Conference Presentation:

MMAP in Track2 in FedCSIS 2019:    Multimedia Applications and Processing

Title of paper:License Plate Detection with Machine Learning Without Using Number Recognition

Authors:Kazuo Ohzeki, Max Geigis, Stefan Alexander Schneider

Abstracts:License plate detection has been enhanced by using deep learning, and my group has achieved the world's best performance in 2018 for caltech data. Many methods detect the plate based on number recognition. However, there is an idea that the method using deep learning is difficult to use for automatic driving and the like because of the problem of deterministic verification complexity. Therefore, development of a method that does not use deep learning has become important again. My group is doing high-performance research using deep learning, but is also working on improvement of the method that does not use deep learning in parallel. The CT5L method is the latest type, that includes techniques of the continuity of vertical and horizontal black-and-white pixel values ​​inside the plate, unique Hough transform, only vertical and horizontal lines are detected, the top five in the order of the number of votes to ensure good performance. We proposed a method to determine the threshold value for binarizing input by machine learning, and obtained good results. The detection rate was improved by about 20 points in percent as compared to the fixed case. It achieved the best performance among the conventional fixed threshold method, Otsu's method, and the conventional method of javaanpr.