New Deep Learning Framework Detects Aggressive Driving from Vehicle Signals
Researchers propose CBANet, a compact attention-based CNN-BiLSTM network for detecting aggressive driving events from multivariate vehicle dynamics signals. The framework addresses data imbalance and driver variability by constructing engineered dynamic features for steering, acceleration, and braking, combined with SMOTE-based oversampling and class-weighted loss. The study aims to improve road safety through more accurate detection of risky behaviors.
Key facts
- CBANet uses CNN-BiLSTM with attention mechanism
- Engineered features capture steering, acceleration, braking
- SMOTE oversampling and class-weighted loss address data imbalance
- Aims to detect aggressive driving from vehicle sensor data
- Published on arXiv with ID 2605.23471
- Aggressive driving is a major cause of traffic accidents
- Deep learning methods often limited by data imbalance and driver variability
- Proposed framework enhances performance in real-world conditions
Entities
Institutions
- arXiv