Uma abordagem bottom-up completa para reconhecimento de atividades humanas em imagens através da pose estimada com redes convolucionais
Nenhuma Miniatura disponível
Data
2019-09-27
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal de Goiás
Resumo
In the last few years, significant improvements in the computer vision were
made, making
it possible to obtain important information from images. Some of the
challenges for a
better understanding of a scene are the detection of people and the
recognition of the
activities they are performing. This work propose a single end-to-end model
able to detect
people, estimate their pose, and recognize each one of their activities by their
pose. The
experiments show that the model has reached the state of the art in the tasks
of person
detection and pose estimation on MSCOCO Dataset 2017, and can recognize
walking,
running, sitting, and standing activities with an F1 score of 0.7344. The model
is real-time
with an inference rate of approximately 20 frames/sec.
Descrição
Citação
SOARES, I. M. Uma abordagem bottom-up completa para reconhecimento de atividades humanas em imagens através da pose estimada com redes convolucionais. 2019. 127 f. Dissertação (Mestrado em Engenharia Elétrica e da Computação) - Universidade Federal de Goiás, Goiânia, 2019.