Robust Visual Ship Tracking with an Ensemble Framework via Multi-View Learning and Wavelet Filter

Area of Science:

• Computer Vision
• Maritime Technology
• Signal Processing

Background:

• Maritime surveillance videos are vital for traffic analysis and safety.
• Conventional ship tracking methods struggle with visual feature reliance and tracking oscillations.

Purpose of the Study:

• To develop an ensemble ship tracking framework to enhance maritime situational awareness.
• To address limitations of conventional methods in handling tracking oscillations.

Main Methods:

• Particle filter for ship candidate sampling.
• Multi-view learning algorithm extracting contour features (LoG, LBP, Gabor, HOG, Canny) and learning intrinsic ship features.
• Wavelet filter for data quality control and correction of abnormal position oscillations.

Main Results:

• The proposed framework effectively tracks ships in maritime surveillance videos.
• The multi-view learning and wavelet filter integration improves tracking accuracy by correcting oscillations.
• Demonstrated performance on typical maritime traffic scenarios.

Conclusions:

• The ensemble ship tracking framework offers a robust solution for maritime surveillance.
• The integration of multi-view learning and wavelet filtering significantly enhances tracking precision.
• This method contributes to improved automated maritime situational awareness and safety.