Iterative Graph Seeking for Object Tracking.

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Existing object tracking methods struggle with large deformations and occlusions.
• Graph-based models often use sequential part selection, matching, and estimation, leading to inaccuracies.
• Lack of global constraints hinders local part selection and matching effectiveness.

Purpose of the Study:

• To propose a novel object tracking method addressing limitations of current approaches.
• To integrate target part selection, matching, and state estimation into a unified framework.
• To incorporate global structural information for improved local part variation analysis.

Main Methods:

• Developed an iterative graph-seeking approach for object tracking.
• Formulated a unified energy minimization framework for integrated tracking components.
• Employed an alternative iteration scheme to minimize the energy function and identify optimal target graphs.

Main Results:

• The proposed method demonstrates improved performance and robustness.
• Experimental results on VOT2015, OTB2013, and OTB2015 benchmarks validate the approach.
• The integration of global constraints enhances tracking accuracy.

Conclusions:

• The iterative graph-seeking method offers a more effective solution for object tracking challenges.
• The unified energy minimization framework successfully integrates key tracking processes.
• The approach provides superior robustness against large deformations and severe occlusions.