A Study Of License Plate Recognition Under Low Resolution

In recent years, with the increment of vehicle, the situation of city traffic is getting worse and worse. The modern intelligent traffic management gradually becomes a hot research topic. License Plate Recognition which uses computer vision, image processing and pattern recognition technology, is an important research topic in the field of intelligent transportation, and plays a critical role in vehicle detection and automatic management. In practical application, often due to camera quality, placement height, weather conditions and other reasons, resolution of license plate images collected by the camera is low. It makes that the recognition technology research of low resolution license plate has important practical significance.This thesis studies three core technologies in license plate recognition under low resolution, including license plate location, character segmentation and character recognition. The specific work is as follows:In the research of license plate location, through using the rich vertical texture in license plate region, this thesis proposes the license plate location algorithm based on vertical texture. Aiming at the characteristics of low resolution images, this algorithm applies a special horizontal mask to extract the vertical edges of the image. Then using the characters information such as the length and spacing, we scan to locate the license plate line by line.In the research of character segmentation, firstly, this thesis corrects the vehicle license plate tilt and removes the license plate frame. Then, this thesis proposes an improved projection method for character segmentation of vehicle license plates. Using projection information of the license plate characters and combining with prior knowledge, the algorithm recursively finds all possible segmentation results, then according to the variance of character spacing between the center lines to determine the final segmentation result.In the research of character recognition, in consideration of the outstanding performance of convolutional neural network in character recognition field, this thesis uses Le Net-5 for the license plate character recognition. In addition, we take experiments to verify that the character recognition algorithm based on Le Net-5 has good recognition accuracy rate.