Monitoramento de peixes-zebra: rastreamento a partir de vídeos

Abstract

This study focuses on the development of a robust zebrafish (Danio rerio) tracking system, a species widely used in scientific research, by applying computer vision techniques, such as the YOLO object detector. It addresses the limitations of traditional methods and the need for strict environmental controls in existing automatic systems. The work proposes a more versatile solution, less dependent on environmental conditions. For the system creation, an in-depth literature review was conducted, discussing various tracking methods and the evolution of convolutional neural networks. The employed methodology involved collecting videos of zebrafish in various environments, manually labeling the fish positions, training the YOLOv4 base, and applying data augmentation to enhance the model’s robustness. The results indicate that accurate labeling and the application of data augmentation techniques improve the model’s accuracy.