Eventos Anais de eventos
COBEM 2017
24th ABCM International Congress of Mechanical Engineering
DENSITY BASED FISH COUNTING WITH COMPUTER VISION FOR AQUARIUM AND BOX STORAGE IN NOISY ENVIRONMENTS
Submission Author:
Thiago Henrique Gomes Lobato , PA , Germany
Co-Authors:
Gustavo da Silva Vieira de Melo, Thiago Henrique Gomes Lobato
Presenter: Thiago Henrique Gomes Lobato
doi://10.26678/ABCM.COBEM2017.COB17-1676
Abstract
When making fishing inventory there is hardly room for optimization. The process takes too long, reaching days when the number of aquariums and boxes are very high. Sometimes fishes need to be removed from their recipient and it can be a harmful process to them, especially for some kinds of ornamental ones. To avoid this time consuming and difficulty procedure, a very imprecise counting can be done with less frequency, counting fishes in the aquariums or in the boxes by eye, which, however, can be very prone to errors because of the fish size and quantity, producing errors that are greater than 40%. To address that problem, a system of fishing counting with computer vision was proposed, using a grid image binarization together with image segmentation techniques such as color and texture segmentation. The binary image could then be filtered and the fishes be counted with the use of element density, reducing the error caused by the fish overlap. The system was able to reach a high accuracy (ca. 93%) even in noisy environments, with high ambience light and reflections. The proposed system is way faster than the conventional way, does not harm the fishes, is more precise than the bare eye and can be automated, proving the algorithm high value in fish counting.
Keywords
Computer Vision, Fish Counting, image segmentation
