A Deep Learning Framework for Accurate Number Plate Recognition using OWL-V2 and PaddleOCR

Recognizing number plates holds significant importance in various applications such as law enforcement, traffic management, toll collection, parking management, and security. Manual recognition is challenging one and is prone to errors. Automation of number plate recognition enables faster inference and timely response from the police personnel. Artificial intelligence techniques using deep learning models are useful in this regard. Typically, Automated Number Plate Recognition (ANPR) technology to identify and process vehicle license plates effectively Such as Image Capture, Image Processing, Character Recognition, Alerts or Actions. This paper presents a vision-language model-based approach for automatic detection. Number Plate Recognition (NPR) is a technique that identifies alphanumeric characters from license plates and converts them into text format. In this work, different deep learning models were utilized for the detection and recognition of number plates. The model was tested on a car image, processed and number plate datasets. For this recognition task, zero-shot models such as OWLVIT and Grounding DINO, were employed. Additionally, techniques like PaddleOCR were integrated. The proposed tests demonstrated that the system can accurately detect number plates with an impressive accuracy of 92.91%, even under challenging conditions.