Design and Implementation of License Plate Positioning Algorithm based on FPGA

Compared to image processing platforms such as CPU and GPU, FPGA have advantages in terms of real-time performance. The license plate positioning method based on Field Programmable Gate Array (FPGA) proposed in this paper only takes 13ms to process a frame of 640∗480 images, which takes 19ms based on GPU and 58ms based on CPU[1]. The image preprocessing method of binarization by single-channel color requires a high degree of chromaticity of the license plate part in the image. Especially when the color of the license plate is similar to the car body color, the recognition accuracy is very low. The method of binarization using two-channel mask proposed in this paper can effectively improve the accuracy of license plate recognition when the color of the license plate is similar to that of the car body or the chroma of the license plate is insufficient. Because in the actual installation of the license plate, there are four rivets on the upper and lower edges. So in the image after Sobel edge detection, the rivets affect the completeness of the upper and lower edges, resulting in irregular lines on the upper and lower edges of the license plate. When using the traditional pole method to locate the upper and lower boundaries, the error is more than 12%. When using the three-order matrix method proposed in this paper to locate the boundaries of the license plate, the error value of the four boundaries is no more than 5%, compared with the manually calibrated coordinates.