Defesa de Exame de Qualificação – Flávio Gabriel Oliveira Barbosa – 5/12/2018
28/09/2018 16:00
| Defesa de Exame de Qualificação | |
| Aluno | Flávio Gabriel Oliveira Barbosa |
| Orientador | Prof. Marcelo Ricardo Stemmer, Dr. – DAS/UFSC |
| Data
Local |
5/12/2018 9h00 (sexta-feira)
Sala PPGEAS I (piso superior) |
| Prof. Leandro Buss Becker, Dr. – DAS/UFSC (presidente)
Prof. Adilson Gonzaga, Dr. – SEL/EESC/USP Prof. Maurício Edgar Stivanello, Dr. – DAMM/IFSC |
|
| Título
|
Still Image Action Recognition for Android Mobile Devices |
| Abstract: We consider the fundamental computer vision problem of enabling computers to visually understand its surroundings. It is interesting to note that we continuously receive a large amount of visual data, and one of the most remarkable features of the human visual system is how rapidly, accurately, comprehensively and almost unconsciously we process this information, recognizing and understanding the complex visual world. To classify an object, a scene or a situation based on a single image can be trivial for humans, however, to enable a computer perform similarly is challenging. The still image action recognition (SIAR) is the task of localizing and classifying a person’s action based on a single image, and it is an extremely challenging problem due to factors such as person pose variation (e.g climbing), person context (e.g. gardening), and object-person interactions (e.g. phoning). Hence, the SIAR based systems have the potential to provide useful meta-data to many applications such as image understanding, human-computer interaction, indexing and searching of large-scale image archives, and it can be used in several other real-world tasks, as helping impaired people perform daily activities. The state-of-the-art algorithms on SIAR are based on deep Convolutional Neural Networks, which requires high computational power, and the use of these CNN based methods on embedded devices can be challenging due to the lower computational capacity and battery life (compared to high-end computers). The Mobile Artificial Intelligence (MAI) is the methodology of using artificial intelligence methods in the mobile context, as robots or smartphones, which requires to propose or adapt methods that are feasible with the hardware limitations. We intend to propose a SIAR that localizes and classifies human actions on an Android mobile platform with accuracy performance comparable to the state-of-the-art . | |
