Multi-parameter prediction of drivers' lane-changing behaviour with neural network model.

Area of Science:

• Automotive Engineering
• Human-Computer Interaction
• Artificial Intelligence

Background:

• Active safety systems require accurate driving behavior prediction for enhanced driver safety.
• Predicting lane-changing behavior is crucial for developing effective lane-changing assistance systems.
• Existing methods, like turn signals, offer limited predictive capabilities.

Purpose of the Study:

• To develop a predictive model for driver lane-changing behavior using naturalistic driving data.
• To determine a reliable time window for detecting lane-changing intent.
• To enhance the predictive accuracy and lead time for lane-changing assistance systems.

Main Methods:

• Utilized naturalistic on-road driving experiment data.
• Extracted visual characteristics from rearview mirror usage to determine lane-changing intent.
• Constructed a prediction index system incorporating visual search, vehicle operation, motion states, and driving conditions.
• Developed a back-propagation neural network model for behavior prediction.

Main Results:

• Identified a lane-change intent time window of approximately 5 seconds.
• The developed model accurately predicts lane-changing behavior at least 1.5 seconds in advance.
• Demonstrated superior accuracy and time-series characteristics compared to using turn signals for prediction.

Conclusions:

• The proposed model offers a significant advancement in predicting driver lane-changing behavior.
• This predictive capability can be integrated into lane-changing assistance systems to improve road safety.
• The model's reliance on visual cues and driving parameters provides a more robust prediction than turn signals alone.