Driving data generation using affinity propagation, data augmentation, and convolutional neural network in communication system

Abstract

In vehicle-driving simulation-based communication systems, vehicles are always driven according to predefined driving styles. However, in the real world, various driving styles exist. To simulate various types of drivers in driving simulation systems, a new driving-data generation method is required. This paper proposes a method that generates a realistic vehicle-driving model. The data augmentation method is utilized to expand the driving dataset, and then the expanded driving data are clustered into several groups. The clustered driving data are inputted into a convolutional neural network to train a driving model. The driving model is utilized to classify another driving dataset into some categories. The driving data within the same categories are utilized to generate new driving data by combining the properties of the driving data. The new driving data thus generated is applied to a vehicle, which can be utilized in virtual driving simulation systems.