TY - GEN
T1 - Deep Learning System for Automatic License Plate Detection and Recognition
AU - Selmi, Zied
AU - Ben Halima, Mohamed
AU - Alimi, Adel M.
N1 - Publisher Copyright:
© 2017 IEEE.
PY - 2017/7/2
Y1 - 2017/7/2
N2 - The detection and recognition of a vehicle License Plate (LP) is a key technique in most of the applications related to vehicle movement. Moreover, it is a quite popular and active research topic in the field of image processing. Different methods, techniques and algorithms have been developed to detect and recognize LPs. Nevertheless, due to the LP characteristics that vary from one country to another in terms of numbering system, colors, language of characters, fonts and size. Further investigations are still needed in this field in order to make the detection and recognition process very efficient. Although this domain has been covered by a lot of researchers, various existing systems operate under well-defined and controlled conditions. For example, some frameworks require complicated hardware to make good quality images or capture images from vehicles with very slow speed. For this reason the detection and recognition of LPs in different conditions and under several climatic variations remains always difficult to realize with good results. For that, we present in this paper an automatic system for LP detection and recognition based on deep learning approach, which is divided into three parts: detection, segmentation, and character recognition. To detect an LP, many pretreatment steps should be made before applying the first Convolution Neural Network (CNN) model for the classification of plates / non-plates. Subsequently, we apply a few pre-processing steps to segment the LP and finally to recognize all the characters in upper case format (A-Z) and digits (0-9), using a second CNN model with 37 classes. The performance of the suggested system is tested on two datasets which contain images under various conditions, such as poor picture quality, image perspective distortion, bright day, night and complex environment. A great percentage of the results show the accuracy of the suggested system.
AB - The detection and recognition of a vehicle License Plate (LP) is a key technique in most of the applications related to vehicle movement. Moreover, it is a quite popular and active research topic in the field of image processing. Different methods, techniques and algorithms have been developed to detect and recognize LPs. Nevertheless, due to the LP characteristics that vary from one country to another in terms of numbering system, colors, language of characters, fonts and size. Further investigations are still needed in this field in order to make the detection and recognition process very efficient. Although this domain has been covered by a lot of researchers, various existing systems operate under well-defined and controlled conditions. For example, some frameworks require complicated hardware to make good quality images or capture images from vehicles with very slow speed. For this reason the detection and recognition of LPs in different conditions and under several climatic variations remains always difficult to realize with good results. For that, we present in this paper an automatic system for LP detection and recognition based on deep learning approach, which is divided into three parts: detection, segmentation, and character recognition. To detect an LP, many pretreatment steps should be made before applying the first Convolution Neural Network (CNN) model for the classification of plates / non-plates. Subsequently, we apply a few pre-processing steps to segment the LP and finally to recognize all the characters in upper case format (A-Z) and digits (0-9), using a second CNN model with 37 classes. The performance of the suggested system is tested on two datasets which contain images under various conditions, such as poor picture quality, image perspective distortion, bright day, night and complex environment. A great percentage of the results show the accuracy of the suggested system.
KW - CNN
KW - Deep learning
KW - LP detection
KW - LP recognition
KW - Segmentation
KW - Tensorflow
UR - https://www.scopus.com/pages/publications/85045211999
U2 - 10.1109/ICDAR.2017.187
DO - 10.1109/ICDAR.2017.187
M3 - Conference contribution
AN - SCOPUS:85045211999
T3 - Proceedings of the International Conference on Document Analysis and Recognition, ICDAR
SP - 1132
EP - 1138
BT - Proceedings - 14th IAPR International Conference on Document Analysis and Recognition, ICDAR 2017
PB - IEEE Computer Society
T2 - 14th IAPR International Conference on Document Analysis and Recognition, ICDAR 2017
Y2 - 9 November 2017 through 15 November 2017
ER -