Alsawadi, Motasem;
(2024)
Advancing Action Recognition through Artificial Intelligence: A Comprehensive Approach for Home Safety Monitoring using Skeleton Data and Spatial Temporal Graph Convolutional Neural Networks.
Doctoral thesis (Ph.D), UCL (University College London).
Preview |
Text
Alsawadi_10185865_Thesis.pdf Download (24MB) | Preview |
Abstract
Accidents resulting from falls are a pressing global concern, especially among the elderly, leading to fatalities, post-fall complications, and limitations in daily activities. Our work introduces an efficient action recognition system, with a primary focus on detecting falls in the fewest possible video frames. Instead of a relying in a single stage (e.g., the classification stage) to solve this issue, we break down the problem into smaller components to enhance the overall action recognition system's accuracy and efficiency. To improve the representation of actions, we utilize skeleton data extracted from RGB images, employing the Spatial Temporal-Graph Convolutional Network. We used the BlazePose topology for action recognition for the first time in the state-of-the art. Moreover, we introduce the Enhanced-BlazePose topology. This innovative approach can represent the actions more accurately. On the other hand, to improve the convolution operation effectiveness, we introduce three new skeleton partitioning strategies: the full-distance, the connection and the index splits. These contributions enhance our ability to recognize human body actions. Recognizing that an abundance of features can hinder machine learning algorithms' performance, we incorporate a feature selection layer, utilizing the Stochastic Fractal Search-Guided Whale Optimization Algorithm (SFS-GWOA) to identify critical joint movements during activities. This feature selection not only enhances performance but also reduces computational costs and processing time. Furthermore, our Multi-Stream Graph Recurrent Neural Network architecture, featuring LSTM units, models spatio-temporal features of skeleton data effectively. Our methodologies and approaches are rigorously evaluated using datasets from restricted and non-restricted environments, demonstrating promising results. Benchmark datasets include NTU-RGB+D, MultiCamera Fall, UR Fall, Kinetics, UCF-101, and HMDB-51. These findings contribute to advancing the field of fall detection and ADL recognition, with practical implications for enhancing the well-being of older individuals living alone.
| Type: | Thesis (Doctoral) |
|---|---|
| Qualification: | Ph.D |
| Title: | Advancing Action Recognition through Artificial Intelligence: A Comprehensive Approach for Home Safety Monitoring using Skeleton Data and Spatial Temporal Graph Convolutional Neural Networks |
| Open access status: | An open access version is available from UCL Discovery |
| Language: | English |
| Additional information: | Copyright © The Author 2024. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request. |
| UCL classification: | UCL UCL > Provost and Vice Provost Offices > UCL BEAMS UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Electronic and Electrical Eng |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10185865 |
Archive Staff Only
 |
View Item |
