New Approach to the Detection and Recognition of Brazilian Mercosur Plates Using Haar Cascade and Tesseract OCR in Real Images

The challenge of license plate recognition (LPR) detection and recognition has different solutions in the literature. LPR systems also address various proposals for technological use, from vehicle tracking on roads to detection and prospecting of routes through plate capture. In this sense, this paper presents a new approach for character detection and recognition based on training synthetic Mercosur model plates, tested on real Mercosur model plates. The proposed study also introduces a new database (LPR-UFC) available by request, containing 1.100 real images of different vehicles with Mercosur plates. The results with the new model in conjunction with perspective adjustment obtained 14% more accuracy than the model without the plate perspective process (713% against 85 %) for character recognition, reaching 90% accuracy for plate detection, surpassing state of the art for real dataset images (LPR-UFC). The results also beat studies found in the literature.