Night Vision Enhancement for License Plate Recognition based on Deep Learning

Abstract

License plate recognition technology plays an important role in traffic management. It is widely used in parking, high-speed and road traffic management. The existing license plate recognition system is easy to be disturbed by the external environment, and the detection performance is poor in the night scene. This paper investigates the problem of license plate recognition in night vision scenarios and license plate tilt. This paper proposes a license plate recognition method with night vision enhancement, which can detect and recognize license plates under extremely poor lighting conditions. The method first uses a night vision enhancement module, Recursive Encoder-Decoder Network (RED-Net), and a set of non-reference loss functions designed for properties of the image. And then the License Plate Location and Recognition (LPLR) system is used to get the license plate number. The algorithm is tested on 75k images of the simulated CCPD night dataset. The testing result that the accuracy of the license plate recognition algorithm after night vision enhancement can reach 72.29% increase 65.5% than without night vision enhancement.