Tracking of multiple targets using online learning for reference model adaptation.

Area of Science:

• Computer Vision
• Artificial Intelligence

Background:

• Existing multiple object tracking methods often require manual initialization.
• Reference model adaptation for single object trackers is an active research area.
• Online learning for appearance changes in tracking is crucial for real-world scenarios.

Purpose of the Study:

• To develop an automated multiple object tracking system capable of online reference model adaptation.
• To enable automatic initialization and termination of individual object tracks.
• To improve tracking robustness by accounting for appearance variations.

Main Methods:

• Utilizes a particle filter for sample distribution propagation during tracking.
• Incorporates online learning to incrementally update object models.
• Employs low-level features (color, size, movement) for automatic track management.
• Discusses the relationship between particle filters and genetic algorithms.

Main Results:

• Demonstrates effective multiple face tracking in meeting scenarios.
• Provides empirical verification of robust reference model learning in diverse scenes.
• Reports on trajectory standard deviation based on learning rate.

Conclusions:

• The proposed approach offers automated and robust multiple object tracking with online adaptation.
• Automatic track initialization/termination enhances usability compared to manual methods.
• The system shows promise for real-world applications requiring adaptive tracking.