FishEye: A Centroid-Based Stereo Vision Fish Tracking Using Multigene Genetic Programming

Investigating the sharp movements and habitat use of active fishes has traditionally been difficult in aquaculture. This study focuses on improving the prediction of tracking and tagging fish in the three-dimensional form (stereovision). This study used two identical devices to capture videos represented as left and right cameras. As with the location of the aquaculture tank, it was in an environmental outdoor lighting condition containing the three sampled fish. The recorded videos have 20 seconds duration showing the movements of fish. Here, extraction of frames occurs and applies computer vision to get the $x$ and $y$ centroid components. The use of the triangulation method was employed to generate the $z$ point of fish images. Multigene genetic programming (MGGP) was utilized and explored in fish trajectory prediction resulting in 7.78%, 13.34%, and 8.90% mean absolute percentage error for fish 1, 2, 3, and respectively. These findings have prompted the authors and researchers to expand their research to use these methods to track fish.