Human activity and anomalous behaviour detection
The number of elderly population is increasing significantly. The elderly living independently need serious attention as in some cases the elderly people are also found dead in their places. Therefore, it is crucial to assist older adults to live a safer independent life. The increasing cost of nursing care and the shortage of health-care workers increase the demand of home-based assisted living in recent time. Thus, home-based health care for elderly people has become an active research domain, among which abnormal activity detection attracts most attention. My thesis aims to provide effective solutions to support independent living for elderly, which consists the following five works.
The first work presents an extensive survey to highlight those technologies for human activity and anomalous behaviour detection which are essential in elderly care. It also discusses the current research trends, their outcomes and effects in elderly care.
The second work predicts human activities from sensor data. I propose and show the effectiveness of employing a new combination of deep learning (DL) methods for human activity recognition (HAR). Specifically, I propose a hybrid architecture which features a combination of Convolutional neural networks (CNN) and Long short-term Memory (LSTM) networks for HAR task. The model is tested on UCI_HAR_dataset which is a benchmark dataset and comprises of accelerometer and gyroscope data obtained from a smartphone.
The third work predicts human activities from video. I use CNN model to predict human activities from Wiezmann Dataset. Specifically, I employ transfer learning to get deep image features and train machine learning classifiers.
The fourth work detects human anomalous behaviour. I use multi-view learning methods to detect human fall. I introduce 3 methods for feature fusion for multiview learning to leverage the inadequacy of single-view learning. I use co-training methods to train each view alternately for better learning.
The fifth work surveys the privacy and security issues that may occur in home care systems which require connected devices exchanging data. Data collection involves information exchange among the connected devices. Therefore, security examination and requirement is important aspect from the data collection perspective.
All the algorithms proposed in this thesis have been validated and evaluated through extensive experiments on publicly available dataset. The results have demonstrated that the proposed algorithms significantly outperform the comparable models in the existing studies in terms of effectively predicting human activities and anomalous behaviour, which are essential for elderly care.