A Scheme of Autonomous Victim Search at Sea Based on Deep Learning Technique Using Cooperative Networked UAVs

Tuan Do Trong, Vu Dinh Khai, Tran Bao Duy, Manh Vu Van

Abstract: In this research paper, we propose the utilization of Multiple Vertical Take-off and Landing Unmanned Aerial Vehicles (VTOL UAVs) suited for maritime operations, combined with Deep Learning algorithms for automatic victim detection and coming up the solution rescue. This work addresses detecting victims with low light intensity - the challenge of Computer Vision algorithms combined with brute-force search flight systems. A cooperative architecture between two UAVs on a local communication network is devised to monitor the drone’s trajectory in the search progress. The modern flight system is based on Software in the Loop (SITL) simulation consisting of X-Plane 11, Ardupilot Firmware, and the autonomous flight algorithm to detect and rescue victims at sea. The simulation results demonstrate the enhancement of mitigating the effect of light intensity for images captured by UAVs to be used by deep learning technique for ameliorating the number of detected victims at sea with up to 62.6% compared with original data.