Sistema de Deteção de Quedas Automático Baseado em Vídeo

• Autor(a) principal: Leal, João Gonçalo Nunes
• 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/10400.22/23979
• topic: 3D CNN; automatic fall detection; computer vision; deep learning; I3D; RNN; Domínio/Área Científica::Engenharia e Tecnologia

The elderly population faces difficulties in completing certain tasks independently, often re quiring supervision to not only assist them but also to mitigate and notify about potential health risks. Falls, a prevalent and severe problem, pose a high risk of causing hospitaliza tions and fatalities. However, the aging population in developed countries is growing at an unprecedented rate, while the proportion of active age individuals continues to decline. Con sequently, elderly care has become less accessible as caregivers are confronted with a larger number of patients. Nonetheless, conventional fall detection methods, typically triggered by victims themselves, are unreliable and inadequate. This thesis proposes an automatic alternative to existing methods, presenting a computer vision-based Fall Detection System (FDS) that utilizes a two-stream Inflated 3D Convolutional Neural Network (I3D) in con junction with a Recurrent Neural Network (RNN). To enhance the available datasets, a new collection of simulated falls was created. Experimental evaluations demonstrate the superi ority of this hybrid model over state-of-the-art fall detection models, achieving an accuracy of 94% and a recall value of 96%. By promptly and accurately detecting falls, a system employing this model could significantly reduce the risk of severe injuries posed to the elderly and physically disabled individuals.