Probabilistic detection of impacts using the PFEEL algorithm with a Gaussian Process Regression Model.

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Human activity identification is crucial for security, event detection, intelligent environments, and health monitoring.
• Existing methods often use wave propagation or structural dynamics, facing challenges like multi-path fading.
• Force-based methods, like PFEEL, offer advantages by estimating impact forces and locations with uncertainty.

Purpose of the Study:

• To present a novel data-driven implementation of the PFEEL algorithm.
• To leverage Gaussian Process Regression (GPR) for enhanced force and event localization accuracy.
• To evaluate the performance of the new PFEEL implementation using experimental impact data.

Main Methods:

• Developed a new PFEEL implementation utilizing Gaussian Process Regression (GPR).
• Collected experimental data from an aluminum plate subjected to 81 distinct impact points (5 cm separation).
• Analyzed localization accuracy by comparing estimated impact locations with actual positions at various probability levels.

Main Results:

• The data-driven PFEEL approach demonstrated effective impact force and event localization.
• Results quantified the localization area relative to actual impact points across different probability thresholds.
• The GPR-based PFEEL provides a measure of uncertainty, aiding in precision determination for practical applications.

Conclusions:

• The GPR-based PFEEL offers a robust and accurate method for identifying human activity through impact analysis.
• This approach addresses limitations of traditional wave propagation methods.
• The quantified uncertainty in localization aids analysts in selecting appropriate precision levels for specific PFEEL applications.