Fuzzy Finite State Machine for human activity modelling and recognition

Mohmed, G.M.O., 2020. Fuzzy Finite State Machine for human activity modelling and recognition. PhD, Nottingham Trent University.

Gadelhag Mohmed 2020.pdf - Published version

Download (6MB) | Preview


Independent living is a housing arrangement designed exclusively for older adults to support them with their Activity of Daily Living (ADL) in a safe and secure environment. The provision of independent living would reduce the cost of social care while elderly residents are kept in their own homes. Therefore, there is a need for an automated system to monitor the residents to be able to understand their activities and only when abnormal activities are identified, provide human support to resolve the issue.

Three main approaches are used for gathering data representing the human’s activities; ambient sensory device-based, wearable sensory device-based and camera vision device-based. Ambient sensory devices-based systems use sensors such as Passive Infra-Red (PIR) and door entry sensors to capture a user’s presence or absence within a specific area and record them as binary information. Gathering data using these sensory devices are widely accepted, as they are unobtrusive and it does not affect the ADLs. However, wearable sensory devices-based and camera vision device-based approaches are undesirable to many users especially for the older adults users as they more often forget to wear them and due to some privacy concerns.

Recognising and modelling human activities from unobtrusive sensors is a topic addressed in Ambient Intelligence (AmI) research. The research proposed in this thesis aims to recognise and model human activities in an indoor environment based on ambient sensory device-based data. Different methods including statistical, machine learning and deep learning techniques are already researched to address the challenges of recognising and modelling human activities. The research in this thesis is mainly focusing on the application of Fuzzy Finite State Machine (FFSM) for human activities modelling and proposes ways for enhancing the FFSM performance to improve the accuracy of human activity modelling.

In this thesis, three novel contributions are made which are outlined as follows; Firstly, a framework is proposed for combining the learning abilities of Neural Networks (NNs), Long Short-Term Memory (LSTM) neural network and Convolutional Neural Networks (CNNs) with the existing FFSM for human activity modelling and recognition. These models are referred to as NN-FFSM, LSTM-FFSM and CNN-FFSM. Secondly, to obtain the optimal feature representation from the acquired sensory information, relevant features are extracted and fuzzified with the selected membership degrees, these features are then applied to the different enhanced FFSM models. Thirdly, binary data gathered from the ambient sensors including PIR and door entry sensors are represented as greyscale images. A pre-trained Deep Convolutional Neural Network (DCNN) such as AlexNet is used to select and extract features from the generated greyscale image for each activity. The selected features are then used as inputs to Adaptive Boosting (AdaBoost) and Fuzzy C-means (FCM) classifiers for modelling and recognising the ADL for a single user.

The proposed enhanced FFSM models were tested and evaluated using two different datasets representing the ADL for a single user. The first dataset was collected at the Smart Home facilities at NTU and the second dataset is a public dataset collected from CASAS smart home project.

Item Type: Thesis
Creators: Mohmed, G.M.O.
Date: August 2020
Divisions: Schools > School of Science and Technology
Record created by: Jeremy Silvester
Date Added: 21 May 2021 09:50
Last Modified: 31 May 2021 15:02
URI: https://irep.ntu.ac.uk/id/eprint/42902

Actions (login required)

Edit View Edit View


Views per month over past year


Downloads per month over past year