Research On Methods Of Recognition And Its Parallel Acceleration For Vehicle License Plate Image

License plate location and character recognition are two key steps of intelligent license plate recognition.In view of the current two difficulties:(1)similar plates and distortion of license plate location;(2)the learning algorithm of character recognition is time-consuming,this article puts forward a new algorithm and complete the overall intelligent license plate recognition system from design to implement.The main innovation points and performances of this paper are as follows:1.Consider the similar and distortion of license plates are difficult to location,this paper improve the positioning accuracy from two aspects: firstly,according to the characteristics of the similar and distortion of license plate images this study proposed two new kinds of license plate image descriptor,targeted improved the positioning accuracy;secondly,this paper also raised a classifier voting method to further enhance the positioning accuracy.Experiments show that the descriptors we have proposed can effectively improve the distortion of license plate positioning accuracy and reduce the similar license plate recognition error rate.Compared to only use single descriptor this paper according to introduce the voting algorithm can improve the positioning accuracy from 96.95% to 99.2%.2.In view of the character recognition learning algorithm time-consuming problem,this paper,by using CUDA to improve the speed of the character recognition learning algorithm R-ELM.Our study analysis and convert the R-ELM to resign it in matrix operation form,and use CUDA to optimize it.The experiments show that this study has proposed method can drastically enhance the speed of the character recognition algorithm learning and does not lose the recognition accuracy.3.We complete the whole license plate recognition system from design to implement base on the above two points.This system architecture clearly,functionally,scalability,can be used as a basic development platform for license plate recognition algorithm,and it also has strong practical value.