Human Detection and Tracking for Drone based Marine Surveillance

Effective object detection and tracking are essential for prompt and precise rescue operations in water-based emergency response scenarios. Acknowledging this necessity, this paper introduces an innovative method using deep learning methodologies. At the heart of this project is the MobDrone dataset, selected to document various situations that occur during water crises. By carefully reannotating a subset of this dataset, new labels of swimmers, rescue victims, and deceased individuals were generated, making it more useful for training and assessing. In this context, the research assesses the effectiveness of YOLOv9 using this specific dataset. By conducting thorough experiments, the study highlights YOLOv9’s exceptional accuracy and resilience, particularly in comparison with its predecessor YOLOv8. This research highlights the possibility of combining advanced object detection algorithms such as YOLOv9 with effective tracking methods like BoT-SORT to transform autonomous systems in water-based emergency response, leading to improved and quicker rescue operations.