Research On Precise Plate Location Algorithms

License plate recognition system plays an important part in Intelligent Transportation System(ITS).It has been widely used in occasions like access control,flow monitoring,elec-tronic toll collection,improving the degree of automation of traffic management.By analysis and process of automobile images taken under complex background,license plate recognition system detects,locates the license plate,recognizes the charcaters of the plate,leading to fast automobile identification.Among those steps,precise plate location is an significant basis of plate alignment and accurate character segmentation.Traditional license plate location methods rely on geometry,color and texture features,taking many steps to get result,as well as having limited application scenarios.Their accuracy need to improve in the scene of low brightness,perspective transformation and low quality photography.To achieve the target of building a more general and precise plate locaiton algorithm,this essay is inspired by shape regression method commonly used in facial landmark detection,proposes to convert the problem from plate precise location to the plate landmark detection(which is detection of four coordinates of the plate).This algorithm which takes full advantage of abundant labeled plate data,learns the feature of plate coordinates,builds multiple stage regression equations and approaches to the real plate coordinates using feedback stage by stage.By combining various feature equations and regression equations,shape regression algo-rithm have different implementation.Compared with traditional plate locaiton algorithms in the the real and complicated dataset,the result shows that,plate location algorithm based on shape regression has higher speed and lower position offset and depends less on the envi-ronment,angle or distance from where photos are taken.It has good generality.Meanwhile,the plate location algorithm based on shape regression sets the stage for improving plate recognition accuracy.