License Plate Positioning and Recognition Algorithm Based on PYNQ

To solve the problem of inaccurate positioning caused by rivets on the upper and lower edges of the license plate, a new matrix threshold method was proposed to define the edges of the license plate. The license plate area was located by color and edge features, and then a three-order matrix operator was constructed to judge the coordinates of the edges of the license plate. The algorithm reduced the error of license plate coordinates from 10% to less than 5%, compared with that of the traditional pole method. On this basis, in order to improve recognition accuracy, the deep-shallow features short-circuit fusion algorithm was proposed to obtain more detailed information of the characters on the plate. Compared with the classic CRNN character recognition algorithm, the accuracy of the algorithm for license plate characters was increased from 89.1% to 89.9%. Finally, for the application requirements of small device embedded systems, the algorithm was deployed on the PYNQ platform based on FPGA. The image acquisition and display was realized on the programmable logic, and the image processing and character recognition was realized on the processing system, and the effectiveness of the algorithm was verified in the real scene.