Hybrid Visual Tracking for Biodiversity Mass Estimation in Underwater ScenariosÂ
Monitoring and tracking marine life is essential for the preservation and management of aquatic ecosystems. A critical aspect of this task involves estimating the size of marine organisms, which serves as an important indicator of their health and overall well-being. However, underwater environments pose significant challenges for accurate detection and measurement due to issues such as occlusion, low visibility, and overlapping objects. In this study, a two-stage framework is proposed to address these challenges. Initially, a U-Net-based architecture is employed to segment the object boundaries with high precision. Nonetheless, due to overlapping instances, accurately determining the number of individual organisms remains difficult. To mitigate this issue, a YOLO-based architecture is integrated to initialize object instances within the scene. The combined use of YOLO for object detection and U-Net for boundary segmentation facilitates more accurate estimation of fish size, even in complex and cluttered underwater environments.