Review of Advanced Techniques in Underwater Object Detection and Tracking

Detecting and tracking underwater objects is difficult because of the changing water conditions, such as low visibility, color distortion, obstructions, and fast-moving objects. These conditions make it hard for traditional computer vision systems to work well. So, we need strong, flexible, and computationally efficient solutions. This review paper looks at three new and interesting methods for tracking objects underwater in detail and compares them: (1) DeepSeaNet, which combines EfficientDet with a Bi-directional Skip-connected Feature Pyramid Network (BiSkFPN); (2) a lightweight Kernelized Correlation Filter (KCF)-based tracking framework that uses feature fusion and adaptive scale estimation to improve its performance; and (3) an occlusion-aware Adaptive Deep SORT tracker that uses Gaussian Mixture Models (GMM) to find objects and LSTM-based memory management to keep track of identities. We look at architectural design of each method, tracking accuracy, ability to adapt to underwater distortions, and ability to be used in real time. This comparison gives us useful information about how to improve intelligent underwater vision systems in the future. This has effects on environmental monitoring, marine biology, autonomous underwater vehicles, and surveillance applications.