Korean License Plate Recognition Algorithm Using Deep Learning

Abstract

License plate recognition technology detects the license plate area of a vehicle that exists in an image and recognizes character information within the detected license plate area. Image recognition technology using deep learning technology, which has recently been rapidly developing, has shown remarkable performance in areas such as object detection and recognition, one of the main areas of interest in computer vision. Research using deep learning technology is also actively underway in the field of vehicle license plate recognition, and research showing high levels of performance is being released. License plate recognition technology is not only used for recognizing vehicle information such as unmanned parking management systems, illegal parking surveillance systems, speeding and signal violation surveillance systems, but also in various fields such as black box analysis. In the case of conventional license plate recognition technology, it recognizes license plate regions from a single image and recognizes character information within the area, but the accuracy decreases when license plate recognition is difficult due to various external factors in the image. However, for input data for license plate recognition, it is mainly image data such as CCTV and black boxes, which consists of multiple images, so the license plate region appears to vary over time. In this work, an algorithm is proposed to utilize a deep learning algorithm to use mutlifframe images as input data to recognize vehicle license plate regions and to recognize character information within license plate regions. Moreover, since collecting various kinds of license plate data is physically limited, a data augmentation method has been proposed that can solve the deficiency problem of learning data with only a small amount of data. The accuracy of the proposed system's performance was evaluated as an indicator and divided into LP detection parts and LP registration parts.