The growing penetration of telematics systems and connectivity in vehicles has enabled a large variety of possible value-added services for drivers and service providers. In particular telematics based real-time driver identification is in interest of entities such as insurance companies, car rentals and public transportation fleet managers. We propose a mechanism for driver identification based on driving dynamics signals currently available in production cars. The system collects and filters sensing data in a sliding window iteration, computes statistical and spectral features and finally provides driver identification for each window frame through a classification process. Finally, a decision function takes individual predictions and outputs a single prediction for the ongoing trip. We evaluate the impact of various elements of the process on identification accuracy, including sliding window size, classifier algorithms and feature sets. Results show that complementing gas pedal signal with steering wheel cepstral analysis improves identification accuracy by 22.4%. We also show that Boosting classifiers provide better predictions for our problem and the best results have been achieved using AdaBoost with 95, 89, 82 percent accuracies for 5, 15, 35 drivers respectively. In terms of real-time identification performance, the proposed system is able to correctly identify 75% of the drivers in less than 65 s in a 5 drivers scenario.