Merged Application of YOLOv7 Object Detection and Deep Text Recognition for Four-Wheeled Vehicle License Plate Recognition

Manual vehicle profiling, identification, and monitoring methods compelled the development of novel and reliable computer vision techniques. Increasing demand exists for accurate and efficient license plate recognition (LPR) methods to automate vehicle profiling. This study investigated a trained YOLOv7 (You Only Look Once) object detection model to detect vehicle license plates. Deep Text Recognition (DTR) was adopted for character segmentation and recognition of license plate texts. DTR, a type of scene text recognition (STR), is more versatile to various scene conditions than Optical Character Recognition (OCR) techniques. This study's proposed LPR architecture can be utilized to substantially improve real-time vehicle identification and profiling, which is essential for applications such as traffic management, security enforcement, and surveillance systems. The trained license plate detection model has an accuracy of 97.71%, as measured by its mAP50 score. In addition, the trained DTR model has a 90.50% accuracy and a 4% Character Error Rate (CER). The calculated overall system accuracy is 88.43%.