A target model construction algorithm for robust real-time mean-shift tracking.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Mean-shift tracking is a popular real-time object tracking technique.
• Background clutter can significantly interfere with the accuracy and reliability of mean-shift trackers.
• Existing methods struggle to effectively mitigate background noise.

Purpose of the Study:

• To propose a novel indicator function generation method for mean-shift object tracking.
• To reduce the impact of background clutter on tracking performance.
• To enhance the robustness and accuracy of mean-shift based object tracking.

Main Methods:

• Utilizes two 'a priori' knowledge elements inherent to kernel support for target model initialization.
• Employs gradient-based label propagation based on assured background labels to differentiate objects.
• Implements a region growing scheme to select the largest target object near the kernel support center.
• Constructs an exact target model using the grown object region as the indicator function.

Main Results:

• The proposed indicator function generation method effectively differentiates target objects from background clutter.
• The exact target model construction significantly improves tracking robustness.
• Simulation results demonstrate enhanced accuracy in mean-shift object tracking.
• The method successfully mitigates interference from background noise.

Conclusions:

• The novel indicator function generation method provides a robust approach for mean-shift object tracking.
• The proposed technique enhances the reliability and accuracy of real-time object tracking systems.
• This work contributes to more effective object tracking in cluttered environments.