Chelonia mydas detection and image extraction from field recordings

Abstract

Wildlife videography is an essential data collection method for conducting. The video recording process of an animal like the Chelonia mydas sea turtles in its habitat requires setting up special camera or by performing complex camera movement whilst the camera operator maneuvers over its complicated habitat. The result is hours of footage that contains only some good data that can be used for further animal research but still requires human input in filtering it out This presents a problem that artificial intelligence models can assist, especially to automate extracting any good data. This paper proposes usage of machine learning models to crop images of endangered Chelonia mydas turtles to help prune through hundreds and thousands of frames from several video footages. By human supervision, we extracted and curated a dataset of 1,426 good data from our video dataset and used it to perform transfer learning on a you only look once (YOLO)v7 pre-trained model. Our paper shows that the retrained YOLOv7 model when run through our remaining video dataset with various confidence scores can crop images in the field video recordings of Chelonia mydas turtles with up to 99.89% of output correctly cropped thus automating the data extraction process.