Design Of License Plate Recognition System In Complex Scene Based On FPGA

License plate recognition technology is widely used in the life of people with the development of intelligent transportation system,but the stability of license plate recognition is easily affected by the weather and the background of license plate image.In this thesis,we design an FPGA-based license plate recognition system for complex scenarios in hazy weather and complex license plate image backgrounds,The main research elements of this thesis are as follows:Firstly,the improved dark channel a priori algorithm is used to defog the license plate images in hazy weather,and since the processed license plate images may lose in terms of contrast and brightness,the darker license plate images after defogging are subjected to an improved histogram equalization process to improve the quality of the number plate images;Secondly,a license plate localization algorithm based on edge detection and license plate features is used to localize number plates,improving the localization accuracy of blue and green license plates against complex backgrounds using information such as edge detection,color features and number plate aspect ratios;Then,using the combination of vertical projection and character width for character segmentation,the problem of partial Chinese character disjunction and character sticking was effectively solved,and character recognition was performed based on the improved Le Net-5 model,and the numerical results showed that the recognition results based on the improved Le Net-5 were better than those of the classical Le Net-5 and SVM;Finally,based on the hardware-software collaborative approach,the license plate recognition system is built on the ACZ702 hardware platform,with image preprocessing,license plate positioning and character segmentation implemented in the Linux system programming,and the character recognition module is implemented using a hardware accelerator based on CNN to realise the hardware-software integration of the system.Through the function test and performance test,it was verified that the license plate recognition system can complete the recognition task of license plate with blue and green background.The recognition accuracy rate is above 93%,and the running time of the system is reduced by more than 50% compared with the software implementation,and the recognition time is 0.852 s,which makes the accuracy rate and speed of number plate recognition have better superiority,and has important significance to the intelligent traffic system.