technology and CNNs is crucial for continuous health monitoring.

Purpose of the Study:

• To comprehensively study the application of CNNs in classifying human activity recognition (HAR) tasks.
• To present the enhancement of CNNs from their origins to current deep learning (DL) systems.
• To propose and evaluate a CNN-based model for HAR classification using sensor data.

Main Methods:

• A multi-layered, two-dimensional CNN model was developed to interpret sensor sequence data.
• The model was designed to capture both temporal and spatial data relevant to human activities.
• The publicly available WISDM dataset was utilized for training and validating the HAR classification model.

Main Results:

• The proposed CNN model achieved a high accuracy rate of 97.20% for HAR classification.
• This accuracy surpasses previously reported state-of-the-art techniques in HAR.
• The study demonstrated the effectiveness of deep learning methods in enhancing HAR accuracy.

Conclusions:

• Deep learning methods, particularly CNNs, significantly improve accuracy in human activity recognition.
• The developed CNN model offers a robust solution for classifying human activities from wearable sensor data.
• The findings suggest broader applications for HAR and highlight future research directions in the field.