A Student's t Mixture Probability Hypothesis Density Filter for Multi-Target Tracking with Outliers

Area of Science:

• Robotics and Automation
• Signal Processing
• Statistical Inference

Background:

• Multi-target tracking is crucial for various applications.
• Probability Hypothesis Density (PHD) filters are widely used but sensitive to outliers.
• Process and measurement noise outliers degrade PHD filter performance.

Purpose of the Study:

• To develop a robust Probability Hypothesis Density (PHD) filter resistant to outliers.
• To enhance multi-target tracking accuracy in noisy environments.

Main Methods:

• Proposed a novel Student's t mixture PHD (STM-PHD) filter.
• Modeled heavy-tailed process and measurement noise using Student's t distributions.
• Approximated multi-target intensity as a mixture of Student's t components.
• Derived a closed PHD recursion based on Student's t approximation.

Main Results:

• The STM-PHD filter effectively models heavy-tailed noise.
• The filter demonstrates resilience against outliers in both process and measurement noise.
• Simulation results confirm improved tracking accuracy compared to standard PHD filters.

Conclusions:

• The STM-PHD filter offers a robust solution for multi-target tracking with outliers.
• This method leverages Student's t distribution properties to handle heavy-tailed noise.
• The proposed filter maintains good tracking performance even with simultaneous process and measurement outliers.