Lane Change Scene Recognition Method for Natural Driving Datasets

Lane change recognition is essential for intelligent transportation systems and autonomous driving. This study presents a novel method leveraging scene data from natural driving datasets to enhance lane change recognition. The approach integrates a 3D Convolutional Neural Network (3DCNN) with an attention mechanism and a Random Forest (RF) algorithm, using Extreme Value Theory (EVT) to improve detection of unknown traffic scenarios. By analyzing decision tree voting patterns for known classes and applying EVT for anomaly detection, the method identifies both known and novel lane change instances. Experiments on the HighD and OpenTraffic datasets show that this method outperforms current state-of-the-art techniques. Additionally, self-supervised learning and data-adaptive similarity measures are employed to analyze lane change patterns, offering insights for accident prevention.