Automatic Number Plate Recognition with Deep Learning

Automatic Number Plate Recognition is a tool to detect, and extract plate numbers into a computer type of data string. Traffic errors and violations are getting worse without the assistance of this technology to classify the vehicle to cross the highways. It helps detect prohibited vehicles more easily. Building the models require preprocessing the image, extracting the number plate. Finally, feeding the cropped number plate to the Optical Character Recognition. Research shows that YOLOv5 is the most accurate and fastest model to extract the number plate with 0.9659 accuracy. EasyOCR came out with 0.7223 accuracy. The other models are found resulting quite low. ResNet50, VGG16, and PytesseractOCR are respectively 0.9544, 0.7556, and 0.3697. While YOLOv5 at average processes the image around 0.76s. ResNet50 and VGG16 are standing around 6.55s and 0.76s. In conclusion, the best combination for number plate recognition is YOLOv5 and EasyOCR.