Research on Vision-Based Object Detection Algorithm for Marine Obstacles

Marine targets Obstacle detection is of great significance to autonomous navigation. Marine obstacles detection based on visual images can improve the perception of the surrounding environment to promote autonomous navigation ability, which is a basic requirement for assisted obstacle avoidance and fully autonomous obstacle avoidance. In the real navigation environment, it is necessary to detect specific types of obstacles, such as sailing ships, marine obstacles, floating objects and other objects. Due to the low cost of the ship-mounted camera and the visual information that provides nearly 70% of the effective information, a target detection algorithm for multiple types of obstacles based on visual images is proposed. The proposed detection algorithm can be quickly implemented on the graphics processing unit (GPU), which can achieve approximately real-time image processing. To simulate the real maritime environment, data collection has been carried out in the real sea, and balloons of different colors were used to indicate different types of obstacles. To improve the detection performance of surrounding ships, the collected ship images from the existing public dataset combined to the our collected dataset to improve ship detection performance. The algorithm uses the single-stage target detection algorithm YOLO-V4 as the baseline algorithm. In order to improve the detection performance of small objects, the target detection regression is performed on the multi-layer cascaded feature layer to improve the detection of distant small target obstacles performance. Finally, it is verified that the algorithm can achieve reliable detection performance on the collected dataset.