A Lane Change Assistance System Based on Prediction of Driver Intention

TL;DR

This study addresses lane-change safety by introducing an intention-aware warning system that predicts a driver's lane-change maneuvers using a personalized fuzzy–random-forest (fuzzy-RF) classifier. The method combines 24 ego- and environment-related features to classify three maneuvers (LK, LCL, LCR) and triggers audible-visual warnings when the predicted action risks a collision, leveraging time-to-collision thresholds ($TTC$) for timing. Two highway scenarios (S1 daytime, S2 foggy daytime) are used to train and test the system, with 44 participants split into experimental and control groups; warnings are delivered via an AR HUD and a 2 s display, along with approval imagery. Results show improved safety for certain lane-change maneuvers but reveal ongoing challenges with trust, workload, and the need to anticipate other vehicles’ behavior; the work highlights interface design, warning timing, and the potential for multi-level warnings to further enhance acceptance and effectiveness. Overall, the paper demonstrates the practical viability of integrating driver intention prediction into lane-change warnings to reduce eminent collisions, while outlining concrete directions for enhancing warning relevance and reliability in real-world driving contexts.

Abstract

Lane change assistance system increase safety by providing warnings and other stability assistance to drivers to avert traffic dangers. In this contribution, lane change intention recognition was performed and applied to generate warnings for drivers to avoid eminent collision. Previous studies have not yet integrated driver's intended lane change actions as an input for determining when to warn drivers about eminent traffic dangers. Thus, if a driver's intended action may result in a collision, the driver should be warned in advance. In this contribution, lane change to left and right and lane keeping intentions were utilized to warn drivers of potential collision using an audio visual interface. The results indicate reduced risk of collision during lane change to left and right except lane keeping maneuvers. Moreover several participant feedback indicate an increased need for improved warnings by additional situational analysis that anticipate other vehicle behaviors such as intended lane changes.

Figures (14)

• Figure 1: Online driver assistance system
• Figure 2: Intention-based lane change assistance (cf. DengSTS:2020)
• Figure 3: Trapezoidal membership function defined by core ($a_2$, $a_3$) and support ($a_1$, $a_4$) parameters
• Figure 4: Intention to eventual lane change DengSTS:2020DengWangSoeJ1819
• Figure 5: Offline training and on-line fuzzy-RF model adapted from DengSTS:2020