Manuscript received November 30, 2025; revised January 16, 2026; accepted March 5, 2026; published May 29, 2026.

Abstract—Underwater object detection plays a vital role in marine exploration, surveillance, and ecological monitoring. However, the presence of light scattering, color distortion, and low visibility makes accurate detection and tracking in underwater environments extremely challenging. To address these issues, this work proposes an efficient underwater object detection and tracking framework integrating the Single Shot MultiBox Detector (SSD), Deep Simple Online and Realtime Tracking (SORT), and Kalman Filter (KF). The SSD model is employed for real-time object detection due to its capability of multi-scale feature extraction and fast inference. Detected objects are then tracked across frames using the Deep SORT algorithm, which combines appearance descriptors and motion information for robust identity preservation. To further enhance trajectory estimation and reduce noise, the Kalman Filter is incorporated to predict object positions in occluded or visually degraded frames. The proposed hybrid approach demonstrates improved stability and accuracy in underwater video sequences utilizing the benchmark Underwater Object Tracking (UOT32) dataset. Performance evaluation based on precision, recall, and F1-Score confirms the model’s robustness in identifying and tracking objects under varying underwater conditions. The obtained results with a precision of 97.39%, a recall 100%, and F1-Score of 0.986 indicate that the integration of SSD with Deep SORT and Kalman filtering offers a reliable and computationally efficient solution for real-time underwater detection and tracking applications.