Eliteun applied OCR and Convolutional Neural Network (CNN) in License Plate Recognition (LPR), relying on self-designed “feature engineering” and multiple leaning model, the system achieved excellent performance even in outdoor scenarios with challenging weather or lighting conditions, such as:
The recognition accuracy of LPR is mainly impacted by two factors: 1. the software platform bearing the algorithms, 2. quality of acquired data which is the image of license plate. The capacities of applied algorithms not only determine the final recognition accuracy, but also affect the diversity of license plates the system supports. License plates follow different plate structure and different format in different countries, creating different plate syntax for computer to analyze. Therefore, recognizing different license plates with both accuracy and efficiency is very demanding. The final recognition result is the multiplication of a series of sub-algorithms such as plate localization, contrast normalization, character segmentation and so on. In addition, the entire process needs to be trained to yield consistent accuracy.