A Study of Driver Behavior Inference Model at Time of Lane Change using Bayesian Networks

Tezuka, Shigeki; Soma, Hitoshi; Tanifuji, Katsuya · 2006 · OpenAlex-citations

DOI: 10.1109/icit.2006.372650

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 challenge of developing driver support systems that are tailored to individual driving characteristics. Current systems often rely on average driver behaviors, leading to user discomfort or confusion because they fail to account for individual psychological functions, skills, and habits. To enable personalized support, the authors propose a method to infer specific driver behaviors during lane changes by capturing time-series steering angle data. The study aims to construct a driver model that captures the context dependency of driving actions, specifically comparing a proposed Bayesian Network (BN) approach against traditional Hidden Markov Models (HMM). The researchers constructed a static conditional Gaussian model within a Bayesian Network framework. This model aligns with a hierarchical driver behavior model, where Level 1 represents operational skills (observable steering angles) and Level 2 represents cognitive judgment (hidden states). The BN uses discrete nodes for steering judgment and Gaussian nodes for observed steering angles, allowing for the representation of continuous feature sequences and context dependency. To evaluate the model, experiments were conducted using a driving simulator with nine participants (ages 21–25). Data was collected at 50ms intervals while subjects drove at approximately 60 km/h on a simulated highway. Three driving behaviors were analyzed: lane keeping, normal right lane change, and emergency right lane change (triggered by a sudden obstacle). The BN model was compared against an HMM baseline using leave-one-out cross-validation. The results demonstrated that the BN model significantly outperformed the HMM in inference accuracy and stability. When the similarity (norm) between the subject’s data and the learned data was below a certain threshold, the BN model inferred driver behavior with nearly 100% probability. Specifically, for emergency lane changes with a norm under 10, the inference rate was nearly 100% with zero incorrect inferences. In contrast, the HMM model frequently produced incorrect inferences, particularly in the early stages of steering input. The BN model provided more stable and accurate predictions for both normal and emergency lane changes, effectively reducing the rate of incorrect inference inclusion. The significance of this work lies in demonstrating that Bayesian Networks can effectively model the context dependency of driver behavior using steering angle data alone. The study confirms that BNs are superior to HMMs for this application due to their ability to handle probabilistic parameter variations and generate smooth feature sequences. However, the authors note that for normal lane changes with small steering angles, error rates remained similar to HMMs, suggesting that steering angle data alone is insufficient for fully capturing context dependency. Future work should incorporate additional vehicle data to enhance the model's ability to represent complex driving contexts, thereby enabling more precise and personalized driver support systems.

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.

• lane changing

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).

• Theoretical Contribution: computational model, theory or model