Licence Plate Detection and Recognition with OCR Using Machine Learning Techniques

An easy-to-integrate licence plate recognition and detection system is proposed in this paper for enhanced safety and security. Several solutions have been presented to date for the recognition and detection of licence plates, but several challenges are still faced by authorities when it comes to tracking licence plates under different circumstances. This study proposes a novel, standalone system designed to overcome most of the drawbacks faced by the other existing systems while maintaining accuracy for detecting and recognising licence plates. Two approaches have been described in this paper for the detection and recognition of licence plates. In the first approach, a pre-trained model called MobileNet SSD FPN-Lite was employed for detection and recognition of plates. It didn't deliver the expected performance. The second approach investigated the application of GLCM for feature extraction to enhance the performance of detection and recognition. The suggested approach made use of several parameters offered by GLCM that help in the detection and recognition of plates. Four classifiers, viz. Decision Tree, SVM, KNN, and Random Forest, were used to recognise the licence plates. The Mobile SSD Net-Pre-trained model gave an average precision of 75.1%, whereas the average precision of GLCM with Random Forest was 89.2%, which came out to be the highest. The testing accuracy that Random Forest offered was 93.5%, which was the highest. In optical character recognition (OCR), EasyOCR was used for OCR as it is a widely used technique. The system successfully detects and recognises the licence plates and identify its characters.