Enhanced Vehicle Trajectory Extraction Using Bi LSTM Network and Deep Learning for Intelligent Transportation Systems

Abstract

Accurate vehicle trajectory extraction is essential for intelligent transportation systems, impacting traffic management, safety, and behavior modeling. Traditional methods face challenges in complex environments and adverse weather conditions. This paper proposes a novel vehicle detection and tracking approach using a Bi LSTM network and deep learning techniques. The method involves extracting vehicle cut-in scene fragments from natural driving data, generating a dataset through a sliding time window method, and using a modified Bi LSTM model to predict entry trajectories in a hybrid driving environment. The approach effectively reduces gradient disappearance and network degradation, improving prediction accuracy and robustness. Experimental results demonstrate significant advancements in vehicle trajectory prediction, highlighting the method's high application value.