Underwater Target Tracking Using Forward-Looking Sonar for Autonomous Underwater Vehicles.

Area of Science:

• Robotics
• Underwater Sensing
• Signal Processing

Background:

• Autonomous underwater vehicles (AUVs) require reliable underwater target positioning.
• Forward-looking sonar (FLS) offers effective long-range tracking despite camera limitations.
• Cluttered underwater environments pose challenges for persistent multiple-target tracking.

Purpose of the Study:

• To develop an online processing framework for underwater target tracking using FLS images.
• To present a novel Gaussian particle filter (GPF) based approach for multiple-target tracking.
• To improve tracking performance by integrating adaptive feature fusion.

Main Methods:

• Modified median filtering and region-growing segmentation for acoustic-vision image processing.
• Generalized regression neural network (GRNN) for evaluating target region features.
• Adaptive fusion strategy to integrate feature cues into the observation model for GPF.

Main Results:

• The proposed method demonstrated feasibility and effectiveness in tracking targets in complex underwater environments.
• Sea trials on a real acoustic-vision AUV platform validated the tracking approach.
• Improved image processing and feature evaluation enhanced tracking performance.

Conclusions:

• The developed framework and GPF-based tracking approach are effective for underwater target localization.
• The adaptive fusion strategy successfully integrates feature cues for robust tracking.
• The method provides a viable solution for persistent multiple-target tracking in challenging conditions.