A Fusion Algorithm of Object Detection and Tracking for Unmanned Surface Vehicles

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Marine Engineering

Background:

• Unmanned Surface Vehicles (USVs) require accurate target sensing for navigation and decision-making.
• Existing detection and tracking algorithms struggle with small USV targets due to background interference, leading to missed detections and drift.

Purpose of the Study:

• To develop a robust fusion algorithm for enhanced detection and tracking of USV targets.
• To address the limitations of single-frame feature ambiguity and real-time processing constraints.

Main Methods:

• Utilized a cross-stage partial network for high-resolution, deep semantic information extraction.
• Improved network architecture (anchor and convolution) tailored for USV characteristics.
• Integrated correlation filtering for real-time tracking and multi-frame correlation feature extraction.
• Fused single-frame semantic features with multi-frame correlation characteristics to correct drift and reduce missed detections.

Main Results:

• Developed and validated a novel fusion algorithm for USV target detection and tracking.
• Demonstrated superior performance of the fusion algorithm over individual detection and tracking methods.
• Achieved a >10% increase in success rate compared to single-algorithm approaches.

Conclusions:

• The proposed fusion algorithm effectively overcomes the challenges of USV target detection and tracking in complex environments.
• The integration of high-resolution semantic information and correlation filtering significantly enhances accuracy and reliability.
• This approach offers a promising solution for improving USV operational safety and efficiency.