Research On License Plate Recognition Algorithm In Complex Environment

As the number of cars in our country is increasing dramatically,traffic problems are becoming increasingly prominent.In order to relieve traffic pressure and improve the management efficiency of vehicles,intelligent transportation system emerges as the times require,and license plate recognition system is an important component of intelligent transportation system,so the research on license plate recognition technology is of great significance.Although China's license plate recognition technology has achieved good research results under long-term research and efforts,and can accurately recognize license plates in most scenes,license plate images obtained in complex environments often have various interference factors,such as harsh lighting conditions,license plate inclination,license plate fouling,etc.Using existing license plate recognition technology to locate and recognize license plate images in complex environments has not achieved satisfactory results.To solve these problems,this paper makes corresponding improvements on the basis of the existing license plate recognition technology to make vehicle license plate recognition more suitable for complex environment.The main work is as follows.Firstly,in order to avoid the problem that texture information is easy to be lost during image enhancement,this paper uses homomorphic filtering algorithm in frequency domain to enhance license plate images.In view of the problems of many filter parameters and complicated calculation in previous homomorphic filtering,the traditional Butterworth filter is improved,and the unknown parameters are reduced from five to two.Experiments show that the algorithm complexity is greatly reduced and the calculation time is saved while the image enhancement is not affected.In license plate location,combined with illumination compensation,the HSV color model of license plate is separated in three channels,the hue component is kept unchanged,the saturation component is stretched,the illumination compensation is carried out on the luminance component by using improved homomorphic filtering,and the license plate region is located by combining the color and structural characteristics of the license plate.Experimental results show that the algorithm has obvious improvement on license plate location effect in complex environment.Secondly,in order to solve the problem that the traditional license plate tilt correction algorithm requires a higher number of license plate borders,this paper introduces an image low rank texture invariance algorithm and applies it to license plate tilt correction.Experiments prove that the algorithm does not rely on license plate borders,and can correct tilted license plate images only according to the low rank of non-tilted license plate images.In character segmentation,this paper uses projection segmentation algorithm,but in traditional vertical projection,if a single character breaks,it will lead to the failure of character segmentation.Therefore,after introducing the character template frame designed in this paper,the complete license plate characters can be accurately and effectively segmented.Finally,the traditional LeNet-5 network is improved and integrated into a double-layer LeNet-5 Convolutional Neural Network.Parameters and excitation functions in the network are adjusted and license plate characters are trained to obtain the optimal model of the network.Experimental results show that the improved double-layer LeNet-5 network is superior to the traditional LeNet-5 network in the recognition rate of license plate characters,and is obviously superior to template matching and BP neural network algorithm in the recognition rate,and has higher robustness.