Object tracking using a many-core embedded system

• Autor(a) principal: Minozzo, Laercio
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10198/14598
• topic: Real-time color-based object tracking technique; RGB-D sensor; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Object localization and tracking is essential for many practical applications, such as mancomputer interaction, security and surveillance, robot competitions, and Industry 4.0. Because of the large amount of data present in an image, and the algorithmic complexity involved, this task can be computationally demanding, mainly for traditional embedded systems, due to their processing and storage limitations. This calls for investigation and experimentation with new approaches, as emergent heterogeneous embedded systems, that promise higher performance, without compromising energy e ciency. This work explores several real-time color-based object tracking techniques, applied to images supplied by a RGB-D sensor attached to di erent embedded platforms. The main motivation was to explore an heterogeneous Parallella board with a 16-core Epiphany coprocessor, to reduce image processing time. Another goal was to confront this platform with more conventional embedded systems, namely the popular Raspberry Pi family. In this regard, several processing options were pursued, from low-level implementations specially tailored to the Parallella, to higher-level multi-platform approaches. The results achieved allow to conclude that the programming e ort required to e - ciently use the Epiphany co-processor is considerable. Also, for the selected case study, the performance attained was bellow the one o ered by simpler approaches running on quad-core Raspberry Pi boards.