Next location prediction plays an essential role in location-based applications. Many works have been employed to predict the next location of an object (e.g. a vehicle), given its historical location records. However, existing methods have not fully addressed the importance of contextual features, such as the short-term traffic flows. In this paper, we propose a deep learning-based model to incorporate contextual features into next location prediction. First, we conduct the similarity mining among candidate locations. Second, we model contextual features among trajectories, including both periodical patterns and dynamic features of trajectories. Third, we adopt both CNN and bidirectional LSTM networks to predict next location in each trajectory with contextual information. Intensive experiments on 197 million vehicle license plate recognition (VLPR) records in Xiamen, China, demonstrate that the proposed method outperforms several existing methods.