Recurrent and Mixture Density Architecture-based Driver Pattern Recognition

Within the framework of Intelligent Transportation Systems (ITS), recognizing driving patterns is becoming of crucial importance, given the alarming rise in accident rates resulting from imprudent conduct behind the wheel. As vehicles become more advanced, the ability to identify and deal with risky behavior have become a key factor in improving road safety. In this work, we propose an approach that distinguishes between safe and unsafe driving, enabling self-control and adaptation of the driver's actions if necessary. The main idea of our solution is the use of LSTM-based Recurrent Neural Networks (RNN) combined with the Gaussian Mixture Method (GMM), which enables classification of driving styles into three distinct categories. The performance of the proposed approach has been evaluated using important evaluation metrics. The results show that the proposal offers encouraging performance, achieving an accuracy of over 93% for different driving styles.