Browsing by Author "Coelho, Paulo"
Now showing 1 - 7 of 7
Results Per Page
Sort Options
- Artificial Intelligence-Driven User Interaction with Smart Homes: Architecture Proposal and Case StudyPublication . Lemos, João; Ramos, João; Gomes, Mário; Coelho, PauloThe evolution of Smart Grids enabled the deployment of intelligent and decentralized energy management solutions at the residential level. This work presents a comprehensive Smart Home architecture that integrates real-time energy monitoring, appliance-level consumption analysis, and environmental data acquisition using smart metering technologies and distributed IoT sensors. All collected data are structured into a scalable infrastructure that supports advanced Artificial Intelligence (AI) methods, including Large Language Models (LLMs) and machine learning, enabling predictive analysis, personalized energy recommendations, and natural language interaction. Proposed architecture is experimentally validated through a case study on a domestic refrigerator. Two series of tests were conducted. In the first phase, extreme usage scenarios were evaluated: one with intensive usage and another with highly restricted usage. In the second phase, normal usage scenarios were tested without AI feedback and with AI recommendations following them whenever possible. Under the extreme scenarios, AI-assisted interaction resulted in a reduction in daily energy consumption of about 81.4%. In the normal usage scenarios, AI assistance resulted in a reduction of around 13.6%. These results confirm that integrating AI-driven behavioral optimization within Smart Home environments significantly improves energy efficiency, reduces electrical stress, and promotes more sustainable energy usage.
- A Deep Learning Approach for Red Lesions Detection in Video Capsule EndoscopiesPublication . Coelho, Paulo; Pereira, Ana; Leite, Argentina; Salgado, Marta; Cunha, AntónioThe wireless capsule endoscopy has revolutionized early diagnosis of small bowel diseases. However, a single examination has up to 10 h of video and requires between 30–120 min to read. Computational methods are needed to increase both efficiency and accuracy of the diagnosis. In this paper, an evaluation of deep learning U-Net architecture is presented, to detect and segment red lesions in the small bowel. Its results were compared with those obtained from the literature review. To make the evaluation closer to those used in clinical environments, the U-Net was also evaluated in an annotated sequence by using the Suspected Blood Indicator tool (SBI). Results found that detection and segmentation using U-Net outperformed both the algorithms used in the literature review and the SBI tool.
- Endoscopy - Brief historical survey, developments and therapeuticsPublication . Libório, Ana; Couto, Sylvie; Cunha, António; Coelho, PauloRapid increase of elder population and the appearance of more diseases needs the creation of new medical devices, as minimal invasive as possible. Nowadays, the endoscopic capsule allows good image and much less stress and pain to the patient than traditional endoscopic catheters. The endoscopy to become as developed as today had many improvements. We present on this paper a brief survey of the historical background of equipment developments, some of the most commonly used endoscopic procedures, their drawbacks and virtues.
- Fiducials Marks Detection to Assist Visually Impaired People NavigationPublication . Costa, Paulo; Fernandes, Hugo; Vasconcelos, Veronica; Coelho, Paulo; Barroso, Joao; Hadjileontiadis, LeontiosAssistive technology enables people to achieve independence in the accomplishment of their daily tasks and enhance their quality of life. Visual information is the basis for most navigational tasks, so visually impaired individuals are at disadvantage due to the lack of information or given unsufficient information about their surrounding environment. With the recent advances in inclusive technology it is possible to extend the support given to people with visual disabilities during their mobility. In this context we propose and describe the SmartVision project, whose global objective is to assist visually impaired people in their navigation through unknown indoor and outdoor environments. This paper is focused mainly on the Computer Vision module of the SmartVision prototype, were we propose a new algorithm to recognise fiducials marks suitably placed on sidewalks, revealing to be a promising solution.
