A study “Deep neural network method for improved accuracy of tracking and classification of vehicles registration plates” of a project No. 1.2.1.1/16/A/007 “Information and Communication Technology Competence Center”.
In collaboration with SIA “SQUALIO CLOUD CONSULTING”, EDI researched and developed image processing methods for detection and recognition of vehicle license plates. The main goal of the research was reaching sufficient accuracy and robustness of the system so that it can be deployed on the roads. The resulting system is based on deep neural networks (fully convolutional network segments the image and localizes the license plate while a bi-directional recurrent network recognizes the characters). In addition, the system is able to determine the type of vehicle (light vehicle, truck, emergency vehicle, motorcycle) and the issuing country of the license plate. On a test set, the system detects number plates with 99.5% accuracy and correctly recognizes all characters on 96.7% of license plates. The technology is described in more details in two collaboration papers:
Vehicle type and licence plate localisation and segmentation using FCN and LSTM