Research And Implementation Of Key Technolo-Gies Of Moving License Plate Recognition In Un-Constrained Scene

Intelligent transportation system has been widely used due to its nu-merous applications.However,traditional license plate recognition ap-proaches for constrained scene are difficult to satisfy the demand for intel-ligent transportation applications.The challenges are due to motion blur,horizontal direction skew and vertical direction shear.Motion deblur and skew correction are key technologies to improve accuracy of license plate recognition in unconstrained scene.Aiming at the above challenges,we design a moving license plate recognition system.Firstly,in the kernel estimation procedure from the motion blurred restoration part,we use the guided filter to remove noise instead of traditional bilateral filter.Then,we optimize the process of cal-culating the blur kernel according to the various features of license plate images that makes the process more applicable.We use Hyper-Laplacian distribution to fit the image distribution of gradients in order to estimate more explicit and smooth latent images.Moreover,we propose a skew&shear warp correction pre-processed algorithm based on the multi-ple color space fusion.By using this algorithm,we can get the mask of plate part in order to detect horizontal and vertical skew angles more pre-cisely then get the accurate correction result.The experimental results il-lustrate the satisfactory effect and good noise robustness of the proposed approach.At last,on the basis of the proposed algorithm,we design and implement the moving license plate recognition system in unconstrained scene,and test the function of this system in real scenes.