Semantic underwater world modeling by using Probabilistic Particle Filter Anchoring

Creating an accurate world model of the scenario where an Autonomous Underwater Vehicle (AUV) is navigating can be considered a crucial stage for understanding the surrounding environment. As a result, the targets detected by a cutting-edge Automatic Target Recognition (ATR) architecture alongside their localized positions, must be handled, selected and filtered to get a symbolic representation of the underwater context. Even though the specific World Modeling (WM) architecture may vary, current WM methodologies usually rely on the 3D localization knowledge of the detected target by introducing a not-negligible constraint. Motivated by the aforementioned considerations, a novel Probabilistic Particle Filter Anchoring (PPFA) approach has been developed. Starting from ATR 2D results, the PPFA methodology aims at providing a semantic 3D representation of the subsea environment by merging the upsides of both Data Association (DA) and object tracking, handled by a custom designed Particle Filter (PF) with resampling.