License plate detection and recognition using the SSD architecture: An approach based on convolutional neural networks

Abstract

This work presents the development and evaluation of a model for the detection and recognition of Chilean license plates, using the Single Shot Multibox Detector (SSD) architecture with MobileNet v2 for license plate detection, together with EasyOCR for Optical Character Recognition (OCR). The SSD model was trained with a set of images of vehicles of various types, using Chilean license plates extracted from .cl sales portals and social networks, and then completed with additional images captured in Arica.
To evaluate the system’s accuracy, the SSD model was integrated with EasyOCR in a single functional system that allows both the detection and recognition of license plate characters.
The objective of this study is to show the efficiency of the SSD model in the detection of license plates and to highlight the potential it has to continue advancing in this area of research, since there are many studies on object detection; however, the vast majority of these are based on better-known approaches such as the YOLO model or Faster R-CNN. The results show a 100% accuracy in license plate detection using the SSD model. For character detection and recognition, an accuracy of 94% was achieved, demonstrating the effectiveness of the SSD architecture with EasyOCR.
The practical implications of this system and possible future improvements in character detection and recognition are discussed, suggesting that the combination of SSD with EasyOCR is an efficient solution for license plate recognition in real-world scenarios.