Recent advancements in maritime surveillance technology have the potential to significantly enhance the safety and efficiency of maritime traffic monitoring. A research team led by Woonyoung Baek from the Department of Computer Engineering at Inha University in South Korea has developed a novel approach to tackle a common problem faced by surveillance cameras: the obscuring of ships by various objects in real marine environments, such as rocks or buoys. Their innovative solution, detailed in a paper published in IEEE Access, introduces a specialized Structure Guided Global and Local Attention Transformer (SGGLAT) designed specifically for image inpainting of obscured vessels.
The SGGLAT is a dual-network system that includes an edge generation network and an image inpainting network. The edge generation network employs an Edge Global and Local Attention Transformer (EGLAT) module, which integrates both global and local image information to produce a more precise edge map. This enhanced edge detection is crucial for accurately identifying the contours of ships that may be partially hidden from view. The image inpainting network then utilizes a Texture Global and Local Attention Transformer (TGLAT) module, focusing on these contours to synthesize realistic textures. This addresses a key limitation of previous models that relied solely on edge maps, which often resulted in overly smoothed textures.
An additional feature of the SGGLAT is the Active Masked Patch Removal Algorithm (AMPRA), which gradually erases the mask covering the obscured areas, thereby improving the overall image restoration process. The model was rigorously tested using a custom ship image dataset developed by MGT, a company specializing in maritime control systems. The results from both qualitative and quantitative experiments indicate that SGGLAT outperforms existing state-of-the-art models, producing more visually plausible inpainting results.
The implications of this research extend beyond academic interest. For sectors involved in maritime operations, such as shipping, logistics, and coastal surveillance, the ability to accurately identify and monitor vessels—even when they are partially obscured—can lead to improved safety protocols and operational efficiency. Companies in these industries could leverage such technology to enhance their surveillance systems, potentially reducing the risk of accidents and improving response times in emergency situations.
“This model produces visually more plausible inpainting results,” Baek noted, highlighting the effectiveness of their approach. As maritime traffic continues to grow, the demand for advanced surveillance technologies will likely increase, creating commercial opportunities for innovations like SGGLAT. The research not only addresses a pressing challenge in maritime surveillance but also opens the door for further developments in computer vision and deep learning applications across various industries.
