Prioritized Real-Time UAV-Based Vessel Detection for Efficient Maritime Search

Real-time vessel detection in maritime environments is crucial for diverse applications requiring speed and accuracy. Static camera views often introduce blind spots, compromising detection efficiency. This paper proposes a novel, real-time UAV-based system that uses a dynamic camera control strategy to address this limitation. This strategy leverages pre-defined search patterns, historical data (if available), and real-time sensor information (e.g., radar or LiDAR) to dynamically adjust the UAV's camera gimbal angles. This ensures comprehensive search area coverage while minimizing the risk of undetected vessels. Beyond dynamic camera control, our system incorporates a unique feature-based prioritization scheme for real-time target vessel identification. This scheme analyzes features extracted from captured images, including object size and shape. Additionally, movement analysis helps distinguish stationary objects from potential vessels. The combined approach of dynamic camera control and feature-based prioritization offers significant advantages. Firstly, it enhances search efficiency by systematically scanning the area and prioritizing promising candidates based on dynamic camera adjustments and feature analysis. Secondly, it improves detection accuracy by employing feature similarity (cosine similarity with a reference vessel stored in the system using a ResNet50 module) to reduce false positives and expedite target identification, especially in scenarios with multiple vessels. A comprehensive evaluation process has been conducted to validate the effectiveness of our proposed system in diverse simulated and real-world environments encompassing various conditions (weather, traffic density, background clutter). The results from this evaluation are highly promising and suggest the system's strong potential for real-time vessel detection in maritime environments.