Vehicle Trajectory Prediction Model under Attention Network Based on Goal Areas Estimation

In this study, we explore the utilization of video signals, which encompass both the historical movement paths and the surrounding context of vehicles within a defined setting. The trajectories of vehicles are conceptualized as sequential data points, which can be analyzed to anticipate their subsequent movements. We introduce a novel graph attention network that leverages goal-directed and interactive perceptions to deduce the collective probability distribution of forthcoming trajectories for multiple vehicles. This model presents several significant advantagesFirstly, it efficiently incorporates geographical mapping and real-time traffic data, and it harnesses the power of a Transformer architecture to capture extended temporal sequences. Secondly, leveraging these interdependencies, the model is capable of producing a variety of potential destinations and focal points, enabling the formulation of control trajectories for multiple vehicles. Lastly, by integrating a multimodal forecasting technique grounded in probabilistic distributions, the model also meticulously assesses the likelihood of each trajectory based on a comprehensive consideration of the various potential future paths linked to tshese focal points.