Fatigue detection system to aid in remote work

• Autor(a) principal: Teixeira, Gonçalo Gomes
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.22/21495
• topic: Covid-19; Artificial Intelligence; Computer vision; Fatigue detection system

During the Covid-19 pandemic there was a noticeable surge in the amount of remote workers. In the aftermath of the pandemic working from home still remains a reality for many workers with noticeable impacts on the mental health of people. With the increased stress caused by current situation and the harder time establishing boundaries there was an increase in the overall stress and fatigue in workers, leading to burnouts. Fatigue detection systems are used in several areas, mainly in the automotive industry as a mean to decrease the number of accidents. This research started by approaching the Artificial Intelligence (AI) area and its domains, followed by a study of the current techniques used in order to predict fatigue. With the main ones utilising eye state, facial landmarks, electrocardiogram or heart rate. After a research into existing Fatigue detection systems was done in order to identify the strengths of solutions currently in the market, whether in the automotive industry or other applications. This thesis proposes the creation of a system able to detect fatigue in a user as well as warn him when fatigue levels increase. This system incorporates a webcam analysing the users face and performing eye state detection in order to calculate the percentage of the time the eyes are closed (PERCLOS). Heart rate data was also analysed and a model was developed in order to incorporate this data, the percentage of time the eyes are closed, the program the user has open and time of day in order to predict the level of fatigue. By combining these two different techniques this system can be more effective and more accurate in giving predictions of the level of fatigue. The review of literature showed that the conjunction of these two techniques in predicting fatigue is novelty. The developed system also contains integration with smartwatch technology in order to both harness heart rate data as well as communicate with the user via pop up notifications to inform him when fatigue levels get too high. The conclusion of this work is that eye state detection using Artificial Intelligence can achieve a high accuracy and be a reliable tool in identifying fatigue in an user. The combination of Heart Rate and PERCLOS allows the system to have a higher accuracy as well as not being completely reliant on one sensor. The creation of a fatigue prediction model was hindered by the lack of existent data in order to train a model, a problem that could be fixed with the adoption of the system in a broader scope.