License plate localization, the core component of license plate recognition system, is valuable in both academic development and potential applications. Though much progress has been made in recent years, challenging problems still exist in the complex scenes, such as low luminance, low resolution and inclination scence of vehicle. This paper proposes a novel fully convolutional neural network to localize license plates accurately by a corner regression algorithm. To guarantee effective training in the proposed model, 45 000 sample images are annotated by one person. Meanwhile, the annotated sample images are processed by four operators, including translating, scaling, rotating and noising, to increase the number and diversity of the training samples. Extensive experiments on the newly collected datasets, trafficmonitoring dataset and the complex scene dataset, demonstrate the effectiveness of the proposed method against other two license plate localization methods.