Ensemble Learning-Based Multi-Cues Fusion Object Tracking in Complex Surveillance Environment.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Current kernelized correlation filter (KCF) trackers often rely on single object features, leading to instability in complex video scenarios.
• Tracking instability arises from challenges like occlusion, illumination changes, and scale variations inherent in real-world videos.

Purpose of the Study:

• To develop an ensemble learning-based multi-cues fusion object tracking method to address the limitations of single-feature KCF trackers.
• To enhance tracking robustness and accuracy by integrating multiple object features and employing adaptive strategies.

Main Methods:

• Utilized ensemble learning to train multiple KCF trackers with diverse features, optimizing tracking parameters.
• Implemented adaptive weighted fusion of response findings using peak side lobe ratio and inter-frame response consistency.
• Incorporated a Bayesian estimation model with a scale pyramid for adaptive object scale determination.
• Employed tracking confidence for adaptive model updates to prevent performance degradation.

Main Results:

• The proposed algorithm effectively mitigates interference from various elements in complex videos.
• Demonstrated superior overall tracking performance compared to existing benchmark algorithms on diverse video datasets.
• The adaptive scale estimation improved the tracker's adaptability to object size changes.

Conclusions:

• The ensemble learning-based multi-cues fusion approach significantly improves KCF tracker stability and accuracy.
• The adaptive fusion and scale estimation mechanisms contribute to robust object tracking in challenging conditions.
• The developed method offers a promising solution for reliable object tracking applications.