In the sixth-generation (6G) network-based intelligent transportation system (ITS), high-precision vehicle localization is essential to ensure vehicle safety. Especially, with the application of large-scale antenna arrays in 6G, such as reconfigurable intelligent surface (RIS), the issue of near-field vehicle localization needs to be considered. In this article, different from the approximate model which only considers the phase difference among antennas, the received signal is modeled in a more practical near-field scenario, where both path attenuation and delay are well described. In this practical model, vehicle localization is a nonlinear problem that cannot be efficiently solved through conventional methods. To address this challenge, an efficient method based on the deep neural network (DNN) is proposed for near-field localization, i.e., near-field localization network (NFLnet). Compared with the existing DNN-based method, which needs substantial computing resources, the proposed NFLnet significantly decreases the calculation and the training time. Moreover, a prototype using the low-complexity NFLnet is established to demonstrate the localization performance in the practical scenario. Both simulation and experimental results show that the proposed NFLnet can localize the near-field vehicles effectively and outperforms the compared methods.