A Two-Layered Approach to Situational Awareness System: Flexible Structure and Modular Algorithms

Abstract

Despite significant progress in deep-learning algorithms for object detection, recognition, and perception, implementing these technologies in real-world environments, such as the ocean, for extended periods remains challenging due to diverse environmental conditions.

In this paper, we propose a situational awareness system for autonomous ships with a two-layer structure. The base fusion layer ensures stable performance for all detected objects at all times, while an additional data fusion and prediction layer manages objects according to various sub-goals, such as size, speed, and behavior, based on specific purposes either continuously or for necessary periods. The proposed structure effectively handles non-mathematical algorithms based on experience with existing manned ships and provides a means to track objects of interest more clearly without affecting the performance of the base fusion layer. This ensures a minimum level of performance through the continuous operation of the base fusion layer. To illustrate the effectiveness of the proposed approach,we present example results for the additional data fusion and prediction layer in tasks such as safe navigation and surveillance/reconnaissance.