A Cognitive Control-Inspired Approach to Object Tracking.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Automatic understanding of dynamic scenes requires accurate target state definition.
• Far object tracking faces challenges due to abrupt target size changes, impacting data association and position estimation.
• Adaptability and self-awareness are crucial for robust tracking modules.

Purpose of the Study:

• To propose a general cognitive dynamic systems (CDS)-based approach for adaptive far object tracking.
• To demonstrate how a CDS-inspired design can enhance the self-adaptability of Bayesian trackers.
• To address challenges posed by target size variations in dynamic scenes.

Main Methods:

• Utilizing the paradigm of cognitive dynamic systems (CDS) to design a continuously learning cognitive module.
• Applying CDS theory components to create a module capable of learning behavioral rules through environmental interaction.
• Developing a CDS-inspired Bayesian tracker for fusing heterogeneous object features.

Main Results:

• The proposed CDS-based framework enables self-adaptability in Bayesian trackers.
• The approach effectively overcomes issues caused by abrupt target size changes.
• Experimental results on infrared sequences show superior performance compared to existing far object tracking methods.

Conclusions:

• A CDS-based approach provides a robust framework for adaptive far object tracking.
• The self-adaptive nature of the proposed tracker enhances performance in dynamic environments with varying target sizes.
• This work offers a promising direction for improving automatic understanding of complex dynamic scenes.