| Abstract [eng] |
This work presents overview of technologies and used methods in ANPR systems. After initial study, it was found that all ANPR systems have the same processing of plate recognition: license plate detection, license plate extraction, character segmentation, character recognition. Every part of the process is crucial, as the ANPR system is as effective as its weakest processing part. In order to research object detection methods and their performance – a license plate recognition system was created which implemented all typical processing parts. Two algorithms were analyzed LBP and HOG to see which method offers better detection rate, performance. After analysis it was found that HOG method yields 96,36% detection rate compared with 74,54% using LBP. HOG descriptors have shown average processing speed of an image at 961 ms/image, while LBP shown only 54 ms/image. Moreover HOG descriptors seems to produce more false positive regions at average 1,2 additional false detection per image, compared to LBP at 0,1 additional false detection per image. |
