An Applicational Approach Toward comparing Deep Learning Based Methods and Classical Machine Learning Methods in Traffic Industry

Deep learning based methods are growing among researchers in different machine vision fields. In traffic industry, many applications are switching from traditional andclassical machine learning methods to deep learning based methods. In this paper, we have experimented an accurate comparison between deep learning based methodsand classical ones on Automatic Number Plate Recognition (ANPR) application. This comparison is based on both accuracy and processing time. For this comparison, a database of ten thousands images from different ANPR cameras installed in Tehran is used. We have implemented these methods on many different hardware platforms. Although in many parts deep learning based methods reach better results, but in this paper it is shown that they still lack some optimization to replace traditional methods completely.