Signal analysis based on recurrence plots for effective driver behavior  detection

Danut-Dragos Damian; Simona Moldovanu; Felicia-Anisoara Damian; Michael Frătița; Luminita Moraru

doi:10.35219/ann-ugal-math-phys-mec.2025.1.02

Danut-Dragos Damian The Modelling & Simulation Laboratory, Dunarea de Jos University of Galati, 47 Domneasca Street
Simona Moldovanu The Modelling & Simulation Laboratory, Dunarea de Jos University of Galati, 47 Domneasca Street, Department of Computer Science and Information Technology, Faculty of Automation, Computers, Electrical Engineering and Electronics, Dunarea de Jos University of Galati, 47 Domneasca Street
Felicia-Anisoara Damian Emil Racovita High School of Galati, 12-14, Regiment 11 Siret Street, 800332, Galati, Romania
Michael Frătița Department of Thermal Systems and Automotives, Faculty of Engineering, Dunărea de Jos University of Galați, 800008 Galați, Romania
Luminita Moraru .The Modelling & Simulation Laboratory, Dunarea de Jos University of Galati, 47 Domneasca Street, 800008 Galati, Romania, Department of Physics, School of Science and Technology, Sefako Makgatho Health Sciences University, Pretoria, South Africa

DOI: https://doi.org/10.35219/ann-ugal-math-phys-mec.2025.1.02

Keywords: Accelerometer, Gyroscope, Sliding Window, Recurrence Plots, Recurrence Matrix Recurrence Quantification Analysis, Recurrence Rate (RR), Determinism (DET, Laminarity (LAM)

Abstract

This paper employs recurrence plots (RPs) generated from both accelerometer and gyroscope data to analyze driver behavior. It integrates visualization with quantitative analysis by extracting key recurrence quantification measures, such as the Recurrence Rate (RR), Determinism (DET), and Laminarity (LAM), to effectively characterize the dynamics of the time-series signals. The accelerometer and gyroscope data are collected along three axes. These recurrence-based features facilitate the discrimination between stable, controlled driving dynamics and irregular, non-deterministic driving behavior. The RPs are generated using a sliding time window.
The epoch length is set to 3000 samples with a window overlap of 80%. The results demonstrate that changes in driving conditions significantly altered the structure of the recurrence plots, with corresponding variations in the recurrence quantification RR, DET, and LAM metrics highlighting the sensitivity of these parameters to behavioral dynamics.

Downloads

Download data is not yet available.