Optical camera systems overlooking a maritime infrastructure, are a good choice for the rapid assessment of the situation with the ability to support in tasks of monitoring, logistics, safety and security. Due to the number of video screens and amount of information typically available, personnel may not be able to easily keep track of everything relevant, such as ships, vehicles or other dynamic elements. It, therefore, becomes important to enable the automatic recognition of objects, including the extraction of their key parameters, which are then presented in an easily understandable format, for example on a map. Two-dimensional (2D) maps have limitations displaying objects in terms of the accuracy of presented information, such as height and object depth. Three-dimensional (3D) vision offers more spatial and semantic information about the situation. This requires the 3D reconstruction of the recognized objects for their display on a 3D static map. Cameras with high-resolution images which are analyzed by a local server may create bandwidth wastages, bottlenecks, and time delays, disabling the real-time capability of the system. Remote edge nodes with embedded architectures and an integrated high-resolution camera, however, allow for a faster and more efficient processing.
The main goal is the research and development of a unified pipeline for the automatic recognition of objects and their features, along with their 3D reconstruction to be presented to a maritime situational awareness system. This will be achieved through the use of an embedded device with an integrated high-resolution camera, machine learning algorithms and novel 3D reconstruction techniques to provide condensed and high-quality situation information to the end user. The investigation of embedded systems will be completed to select a suitable remote edge node capable of running the unified pipeline in near-real-time with a monocular high-resolution camera. The developed pipeline will be integrated into a maritime situational awareness tool using web interfaces to communicate with the visualization clients.
Duration of the Project
01.01.2023 - 31.12.2025
The resulting technology will be an embedded end-to-end pipeline for near-real-time, high resolution maritime object recognition along with monocular 3D reconstruction. The key application of MAREVIS 3D is suitable for authorities or companies requiring a situational awareness picture for the improvement of the safety and security level of critical maritime infrastructures. End users could include port authorities, emergency services or the autonomous driving/naval industry.