| License plate recognition is one of the core parts of intelligent transportation,which depends on the license plate information collection accurately.In a complex traffic environment,there are many factors,such as day and night alternation,light changes,bad weather,and license plate contamination and alteration,will affect the accuracy of license plate data that obtained by visible/infrared imaging systems,and then affect license plate recognition.Therefore,how to solve these problems has been paid widely attention to by the domestic and oversea researchers.In this research,spatial-spectral point cloud data of altered license plate are obtained by an active hyperspectral Li DAR(HSL)system,which is not affected by external light.The license plate image reconstruction is completed by using the differences in the spatial and spectral characteristics of character and background point cloud.An altered license plate recognition method is designed based on the spectral reflectance difference between license plate and altered materials.The main work is as follows:(1)The structure of HSL and the principle of HSL point cloud data acquisition in spatial-spectral domain are introduced.The point cloud data of different color retroreflection materials at different positions were collected,and the distance was calculated using the time difference between main wave and echo peak.It is found that there is a difference between the calculated and measured distance,which is affected by actual distance and material.The reasons for the difference between the reconstructed distance and measured distance of different color retroreflection materials were analyzed from the characteristics of retroreflection materials and the quantizer of HSL system.(2)A license plate point cloud separation method based on the measurement distance is realized based on the distance difference between license plate character and background.Firstly,select a characteristic wavelength,the histogram distribution of the measured distance is introduced to determine the threshold,which is used to separate these two parts,and then the license plate image is reconstructed by two-dimensional image reconstruction method based on camera model.Experimental results show that the proposed method can effectively separate license plate characters from license plate background point cloud data,but the reconstruction image is greatly affected by laser power fluctuation,structural integrity of license plate retroreflection material,distance and other factors.A method of license plate point cloud separation is proposed based on the difference of their spectral reflectance.Select average reflectance in the characteristic band as threshold,and a threshold method is used to complete plate point cloud separation,and then the license plate image is reconstructed by two-dimensional image reconstruction method based on camera model.Experimental results show that the method can effectively separate plate point cloud data,and the reconstructed image is better than the former method,which can alleviate the effects from laser power fluctuation,structural integrity of retroreflective material,distance and other factors.(3)Some common altered license plates were simulated by altering plate with different materials,and whose point cloud data were collected by HSL in a laboratory environment.According to the difference of spectral reflectance between license plate surface and altered material,a multi-threshold recognition method based on hyperspectral characteristic is proposed.The altered and non-altered part are separated with average reflectance,and then alteration type is distinguished with the gradient of reflectance.According to alteration type,the characters or background are removed or compensated to restore original point cloud data of license plate.The restored point cloud information is converted into a two-dimensional image,and which is used as input for an improved license plate recognition system.The recognition experiments are conducted on some common altered plates,the proposed method can reconstruct original license plate information and recognize them.Figure[45]table[7]reference[55]... |
PDF Full Text Request