Automatic license-plate recognition

Introduction. The problem of automatic license plate recognition is considered. Its solution has many potential applications from safety to traffic control. The work objective was to develop an intelligent recognition system based on the application of deep learning algorithms, such as convolution neural networks that consider automotive standards for license plates in various countries and continents, and are tolerant to camera locations and quality of input images, as well as to changing lighting, weather conditions, and license plate deformations.

Materials and Methods. An integrated approach for the problem solution based on the application of convolution neural network composition is proposed. An experimental analysis of neural network models trained to meet the requirements of the universal license plate recognition task was conducted. Based on it, models that showed the best ratio of quality and speed were selected. Quality of the system is provided through the optimization of various models with different modifications. In particular, convolution neural networks were trained using images from several datasets. In addition, to obtain the best results, the models used were pre-trained on a specially generated synthetic dataset.

Results. The paper presents numerical experiments, the results of which imply the superiority of the developed algorithm over the commercial OpenALPR package on public datasets. In particular, on the 2017-IWT4S-HDR_LP-dataset, license plate recognition accuracy was 94 percent, and on the Application-Oriented License Plate dataset, 86 percent.

Discussion and Conclusions. The resulting algorithm can be used to automatically detect and recognize license plates. The experiments show that the algorithm quality meets or exceeds quality of the commercial OpenALPR package. The developed algorithm quality can be improved through increasing the training dataset, which does not require the participation of the developer.

1. Bernstein, D. Automatic vehicle identification: technologies and functionalities / D. Bernstein, A. Y. Kanaan // Journal of Intelligent Transportation System. — 1993. — 1 (2). — P. 191–204.

2. Development of vehicle-license number recognition system using real-time image processing and its application to travel-time measurement / K. Kanayama [et al.] // IEEE Vehicular Technology Conference. — May, 1991. — P. 798–804.

3. Kessentini, Y. A two-stage deep neural network for multi-norm license plate detection and recognition / Y. Kessentini, M.D. Besbes, S. Ammar, A. Chabbouh // ExpertSystems with Applications. — 2019. — Vol. 136. — P. 159– 170.

4. Tian, J. A two-stage character segmentation method for Chinese license plate / J. Tian, R. Wang, G. Wang, J. Liu, Y. Xia // Computers & Electrical Engineering. — 2015. — Vol.46. — P. 539–553.

5. A CNN-based approach for automatic license plate recognition in the wild / M. Dong [et al.] // British Machine Vision Conference (BMVC). — 2017. — P. 1–12.

6. LeCun, Y. Deep learning / Y. LeCun, Y. Bengio, G. Hinton // Nature. — 2015. — Vol. 521, no. 7553. — P. 436–444.

7. Holistic recognition of low quality license plates by CNN using track annotated data / J. Španhel [et al.] // 14th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS). — 2017. — P. 1–6.