Visual perception for autonomous surface vehicles in complex waterway environments: A review

With the development of maritime inspection, environmental monitoring, and intelligent shipping, visual perception has become a key capability for safe and reliable autonomous navigation. However, real-world waterways involve complex dynamic conditions that degrade the robustness and generalization of existing vision-based perception systems. Although numerous studies have investigated individual perception tasks and algorithmic paradigms, a review from an environmental and application-oriented perspective is still lacking. Our study presents a comprehensive survey of visual perception in complex waterway environments through a unified analytical framework. First, major waterway perception tasks are summarized, and representative challenging conditions are categorized into dynamic water-surface disturbances, complex illumination, adverse weather, and small-target environments. Second, publicly available real-world waterway datasets are systematically organized with emphasis on data diversity, environmental complexity, and task suitability. Third, representative perception methods are comprehensively reviewed under different environmental conditions, focusing on perception bottlenecks, robustness limitations, dataset applicability, and practical deployment constraints. Furthermore, we analyze how environmental complexity affects perception failure mechanisms and discuss the interplay among model design, data quality, and deployment constraints. Finally, emerging directions such as multimodal perception, domain generalization, and robust adverse-condition perception are highlighted. With this review, we aim to provide practical guidance for dataset construction, method design, and system level optimization, thereby supporting safer and more reliable autonomous surface vehicle navigation in complex waterway and maritime engineering applications.