Automated Counting of Fish in Diver Operated Videos (DOV) for Biodiversity Assessments
Résumé
Underwater video transects are crucial to assess marine biodiversity. The counting of fish individuals in these videos is labour-and time-intensive. An automation of said counting would create non-biased biodiversity data.-For this purpose, we explored traditional methods of counting animals as well as introduced three new methods to count fish from computer vision derived data (single frame detections) resulting in a holistic and fully automated pipeline for fish abundance extraction. The different methods 1) traditional N max , 2) 1d k-means clustering method, 3) an intuitive clustering approach N Heuristic and 4) a Temporal Convolutional Neural Networks (TCN) counting method are proposed on transect data of three Mediterranean species with different ecological niches. Our results shows evidence of underestimation by the traditional N max while the other methods showed better overall results with the proposed N Heuristic and TCN methods representing the reality the most. With an absolute variation comparable to inter-observer variation, we demonstrated reliable methods for quantifying fish counts within the framework of three different species. For future projects, incorporating a stereo system could provide more detailed insights into species recovery, and the analysis should be expanded to encompass a broader range of species, including both marine and terrestrial ecosystems.
Origine | Fichiers produits par l'(les) auteur(s) |
---|