Correlation-Based Tracker-Level Fusion for Robust Visual Tracking.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Current visual object tracking algorithms struggle with simultaneous challenging scenarios.
• Online tracking-by-detection methods face drift due to target detection and model update issues.

Purpose of the Study:

• To propose a novel, robust visual object tracking algorithm.
• To mitigate the drift phenomenon in object tracking.
• To enhance tracking performance by combining complementary strategies.

Main Methods:

• Fusion of frame-level detection (tracking, learning, & detection) with systematic model updates (Kernelized Correlation Filter).
• Mitigation of drift by prioritizing model updates from spatial neighborhood detections.
• Utilizing a conservative correspondence measure for strategic model updates.

Main Results:

• The proposed algorithm demonstrates superior robustness and success rates compared to state-of-the-art trackers.
• Outperforms individual trackers by leveraging their complementary strengths.
• Extensive evaluations on ALOV300++, Visual Tracker Benchmark, and Visual Object Tracking datasets confirm performance.

Conclusions:

• The fused approach effectively addresses limitations of existing visual object tracking methods.
• Achieves significant improvements in robustness and accuracy for challenging tracking scenarios.
• Represents a promising advancement in real-time visual object tracking.