Enhancing ALPR: a two stage YOLO model with data augmentation for improved accuracy and robustness
Swati Bansal,Abhilasha Jain,Manoj Sharma,Gautam Kumar,Shivam Ojha,Hemant Walia
DOI: https://doi.org/10.1007/s11042-024-19415-1
IF: 2.577
2024-05-24
Multimedia Tools and Applications
Abstract:Automatic licence plate detection and recognition (ALPR) systems becomes an important task for a variety of applications such as traffic control, toll collection, law enforcement, and parking management. A growing number of innovative licence plate identification systems are now capturing the interest of scientists worldwide. Therefore, ALPR has become a popular study subject for researchers. However, a large portion of the continuing solutions, which often rely on many imperatives, are still not resilient in the current environment. In this paper, an ALPR system based on the cutting-edge YOLO object detector has been presented. YOLO model is trained and fine-tuned for each ALPR stage to ensure robustness under different conditions such as variations in camera, lighting, weather conditions and background. Especially for character segmentation and recognition, we design a two-stage approach employing double data augmentation tricks such as inverted License Plates (LPs) and flipped characters. Noise is also added to the images to improve the confidence level. The most popular datasets and benchmarks are used to train and evaluate the proposed ALPR system. The recall and accuracy attained for the proposed system is 80% and 94% respectively. our proposed system is well suited to improving safety and efficiency in transportation and law enforcement.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering