Robust Interacting Multiple Model Filter Based on Student's t-Distribution for Heavy-Tailed Measurement Noises

Area of Science:

• Signal Processing
• Statistical Inference
• Target Tracking

Background:

• Traditional interacting multiple model (IMM) filters struggle with heavy-tailed measurement noises from outliers.
• This performance degradation significantly impacts maneuvering target tracking applications.

Purpose of the Study:

• To develop a robust IMM filter capable of handling heavy-tailed measurement noises.
• To enhance the accuracy of maneuvering target state and mode probability estimation.

Main Methods:

• Proposed a robust IMM filter employing Student's t-distribution for measurement noises.
• Utilized gamma and inverse Wishart distributions for degrees of freedom and scale matrix.
• Employed variational Bayesian approach for posterior distribution estimation and joint parameter estimation.
• Incorporated unscented transform for nonlinear estimation problems.

Main Results:

• The proposed filter demonstrated improved estimation accuracy for target state compared to existing filters.
• Accurate estimation of mode probability was also achieved under heavy-tailed noises.
• The filter effectively mitigated the negative impact of outliers in measurement data.

Conclusions:

• The robust IMM filter utilizing Student's t-distribution is effective for maneuvering target tracking in heavy-tailed noise environments.
• This approach offers a significant advancement in robust filtering techniques.
• The proposed method provides more reliable target state and mode probability estimates.