Robust Visual Tracking via Multitask Sparse Correlation Filters Learning.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Visual tracking is crucial for many applications.
• Existing methods struggle with complex appearance changes and background clutter.
• Integrating multiple feature types and sparse learning can improve robustness.

Purpose of the Study:

• To propose a novel multitask sparse correlation filters (MTSCF) model for enhanced visual tracking.
• To leverage multitask learning for improved feature utilization and filter complementarity.
• To implement dynamic spatial feature selection for better target discrimination.

Main Methods:

• Introduced multitask sparse learning into the correlation filters (CFs) framework.
• Exploited interdependencies among Histogram of Oriented Gradient (HOG), color names, and CNN features.
• Utilized an l2,1 regularization term for sparse learning and alternating direction method of multipliers for optimization.

Main Results:

• The MTSCF model effectively integrates diverse visual features.
• Dynamic spatial feature selection improved target-background distinction.
• Achieved competitive tracking performance on multiple benchmark datasets compared to state-of-the-art trackers.

Conclusions:

• The proposed MTSCF model offers a robust approach to visual tracking.
• Multitask sparse learning enhances the utilization of feature strengths.
• The method demonstrates superior performance by effectively modeling target appearance and discriminating from background.