PROJECT TITLE :

Driver Identification and Verification From Smartphone Accelerometers Using Deep Neural Networks

ABSTRACT:

This paper discusses the identification and verification of drivers through the application of Deep Learning (DL) to the tri-axial accelerometer signals obtained from drivers' smartphones. The ResNet-50 network, followed by two stacked gated recurrent units, is a component of the driver identification architecture that has been proposed (SGRUs). ResNet is able to extract rich features from accelerometers thanks to shortcut connections, and its GRU layers model the dynamics of drivers' behavior. This is all made possible by the deep layer model that ResNet provides. Two different strategies for mapping 1D accelerometer signals into 2D images have been tested and evaluated using ResNet-50, which was pre-trained on image classification. For the purpose of driver verification, Siamese Neural Networks and Triplet Loss Training have been suggested as viable options. In contrast, the Siamese architecture relies on the same ResNet-50 + GRU model of driver identification, whereas the Triplet loss necessitated the acquisition of embeddings at the journey level. The results of experiments have been obtained for a dataset consisting of 25 drivers who traveled a total of 20,025 miles, with each driver covering more than 800 kilometers. Driver verification was able to achieve a score of 74.09% on the F1 scale, while driver identification achieved top-1 and top-5 accuracies of 71.89% and 92.02%, respectively. In general, state-of-the-art research has been conducted using much smaller databases (in many cases, based only on predefined routes), and it has relied on information sources other than accelerometers, such as gyroscopes, magnetometers, and GPS. These results are competitive with those of that research. Because of this, we think that the proposed DL architectures are appropriate for developing energy-efficient driver monitoring applications that are based solely on the signals from energy-efficient smartphone accelerometers.