Intelligent Dynamic Recognition Technology of License Plate under Complex Background

Abstract

In view of the low efficiency of traditional vehicle recognition methods for license plate detection, the low accuracy and robustness of license plate recognition in complex environments, an improved Faster R-CNN algorithm is proposed under complex conditions. Usually, when the vehicle is far, the license plate cannot be detected. In order to make the smaller target feature to be retained on the final feature data, different numbers of convolutional layer are used for target feature fusion, the ratio and size of the Anchor are adjusted, and small target detection rate is increased. In order to locate the license plate which rotates a large angle, the mapping function of the RPN layer is redesigned. In which, the center point coordinates and three parameterized vertex coordinates are used to cover the parallelogram surrounding the license plate. In order to reduce the false detection rate and meet real-time detection speed, a lightweight network is used to remove the redundant background of the picture, which will reduce the detection area of license plate. Then the improved Faster R-CNN algorithm is used to detect the reduced area, and the specific location of the license plate will be obtained. Experimental results show that the method proposed in this paper can effectively improve the speed and accuracy of license plate positioning under complex conditions.