| dc.description.abstract | Unpredictable ship accidents still claim a lot of human life every year even with so many technological advancements. Maritime Search and Rescue missions during such hazards are mostly carried out with costly equipment and manpower that have some inherent estimation biases in many physical quantities. With small-size, lower operational cost, flexible aerial maneuverability, wireless communication, and mathematical computation ability, drones can be useful to minimize the costs and speed up the SAR operations without physical intrusion in dangerous post disaster scenarios. And due to the risky nature of the problem, simulation was the rational path initially.
But there was a shortage of previous literature that tried to especially solve this problem in proper simulation platform. Therefore, in the beginning a high fidelity dynamic marine simulation environment was created using Unreal Engine 4.27, Microsoft AirSim, and ROS which contained a Post Disaster Ship, other many debris, and human victims floating. Then, an autonomous SAR mission was planned and implemented for the drone with various pretrained YOLOv7 models that achieved high accuracy of victim detection. This work was published in IEEE/CVF WACV Conference, 2023. After that another iteration of autonomous simulation for tracking both treading and swimming victims with YOLOv8 pretrained models was carried out in custom environment in Unreal Engine 5.1 which also had satisfactory results. Furthermore, the ID and detected location in latitude and longitude of the tracked victim was made easily accessible for use in concerned places. Finally, the possibility for the cooperation and control of multiple drones working together for SAR missions was thoroughly discussed in the end. | |
