Human activities recognition in smart living environment

• Main Author: Gao, Shiyao
• Other Authors: Soh Yeng Chai
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/139046
• Institution: Nanyang Technological University

Over recent years, Human Activity Recognition (HAR) has played a very significant role in this emerging field by connecting the underlying sensors data into applications. Previous researches were focused mainly on wearable sensors that may be an inconvenience to users to record data and detect human activities. Therefore, with today’s continuous advancements in technology, mobile devices are getting smarter by the day with increased functions which include more significant memory space and built-in sensors are continually developing and updating to better influence our daily lives with ease. The use of smartphones has been used in the most appealing and challenging application in ways to be able to sense human body motion and gathering information from people’s actions. With these sensors, the identification of movements by smart occupants is captured together with their surrounding environment. In this Final Year Project, a holistic perspective on human activity recognition with the use of smartphone sensors approached by different machine learning classification methods is proposed. Accelerometer and gyroscope sensors that are embedded in the user’s smartphone allow us to retrieve data information to be classified to recognize human activities. In addition, to infer a user’s situation in daily life, information can be retrieved from a microphone which can classify almost all activities that process specific sound patterns. Results of the various approaches are compared in terms of accuracy precisions, indicating the effectiveness of the proposed integrated system. In the cross-validation test, the highest accuracy achieved for Human Activity Recognition is 88.5% with the use of Ensembles whereas in the case of Audio Recognition of Energy-Related Equipment with the help of Support Vector Machines, high accuracy of 94.3% is observed as well.