Research And Implementation Of Motion-Blurred License Plate Automatic Identification

With technology progressing, more and more vehicles enter people's lives, people's requirements on traffic control and traffic safety management are also increasing, and license plate recognition system as an important part of intelligent transportation plays a more and more important role.Motion-blurred license plate recognition system includes three main parts: restoration of motion-blurred vehicle images, location and segmentation of the license plate, recognition of the license plate. In this paper, use the direction and length of motion blur algorithm for automatic identification, license plate location algorithm, skew correction and character segmentation algorithm, BP neural network algorithm to solve the problem of motion blur license plate recognition. Based on requirements of all aspects on the Motion blur license plate recognition, achieve functional, screening methods and algorithms, and analysis experiments of the key and algorithm, get the final recognition results.1. Through the analysis of degraded image spectrum and using of Radon transform, automatically identify motion blurred images' fuzzy angle. Demand the autocorrelation function of the image after derivatived along the direction of motion with using the fuzzy angle has been automatically identify, finally get fuzzy length of the motion blurred image. Construct PSF using the identified parameters, the last use Wiener filtering to achieve restoration of blurred images. Through simulation experiments, can automatically restore degraded image, there is a certain degree of noise immunity, and recovery effect is obvious.2. For the features of vehicle images have been restored, locate the vehicle license plate and segment image. Mainly use the method based on hybrid characteristics to accurately locate and segment the license plate image. By the processing of multiple vehicle images, showing the way to adapt the diffrent images is strong. It can both deal with the ideal image, but also can handle the image which quality is not good. For ring effect which is generated by the Wiener filtering restoration, it has some flexibility. Use the method of inertia ellipse to obtain the tilt angle. Compared with the Hough transform method, it has a stronger anti-interference ability, and it is more suitable for the image plate tilt correction. Before segmenting characters of the license plate, use Otsu method to get binary license plate, and get a good result. In the traditional method based on vertical projection character segmentation algorithm, do a little improvement, the effect is better for the unconnected character segmentation, and is conducive to the subsequent character recognition process.3. In the character recognition part, use three layers of BP neural network, set the network parameters, after a large number of samples of training, convert the fine grid feature of the character to one-dimensional array, finally input one-dimensional array to network and output using "M taking of one" approach to identify, and ultimately complete the identification, achieve some results.4. Based on the Labview development platform for the realization of system, apply graphical programming language to develop system. system structure is easy to understand.