Lane Change Detection Using Naturalistic Driving Data

Lane Change (LC) detection is the foundation of LC studies using real-world data. Most current studies use rule-based methods for LC detection from experimental data. In this study, we propose a learning-based method to detect LC using large-scale naturalistic driving data. The dataset is analyzed using big data analytics method, and the potential LC maneuvers are extracted. The LC detection is reformulated as a one-class classification problem, and an autoencoder-based anomaly detection method is developed to solve it. The proposed method is robust to data noises and can achieve better detection performance than the one-class Support Vector Machine (SVM). This work lays the groundwork for future LC studies, such as driving behaviour modeling and traffic safety solutions.