Human activities recognition of the aged

• Main Author: Neo, Jerryl Kai Wen
• Other Authors: Soh Yeng Chai
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/167204
• Institution: Nanyang Technological University

This final year project report presents two separate Human Activity Recognition (HAR) systems for monitoring elderly individuals: (1) based on sound and (2) based on smartphone sensors (accelerometer and gyroscope). The sound-based HAR leverages critical sound patterns of everyday activities and objects to detect and recognise human activities and events, while the smartphone Sensor-based system leverages powerful sensors readily available in the modern smartphone for HAR. Both systems are ideal for continuous long-term monitoring of elderly individuals to identify changes in their everyday activities over time. The sound-based HAR system combines Convolution Neural Network (CNN) with Transfer Learning to adapt to problems with smaller datasets to effectively reduce overfitting and improve classification accuracy, advancing beyond the potential of sound-based HAR. This methodology achieved significant improvements in classification accuracy, computational time, as well as the overall fit of the data to the model. The smartphone sensor-based HAR system employs a CNN model to accurately recognise and classify human activities using gathered data from sensors embedded in the modern smartphone, specifically the accelerometer and gyroscope. While the two systems remain separate, the development of both systems is aligned with the aim of developing a comprehensive HAR system, with the data collected from both systems having the potential for use in the continuous long-term monitoring of elderly individuals.