Measurement Noise Recommendation for Efficient Kalman Filtering over a Large Amount of Sensor Data

Area of Science:

• Signal Processing
• Data Analysis
• Machine Learning

Background:

• Real-time sensor data analysis requires effective filtering techniques.
• Kalman filtering is a key technique for correcting sensor data inaccuracies.
• Filtering performance heavily relies on accurate noise parameter estimation.

Purpose of the Study:

• To propose novel methods for recommending measurement noise for Kalman filtering.
• To address the challenge of inaccurate noise parameter estimation impacting filtering accuracy.
• To improve Kalman filtering accuracy by analyzing past sensor data.

Main Methods:

• A transform-based method utilizing wavelet transform for noise variance estimation.
• A learning-based method employing a denoising autoencoder for noise variance estimation.
• Analysis of past sensor data to determine optimal measurement noise variance.

Main Results:

• The proposed methods accurately estimate measurement noise variance.
• Both transform-based and learning-based methods demonstrate superior performance.
• Experimental results confirm improved filtering accuracy over experience-based methods.

Conclusions:

• Novel methods for estimating measurement noise variance in Kalman filtering are presented.
• The proposed techniques offer accurate noise estimation and enhanced filtering performance.
• This research provides a data-driven approach to optimize Kalman filtering for sensor data.