Real-Time Robust Tracking for Motion Blur and Fast Motion via Correlation Filters

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Visual tracking is crucial for intelligent systems.
• Correlation Filter (CF) methods offer good performance but struggle with scale variation, motion blur, and fast motion.
• Existing trackers require improvements for real-world dynamic environments.

Purpose of the Study:

• To develop a real-time, robust visual tracking scheme.
• To significantly improve performance on motion blur and fast motion challenges.
• To address target scale variation estimation effectively.

Main Methods:

• Proposed a novel scheme based on Kernelized Correlation Filter (KCF).
• Fused KCF with Spatio-Temporal Correlation (STC) trackers for scale variation estimation.
• Utilized a point sharpness function to evaluate target motion state.
• Developed an efficient scheme to handle motion and scale variations.

Main Results:

• The algorithm significantly improves performance on motion blur and fast motion.
• Successfully addresses scale variation estimation in various scenarios.
• Maintains the desirable properties of KCF while handling special motion conditions.
• Experimental results on VOT datasets demonstrate superior performance compared to top-rank trackers.

Conclusions:

• The proposed method offers a robust and efficient solution for visual tracking.
• It effectively overcomes limitations of traditional CF methods in dynamic scenarios.
• The fusion approach and motion evaluation enhance tracking accuracy and speed.