Fish Behavior Detection through Video Frames and Trajectories

November 19, 2021

Detecting automatically fish behaviors can be helpful to monitor fish in tanks, which can save considerable time for biologists. In this project, we developed a system capable of detecting abnormal behaviors, feeding periods, and also interesting moments to be analyzed by biologists. Due to its importance to Lisbon Oceanarium, we focused on sharks and manta rays. The system relies on video frames, and fish trajectories which are represented in a features vector format. To detect abnormal behaviors, we resorted to clustering approaches or a switching vector model, and several classifiers were trained to detect interesting moments. Additionally, it is possible to define a set of species-specific rules regarding the extracted features. Feeding periods are detected using a convolutional neural network, or based on aggregation/motion variability. To evaluate each of the approaches, several metrics are extracted such as accuracy, precision, and recall.

Authors in the community:

• 

  G

  Gonçalo Filipe Antão Adolfo

  ist197090

Supervisors of this institution:

• 

  H

  Helena Sofia Andrade Nunes Pereira Pinto

  ist12906

• 

  A

  Alexandre José Malheiro Bernardino

  ist13761

electrical-engineering-electronic-engineering-information-engineering - Electrical engineering, electronic engineering, information engineering

eng - English

Instituto Superior Técnico