Institutional Repository of Vilnius University
Super-rezoliucijos neuroniniai tinklai automobilių valstybinių numerių didinimui /
Abstract
dc:descriptionLicense plate recognition (LPR) is used in intelligent transportation systems such as speed or private parking barrier cameras. The speed and accuracy of the algorithms in these systems are of big importance. Super-resolution methods that increase the quality of given video or photo can be used to improve the accuracy of license plate detection and reading algorithms. In this thesis, generative adversarial and shifted window transformer neural networks are examined for car license plate super-resolution. First, generative adversarial network ESRGAN is examined. ESRGAN with perceptual loss function can upsample fine image details. Second examined super-resolution neural network model is SwinIR, which uses shifted window transformers for feature map extraction. MobileNet SSD neural network was chosen for license plate detection, considering the efficiency and speed of the algorithm and for optical character recognition (OCR) EasyOCR algorithm was chosen. Conducted experiments have shown that SwinIR performs about 7 % better than ESRGAN super-resolution model. It was noticed that super-resolution method does not improve the accuracy of detecting car license plates, therefore, to increase the speed of LPR, super-resolution is applied after detection. However, conducted experiments show that super-resolution algorithms improve the accuracy of OCR by up to 19 %. The speed of upscaling methods was also measured. Applying the fastest super-resolution neural network from the models used in the experiments, the LPR inference time increases by 70 %. Experiments have also shown that that images, upsampled with bicubic interpolation algorithm also improve the accuracy of OCR algorithm.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Vingis, Simonas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD146243003&prefLang=en_US