Driver Decision Space Inversion Method Based on DC-GAN

Abstract

In the driving process, drivers need to constantly perceive the surrounding environment to make decisions and perform operations. This cognitive space formed in the driver's mind is the driver's decision space. For the actual driving environment, factors such as complex road sign settings, unreasonable road planning, long-time fatigue driving, and reduced reaction ability of elderly drivers may interfere with the normal perception of the driver's decision space, resulting in reduced driving safety. In this paper, a driver decision space inversion method based on deep convolutional neural networks and generative adversarial networks is proposed to study driver perception in near-domain traffic scenarios. The steps of the method include: near-domain target element extraction, driving data collection, sample data generation, adversarial generative network model learning, data enhancement and decision space inversion. The experimental results show that the method in this paper can accurately identify the driver's decision space in both real and simulated driving scenarios by implementing real-time monitoring of driver perception. The method is important for studying the driver's decision space, which can promote the development of intelligent driving technology and the synergistic development of human-vehicle-road-loop. In the context of today's sustainable transportation and smart mobility, driver decision space research is not only an important basic research, but also an inevitable requirement to promote innovation and upgrading in the transportation field.