Connected vehicles and new paradigms in the mobility sector have recently pushed forward the need for accurately identifying who is behind the steering wheel on any driving situation. Driver Identification becomes part of a building block in the mobility area to enable new smart services for mobility like dynamic pricing for insurance, customization of driving features and pay-as-you-drive services. However, existing methods for driver identification depend on complex and high sample-rate vehicle data coming either from CAN-bus or from external devices. In this paper we propose to explore the potential for high accuracy driver identification with low-cost and low sample-rate data, mainly GPS trajectories obtained from smartphone. In this approach we contextualize each location data-points in a trip. Then we construct a set of continuous, categorical, and sequential features to represent the whole trip. For driver identification, we propose a Deep Learning~(DL) architecture composed of embedding and recurrent neural networks~(RNNs) layers. Our proposed approach, outperforms, LightGBM and HMM-based baselines. We obtain overall error-rate of 1.9, 3.87, 5.71, 9.57, 13.5% for groups of 5, 10, 20, 50, 100 drivers. The results show outstanding accuracy and performance, enabling a fast and low-complex deployment.