Fast and Robust Object Tracking via Probability Continuous Outlier Model.

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Linear representation models are susceptible to outliers in visual tracking.
• Existing methods struggle with continuous outliers, impacting tracking accuracy and robustness.

Purpose of the Study:

• To propose a novel visual tracking method robust to continuous outliers.
• To introduce a Probability Continuous Outlier Model (PCOM) for enhanced linear representation.
• To improve the accuracy and speed of visual tracking systems.

Main Methods:

• Developed a Probability Continuous Outlier Model (PCOM) incorporating Markov Random Fields for spatial consistency.
• Derived an objective function solvable via outlier-free least squares and max-flow/min-cut.
• Designed an observation likelihood function and update scheme for visual tracking.

Main Results:

• The PCOM method effectively models and handles continuous outliers in linear representations.
• Iterative solutions using max-flow/min-cut demonstrate computational efficiency.
• Evaluations show significant improvements in both tracking accuracy and speed.

Conclusions:

• The proposed PCOM-based visual tracking method offers superior performance compared to existing approaches.
• The novel outlier model enhances robustness in challenging visual tracking scenarios.
• The method provides a computationally efficient and accurate solution for real-time visual tracking.