- Landmarks Detection to Assist the Navigation of Visually Impaired PeoplePublication . Costa, Paulo; Fernandes, Hugo; Vasconcelos, Verónica; Coelho, Paulo; Barroso, João; Hadjileontiadis, LeontiosAssistive technology enables people to achieve independence in the accomplishment of their daily tasks and enhance their quality of life. Visual information is the basis for most navigational tasks, so visually impaired individuals are at disadvantage due to the lack of information or given insufficient information about their surrounding environment. With the recent advances in inclusive technology it is possible to extend the support given to people with visual disabilities during their mobility. In this context we propose a new algorithm to recognize landmarks suitably placed on sidewalks. The proposed algorithm uses a combination of Peano-Hilbert Space Filling Curves for dimension reduction of image data and Ensemble Empirical Mode Decomposition (EEMD) to pre-process the image, resulting on a fast and efficient recognition method and revealing a promising solution.
- Low-Resolution Retinal Image Vessel SegmentationPublication . Zengin, Hasan; Camara, José; Coelho, Paulo; Rodrigues, João M. F.; Cunha, AntónioSegmentation process serves to aid the pathology diagnosing process since segmentation filters the interference from other anatomical structures and helps focus on the posterior segment structures of the eye, highlighting a set of signals that will serve for diagnosis of various retinal pathologies. Automatic retinal vessel segmentation can lead to a more accurate diagnosis. This paper presents a framework for automatic vessel segmentation of lower-resolution retinal images taken with a smartphone equipped with D-EYE lens. The framework is evaluated and the attained results were presented. A dataset was assembled and annotated of train models for automatic localisation retinal areas and for vessel segmentation. For the framework, two CNN based models were successfully trained, a Faster R-CNN that achieved a 96% correct detected of all regions with an MAE of 39 pixels, and a U-Net that achieved a DICE of 0.7547.
- A multidisciplinary engineering-based approach for tunnelling strengthening with a new fibre reinforced shotcrete technologyPublication . Barros, Joaquim; Costelha, Hugo; Bento, David; Brites, Nelson; Luís, Rui; Patrício, Hugo; Cunha, Vítor M.C.F.; Bento, Luís; Miranda, Tiago; Coelho, Paulo; Azenha, Miguel; Neves, Carlos; Salehian, Hamidreza; Moniz, Gonçalo; Nematollahi, Mojtaba; Teixeira, Abel; Taheri, Mahsa; Mezhyrych, Anton; Hosseinpour, Emad; Correia, Tales; Kazemi, Hamid; Hassanshahi, Omid; Rashiddel, Alireza; Esmail, BriarThis paper describes the relevant research activities that are being carried out on the development of a novel shotcrete technology capable of applying, autonomously and in real time, fibre reinforced shotcrete (FRS) with tailored properties regarding the optimum structural strengthening of railway tunnels (RT). This technique allows to apply fibre reinforced concrete (FRC) of strain softening (SSFRC) and strain hardening (SHFRC) according to a multi-level advanced numerical simulation that considers the relevant nonlinear features of these FRC, as well as their interaction with the surrounding soil, for an intended strengthening performance of the RT. Building information modelling (BIM) is used for assisting on the development of data files of the involved design software, integrating geometric assessment of a RT, damages from inspection and diagnosis, and the characteristics of the FRS strengthening solution. A dedicated computational tool was developed to design FRC with target properties. The preliminary experimental results on the evaluation of the relevant mechanical properties of the FRS are presented and discussed, as well as the experimental tests on the bond between FRS and current substrates found in RT. Representative numerical simulations were performed to demonstrate the structural performance of the proposed FRS-based strengthening technique. Computational tools capable of assuring, in real time, the aimed thickness of the layers forming the FRS strengthening shell were also developed. The first generation of a mechanical device for controlling the amount of fibres to be added, in real time, to the FRS mixture was conceived, built and tested. A mechanism is also being developed to improve the fibre distribution during its introduction through the mechanical device to avoid fibre balling. This work describes the relevant achievements already attained, as introduces the planned future initiatives in the scope of this project.