Research Article
Real-Time Detection and Recognition of License Plates for Traffic Monitoring
@INPROCEEDINGS{10.1007/978-3-031-18123-8_32, author={Nam Van Nguyen and Quan Minh Vu}, title={Real-Time Detection and Recognition of License Plates for Traffic Monitoring}, proceedings={Multimedia Technology and Enhanced Learning. 4th EAI International Conference, ICMTEL 2022, Virtual Event, April 15-16, 2022, Proceedings}, proceedings_a={ICMTEL}, year={2022}, month={10}, keywords={License Plate Detection and Recognition Convolutional neural networks Attention mechanism and transformer}, doi={10.1007/978-3-031-18123-8_32} }- Nam Van Nguyen
Quan Minh Vu
Year: 2022
Real-Time Detection and Recognition of License Plates for Traffic Monitoring
ICMTEL
Springer
DOI: 10.1007/978-3-031-18123-8_32
Abstract
We address the task of real-time detection and recognition for heterogeneous license plate images of diverse vehicles with characters arranged in multiple lines and captured in all day and night conditions. This paper presents MixLPR (Mixed License Plate Recognition), a framework to develop a real-time system deployable in dense urban traffic to fill that gap. MixLPR consists of two components, a license plate detector, and an OCR. The plate detector includes new Mish-enhanced residual nets equipped with geometric transformations to deal with view-induced distortions. The OCR component is a new segmentation-free method based on the transformers, which work directly on the 2D character block. We trained and validated MixLPR on two large public and private datasets. Our results exhibit both improvements in accuracy and inference speed compared to state-of-the-art approaches.
