Trajectory Prediction Based on Multi-Layer Fusion of Spatiotemporal Features

Trajectory prediction is a crucial module in autonomous driving technology, enabling intelligent vehicles to choose the optimal path and driving strategy, thus achieving more efficient and safer passage in complex traffic environments. Currently, trajectory prediction algorithms face challenges in adequately utilizing and processing interaction information. In response, this paper proposes a trajectory prediction model based on multi-layer fusion of spatiotemporal features, considering the high interactivity between the target vehicle and the traffic scene. To address the insufficient handling of spatiotemporal interaction information in traditional prediction models, the Long Short-Term Memory (LSTM) network is used to extract temporal features from the trajectory sequences of the target and related vehicles. The temporal information is then embedded into a spatial grid to complete spatial relationship modeling, yielding spatiotemporal features. By designing a multi-layer fusion mechanism using attention, involving interaction fusion and global fusion steps, the model thoroughly exploits the spatiotemporal interaction information of the target and related vehicles' trajectory data, enhancing the prediction model's performance. Finally, simulations are conducted to validate the proposed algorithm; comparisons with other prediction models using the NGSIM dataset and Root Mean Square Error (RMSE) demonstrate that the proposed model has better prediction performance.