Assessing the security status of maritime infrastructures is a key factor for maritime safety and security. Facilities such as ports and harbors are highly active traffic zones with many different agents and infrastructures present, like containers, trucks or vessels. Conveying security-related information in a concise and easily understandable format can support the decision-making process of stakeholders, such as port authorities, law enforcement agencies and emergency services. In this work, we propose a novel real-time 3D reconstruction framework for enhancing maritime situational awareness pictures by joining temporal 2D video data into a single consistent display. We introduce and verify a pipeline prototype for dynamic 3D reconstruction of maritime objects using a static observer and stereoscopic cameras on an GPU-accelerated embedded device. A simulated dataset of a harbor basin was created and used for real-time processing. Usage of a simulated setup allowed verification against synthetic ground-truth data. The presented pipeline runs entirely on a remote, low-power embedded system with $$\sim $$ ∼ 6 Hz. A Nvidia Jetson Xavier AGX module was used, featuring 512 CUDA-cores, 16 GB memory and an ARMv8 64-bit octa-core CPU.

Camera systems support the rapid assessment of ship traffic at ports, allowing for a better perspective of the maritime situation. However, optimal ship monitoring requires a level of automation that allows personnel to keep track of relevant variables in the maritime situation in an understandable and visualisable format. It therefore becomes important to have real-time recognition of ships present at the infrastructure, with their class and geographic position presented to the maritime situational awareness operator. This work presents a novel dataset, ShipSG, for the segmentation and georeferencing of ships in maritime monitoring scenes with a static oblique view. Moreover, an exploration of four instance segmentation methods, with a focus on robust (Mask-RCNN, DetectoRS) and real-time performances (YOLACT, Centermask-Lite) and their generalisation to other existing maritime datasets, is shown. Lastly, a method for georeferencing ship masks is proposed. This includes an automatic calculation of the pixel of the segmented ship to be georeferenced and the use of a homography to transform this pixel to geographic coordinates. DetectoRS provided the highest ship segmentation mAP of 0.747. The fastest segmentation method was Centermask-Lite, with 40.96 FPS. The accuracy of our georeferencing method was (22±10) m for ships detected within a 400 m range, and (53±24) m for ships over 400 m away from the camera.