Abstract
Tracking maneuvering underwater targets is challenging due to the inherent lag in standard filtering techniques during sudden motion changes. This paper proposes a Feature-Aided Multiple Model (FAMM) algorithm that leverages target acoustic intensity as a real-time feature to improve model switching. By integrating signal features with traditional tracking data, FAMM achieves faster response times to maneuvers and reduces tracking errors by up to 18% compared to state-of-the-art adaptive methods.
Introduction
In underwater acoustics, tracking non-cooperative targets like submarines or autonomous underwater vehicles (AUVs) often relies on passive sensor networks. The Interacting Multiple Model (IMM) algorithm is the standard approach, running multiple filters in parallel (e.g., Constant Velocity, Constant Acceleration) and blending their outputs. However, IMM suffers from a "lag" problem; it typically switches models only after observing a significant increase in tracking error. This delay leads to accuracy loss during critical transition periods. This research introduces an external "trigger"—acoustic intensity—to signal maneuvers as they occur, minimizing this delay.
Methods: Feature-Aided Tracking
The proposed FAMM method treats the target's acoustic signal as a primary feature for model selection rather than just a localization data point.
- Acoustic Intensity as a Motion Proxy: Changes in received sound intensity ($\Delta P_k$) are physically linked to target acceleration. Abrupt changes in intensity often correlate with transitions to a Constant Acceleration (CA) state.
- Two-Stage Switching Strategy: The system uses thresholds to toggle between motion models. Low intensity changes trigger switches between Constant Velocity (CV) and Constant Turn (CT), while rapid spikes prioritize the CA model.
- Multi-Node Information Fusion: A central node fuses data from multiple passive hydrophones, weighting nodes closer to the target more heavily due to their superior signal-to-noise ratio.
- Hybrid Probability Estimation: Model selection is determined by a weighted combination of feature-based probability (from intensity) and error-based probability (from the Kalman filter's likelihood function).
Experimental Setup
The algorithm was validated using Monte Carlo simulations in a 3D oceanic environment with four passive sensor nodes. The target followed a complex 150-second trajectory involving multiple transitions between CV, CA, and CT models.
Key Results
The FAMM algorithm was benchmarked against standard IMM and two variants of Adaptive IMM (AIMM).
- Superior Accuracy: FAMM consistently achieved the lowest Root Mean Square Error (RMSE) across the entire trajectory.
- Rapid Maneuver Response: During model transitions, FAMM reduced tracking errors by 18.07% compared to the best existing adaptive method (AIMM2).
- Consistent Performance: Even during stable, non-maneuvering phases, FAMM improved accuracy by 3% to 4% over AIMM2.
Discussion and Future Work
The core innovation of FAMM is its ability to "anticipate" maneuvers by recognizing that acoustic signature shifts often precede velocity changes. By breaking the reliance on fixed Markov transition probabilities, the algorithm becomes truly adaptive.
However, the method currently requires manual tuning of intensity thresholds, which may vary by environment. Future work will explore more complex acoustic features and automated parameter optimization to handle extreme underwater conditions, such as heavy biological noise or adverse weather, which might otherwise trigger false positives in model switching.
Reference: Y. Tian, M. Liu, S. Zhang, R. Zheng, and S. Dong, “A feature-aided multiple model algorithm for maneuvering target tracking,” IEEE/CAA J. Autom. Sinica, vol. 11, no. 2, pp. 566–568, Feb. 2024.
Summary updated