Deep Learning-Based Subject Independent Human Activity Recognition using Smart Lacelock Data.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Wearable Technology

Background:

• Human Activity Recognition (HAR) is vital for healthcare and rehabilitation applications.
• Traditional HAR systems often rely on Inertial Measurement Units (IMU) data.
• There is a need for unobtrusive and effective HAR solutions.

Purpose of the Study:

• To introduce and evaluate a novel wearable device, the Smart Lacelock, for deep learning-based HAR.
• To explore the efficacy of combining IMU and loadcell data for improved activity classification.
• To develop and validate a Convolutional Neural Network (CNN) model for HAR using data from the Smart Lacelock.

Main Methods:

• A novel Smart Lacelock device integrating IMU and loadcell sensors was developed.
• Data was collected from eight participants performing various activities (walking, stair climbing/descending).
• A CNN model with convolutional, max-pooling, ReLU, normalization, dropout, and dense layers was designed and trained.

Main Results:

• The proposed CNN model achieved a high average recognition accuracy of 98.4% on the collected dataset.
• The leave-one-out (L1O) validation technique was employed to assess model generalization.
• The Smart Lacelock device demonstrated significant feasibility for recognizing distinct human activities.

Conclusions:

• The Smart Lacelock device is a feasible and effective tool for Human Activity Recognition.
• The integration of IMU and loadcell data enhances HAR capabilities.
• Further research into the Smart Lacelock's applications in healthcare and rehabilitation is warranted.