Scale-Aware Tracking Method with Appearance Feature Filtering and Inter-Frame Continuity.

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Visual object tracking is crucial in computer vision, but scale variation poses a significant challenge.
• Discriminative Correlation Filter (DCF) based trackers struggle with scale changes, especially under computational constraints.
• Existing multi-scale feature methods for scale estimation are computationally intensive, hindering real-time performance.

Purpose of the Study:

• To develop a practical and efficient solution for visual object tracking that addresses scale variation.
• To create a plug-in module compatible with any DCF-based tracker, enhancing robustness without multi-scale features.
• To reduce the computational cost of DCF trackers while maintaining high accuracy and real-time performance.

Main Methods:

• Utilized color name (CN) features and salient features to reduce target appearance model dimensionality.
• Estimated target scale using a Gaussian distribution model, incorporating global and local scale consistency assumptions.
• Fused proposed scale estimation with DCF tracking results for updated target position and scale.

Main Results:

• Achieved competitive accuracy and robustness on the Temple Color 128 benchmark dataset.
• Significantly reduced computational cost compared to existing methods.
• Demonstrated the effectiveness of the proposed method as a plug-in module for DCF trackers.

Conclusions:

• The proposed method offers an efficient and practical solution for visual object tracking with scale variation.
• The approach enhances DCF tracker performance by effectively handling scale changes without multi-scale features.
• This method is suitable for real-time applications with limited computational resources.