Systematic Number Plate detection using improved YOLOv5 detector

The count of automobiles has risen over the past decade on the road. There must be more automobiles on Indian roadways than the citizens living in. It is necessary to automate the fine-collecting procedure which minimizes vehicles from driving too fast and exceeding the posted speed limit by identifying the license plate. In this paper, a systematic number plate recognition (SNPR) methodology was proposed. A system based on YOLOv5s is used for training the model with annotated images in the dataset. The process was divided into several steps, comprising acquisition, detection, segmentation, and finally text recognition in an image. The automobile is recognised from each photograph in the first stage. The next stage is to identify the automobiles' license plates from the identified cars. After the segmentation, the license plates are cropped. The characters are recognised in the last phase from the collected number plates. YOLOv5 is used by the system for number plate detection and Keras for character recognition. The characters from a number plate are retrieved and entered into an excel spreadsheet. Images of Indian license plates are used to evaluate the model's performance. The accuracy for automobile detection, number plate identification and character recognition are 97.6%, 98.2%, and 99.1%.