License Plate Detection and Recognition Using CRAFT and LSTM

This work proposes a solution for developing a license plate detection and recognition system (LPDR). In the first stage, the poly region of the license plate's word line(s) can be detected by Character-Region Awareness For Text Detection (CRAFT). Specifically, the text line of the one-line license plate and two lines of the multi-line license plate can be detected effectively. Secondly, each region proposed as a plate number region by CRAFT will be passed to Mobilenet architecture to extract features. Finally, these features will be fed to Bi long short-term memory (Bi-LSTM) architecture with Connectionist Temporal Classification to predict output text in each input region. By applying this solution, the problem of multi-line license plates can be appropriately handled.