The accuracy forecast of surrounding vehicles’ trajectories is of great significance for an autonomous vehicle when making safe and reasonable driving decisions. The uncertainty of driver behavior is a major challenge to accurately predict the trajectory. This paper proposes a trajectory prediction algorithm that combines convolutional social pooling and driver's driving intention. The algorithm uses Long-Short Term Memory (LSTM) as the encoder and decoder of vehicles' trajectory, then uses the integrated convolutional social pooling as the trajectory prediction module. Firstly, the conventional convolution social pooling model is used to predict the driving intention of the vehicle in the surrounding area of the target vehicle. Then the driving intention of the surrounding vehicle is merged with the convolutional social pooling model to obtain the integrated convolutional social pooling model. The trajectory prediction of the target vehicle is used to obtain a better trajectory prediction effect. We will use the real vehicle trajectory dataset to train the model. Besides, the model is compared with the existing algorithms to verify the contribution of the driver’s intention, which is used to enhance the accuracy of the trajectory. Finally, we use the root of the mean squared error (RMSE) and negative log-likelihood (NLL) as evaluation indicators. The proposed algorithm prediction accuracy is estimated, and it is expected to obtain a higher trajectory prediction accuracy than the previous algorithms. Keywords: trajectory prediction; convolutional social pooling; driving intention; LTSM

CICTP