Intelligent Vehicle Moving Trajectory Prediction Based on Residual Attention Network

Skilled drivers have the driving behavioral characteristic of pre-sighted following, and similarly intelligent vehicles need accurate prediction future trajectories. The LSTM (Long Short-Term Memory) is a common model trajectory prediction. existing models pay less attention to interactions between target surrounding vehicles. Furthermore, impacts on trajectories vehicle also barely been focus current models. On these bases, Residual Attention-based Long Memory (RA-LSTM) was proposed, an interaction tensor based surroundings at predictive moments constructed weight coefficients for relative were calculated re-programmed in this study. proposed RA-LSTM can implicitly represent different degrees influence vehicle; probability distributions coordinates predicted extracted features. tested verified multiple scenarios by using NGSIM (next generation simulation) public dataset, results showed that accuracy significantly improved compared with