The Study of Trajectory Prediction in V2X Collision Warning Based on Pruned SR-LSTM

With the development of intelligent transportation systems and autonomous driving technology, V2X (Vehicle-to-Everything) technology enhances traffic efficiency and safety through information exchange between vehicles and other road users. In V2X application scenarios, vehicle trajectory prediction is the core of collision warning systems. However, current communication systems face multiple challenges such as transmission delays, data packet loss, and limited computing power at the vehicle end, making it difficult to ensure real-time performance and accuracy. Therefore, how to achieve high-precision trajectory prediction in a constrained computing environment is an urgent issue to be resolved. This paper proposes an optimized model based on pruned SR-LSTM, significantly reducing the model's computational complexity to meet the deployment requirements at the vehicle end. During the prediction phase, this pruned SR-LSTM model improves computation speed by reducing unnecessary connections, ensuring the accuracy and real-time nature of trajectory predictions, thereby enhancing the responsiveness of the collision warning system. By combining time steps and the predicted trajectory point sequences, the system calculates the time to collision (TTC) of vehicles within a reasonable safety distance and outputs warning signals in real-time, verifying the effectiveness of the pruned LSTM in handling dynamic and complex scenarios in a V2X environment.