A data-driven approach to traffic vehicle intent recognition and trajectory prediction

Automated driving systems require accurate and real-time identification and prediction of the motion state of surrounding vehicles to avoid potential accidents and enhance driving safety in complex environments. However, the motion of surrounding traffic vehicles exhibits unpredictable behaviors, such as sudden lane changes and emergency braking, significantly increasing the difficulty of motion prediction. Existing motion prediction methods rely on physics-based and statistical probability modeling, facing challenges such as spatiotemporal solid interactions and long-tail distributions. Therefore, this paper proposes a data-driven method for recognizing traffic vehicle intent and trajectory prediction, mining data to train spatiotemporal change models, and extracting traffic vehicles movement patterns. The paper first utilizes a fuzzy inference system and an extended short-term memory network to recognize spatiotemporal sequences of driving behavior, establishing a long-term semantic intent model adapted to the current scenario. Subsequently, we develop an attention-based short-term trajectory prediction encoder-decoder based on the recognized behavioral intent, outputs the probability distribution of the vehicle's future trajectory, achieving real-time prediction of traffic vehicle trajectory. Finally, the effectiveness of the prediction model is validated through database and system integration testing.