Detalhes bibliográficos
| Ano de defesa: |
2015 |
| Autor(a) principal: |
Feijó, Gregory de Oliveira
 |
| Orientador(a): |
Pinho, Márcio Sarroglia
|
| Banca de defesa: |
Não Informado pela instituição |
| Tipo de documento: |
Dissertação
|
| Tipo de acesso: |
Acesso aberto |
| Idioma: |
por |
| Instituição de defesa: |
Pontifícia Universidade Católica do Rio Grande do Sul
|
| Programa de Pós-Graduação: |
Programa de Pós-Graduação em Ciência da Computação
|
| Departamento: |
Faculdade de Informática
|
| País: |
Brasil
|
| Palavras-chave em Português: |
|
| Área do conhecimento CNPq: |
|
| Link de acesso: |
http://tede2.pucrs.br/tede2/handle/tede/6806
|
Resumo: |
Fish monitoring has been recently used in many biological research fields to understand the effects of drug usage, for example. Monitoring tasks require the full trajectory of these animals for later evaluation. Evaluation by human observers is the main approach used nowadays. However, this is not a reliable approach because humans can not maintain focus on a source of information for too long. For this reason, digital image processing techniques have become a popular approach for monitoring tasks. The tracking of a single fish is a relatively simple problem that may be solved with traditional image processing techniques. On the other hand, the tracking of a group of fish is much more challenging. The biggest problem is to maintain each individual’s identity due to the frequent overlapping(occlusion) situation that occurs while these animals move inside the tank. Some known approaches use three-dimensional information obtained by multiple cameras which requires a laborious camera calibration step. Other approaches based on a single camera, can not correctly handle occlusion, resulting in a frequent identity swap between fish. This work presents a multi-object tracking method to track a group of fish in a tank. The proposed method is capable of maintaining the correct identity of each fish even in partial and full occlusion situations. In order to keep the correct identity, we take advantage of the Kalman Filter by estimating the future position of each fish based on its previous one. When there are more than one fish in the same region in the frame image, a partitioning algorithm is responsible for re-establishing each animal’s pose. The proposed algorithm was compared against a manually labeled ground truth in two videos. Preliminary tests show that the proposed method is capable of maintaining the animals identity in 98,04% of the occlusion cases. |
