Cepstral Analysis of Driving Behavioral Signals for Driver Identification

Miyajima, C.; Nishiwaki, Y.; Ozawa, K.; Wakita, T.; Itou, K.; Takeda, K. · 2006 · OpenAlex-citations

DOI: 10.1109/icassp.2006.1661427

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the problem of driver identification by analyzing driving behavioral signals, specifically gas and brake pedal operations. Motivated by the need for intelligent transportation systems that can provide personalized assistance, the authors propose a method to extract individual driver characteristics through spectral analysis. Unlike previous approaches that modeled raw signal distributions, this study applies cepstral analysis—commonly used in speech recognition—to pedal operation signals to capture the spectral envelope of driving behaviors. The methodology involves collecting driving data from two sources: a driving simulator and a real vehicle (Toyota Regius). In the simulator, twelve participants drove for 20 minutes each, following a lead vehicle on a simulated expressway. In the real-world experiment, data was collected from 276 drivers on city roads. Signals including pedal force, velocity, and steering angle were sampled at 100 Hz. The authors extracted cepstral coefficients from the gas and brake pedal signals, representing the spectral envelope, and modeled these features using Gaussian Mixture Models (GMMs). The models incorporated both static cepstral coefficients and dynamic features (delta coefficients). Driver identification was performed by calculating the weighted log-likelihood of the observed signals against the trained GMMs for each driver. The experimental results demonstrate that cepstral features significantly outperform raw signal features. In the driving simulator, the proposed model achieved an identification rate of 89.6%, compared to 72.9% for the conventional model using raw signals, representing a 61% error reduction. In the real vehicle experiments with 276 drivers, the cepstral-based model achieved a 76.8% identification rate, while the raw signal model achieved only 47.5%, resulting in a 55% error reduction. The study also found that gas pedal signals provided better identification performance than brake pedal signals due to more frequent usage. Furthermore, the cepstral model maintained superior performance even with shorter test durations; a one-minute test using cepstral features yielded a 59.5% identification rate, which exceeded the performance of raw signals using three minutes of data. The significance of this work lies in demonstrating that the spectral envelope of pedal operations contains distinct, stable characteristics unique to individual drivers. By adapting speech processing techniques to driving signals, the authors provide a more robust method for driver identification. This approach improves accuracy and reduces the amount of data required for identification, which has implications for developing personalized intelligent driving assistance systems and enhancing vehicle security through biometric-like driver verification.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: tool software
• Theoretical Contribution: computational model