Dynamic Content-based Indexing in Mobile Edge Networks

• Autor(a) principal: Almeida, José Duarte Farinha de
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/160377
• topic: Indexing; Edge Computing; Mobile Computing; Facial Recognition; Clsutering; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

As the number of technological devices continues to grow exponentially, the demand for edge computing has become increasingly urgent. This is due to the need for efficient and rapid data processing, especially in scenarios where latency and bandwidth constraints prevent the use of cloud-based solutions. In this context, we present in this work a study above face recognition and clustering algorithms to sustain the reliability to introduce them in a Face-Based Indexing system, built on top of an ecosystem for sharing data and computation named EdgeGarden. Our study relies on the future implementation of a Face-Based Indexing system with an efficient and scalable solution for face recognition and indexing at the edge, enabling real-time processing of facial data and reducing the burden on centralized systems. We believe that this system has the potential to significantly improve the performance and efficiency of edge computing systems, paving the way for the development of more so- phisticated applications in the future. A study among several face detection and extraction models led to the implementa- tion of the HaarCascade face detection model and the ArcFace with 50-Residual Neural Network, a Convolutional Neural Network (CNN)-based model for face feature extrac- tion. For the indexing phase, we compare and analyze the performance of two clustering density-based algorithms, DBSCAN and Denstream. This study shows that the Den- stream algorithm performs better in terms of computational efficiency, however, in terms of accuracy it did not remark as good as DBSCAN. Therefore this study essentially highlights the selection of face detection and feature extraction models focusing on the creation of a Face-Based Indexing system and the evidence that it works using a static data clustering algorithm as DBSCAN.