Improved Deep Learning Based Model for Vehicle Plate Detection, Recognition, and Authentication

Abstract

In recent years computer vision models have made our daily life easy in various ways, especially in reducing roadside problems. Many research works are already completed to achieve the goal of automated road surveillance. But these models' actual implementation has failed due to the poor accuracy of the model and other relevant factors. This paper presents an improved model to detect, extract, recognize and validate Bengali license plates from vehicles. In order to recognize vehicle plates more accurately and for various uses, including automated vehicle monitoring, roadside assistance, toll collection, parking management, etc., we implemented a Yolo-based CNN model to detect Bangla license plates and mask R-CNN for recognition of license characters. A total of 6528 images were used in training our model. Based on roadside test images, the experiments can detect at a rate of 98.2%, recognition of 95.6%, and a validation rate of 100%, respectively.