Estratégia de alcance para próteses de membro superior baseada em visão neuromórfica

• Autor(a) principal: Gouveia, Eduardo Borges
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFU
• Texto Completo: https://repositorio.ufu.br/handle/123456789/34247; http://doi.org/10.14393/ufu.di.2022.106
• topic: Sistema de controle semiautônomo; visão neuromórfica; controle de próteses; Semi-autonomous control system; control of prostheses; neuromorphic vision; CNPQ::ENGENHARIAS::ENGENHARIA BIOMEDICA::ENGENHARIA MEDICA::TECNOLOGIA DE PROTESES; Engenharia biomédica; Próteses

The loss of a limb has a devastating impact on the individual's autonomy to perform day-to-day tasks. The social, emotional, and physical impacts related to it are extremely hard, especially in cases of loss of upper limbs. The human hand is an interface for using tools and allows social and affective interactions. The usual way of trying to recover some of the functionality lost with amputation is the use of prostheses. However, even the most sophisticated commercial upper limb prostheses present usability obstacles for users, either because of the difficulty of controlling all the degrees of freedom present in sophisticated models or the lack of them in simpler models. The prostheses do not always meet the expectations of users to regain their autonomy. In this way, several approaches aim to develop mechanisms and systems that improve usability and try to give back the functionality of the lost limb by improving the control interfaces of prosthetic arms. The work developed in this dissertation proposes a reach model for upper limb prostheses based on neuromorphic vision, intending to implement a semi-autonomous model for prosthetic devices for individuals who have suffered amputations above the elbow level. The use of neuromorphic vision systems is a good match to embedded systems because of their power and computational efficiency. However, an event-based system requires new computational approach to deal with event-data. Therefore, contributing to the development of new approaches and algorithms for neuromorphic computer vision systems and proving to be an interesting interface for implementation in embedded devices whose parameters of computational requirements and available power are scarce. The model developed by this work was validated using an event-based camera (Dynamic Vision Sensor 128) and a commercial robotic arm (Widow X) which, combined with object detection and tracking system, performs reaching objects task in five experiments that present challenging environments for a system with this purpose. The results obtained using the semi-autonomous model for upper limb prostheses suggest that neuromorphic systems can guide the development of intelligent control devices with low energy consumption and high computational performance.