Automatic License Plate Detection and Recognition for Jordanian Vehicles

Abstract

Jordanian license plates OCR (optical character recognition) rectangularity make of the vehicle Nowadays, automatic number plate recognition (ANPR) is very important especially in the era of smart cities and intelligent transport systems. Fully automated number plate detection and recognition system helps in reducing time, error, and cost for tracking of vehicles and for recording traffic violations. The main goal of this paper is to design a low cost fully automated number plate detection and recognition system targeting the Jordanian license plates. Several problems (e.g., cost, wasted efforts, manual intervention, and possible errors) were identified in the currently used Jordanian number plate recognition for recording traffic violations. We hope that the proposed system would mitigate such problems. The proposed system performs two main tasks. First, it automatically detects and recognizes the license plate number of a given Jordanian vehicle using a robust metric; the rectangularity measurement, and identifies the vehicle's type (e.g., governmental, private, visitor, public, etc.). Second, it recognizes a selected number of trained classes for the make of the vehicle whenever applicable. The experimentation results and the performance evaluations compared to other ANPR approaches show that proposed system achieves the best performance among the tested systems with a plate detection accuracy of 95%, OCR recognition accuracy of 94.68%, make recognition accuracy of 86.84%, and an overall ANPR accuracy of 89% excluding the make results.

Improved Automatic License Plate Recognition in Jordan Based on Ceiling Analysis

“…Similarly, the recognition accuracy of American-and European-standard plates are respectively found to be 74.8% and 84% compared to 36.9% and 43.6% for the old ALPR system. The obtained results are also compared to other Jordanian plate-recognition systems reported in [5,6,20]. In [5] the authors used a dataset consisting of only 46 images and reported an overall system accuracy of 70%.…”

“…In [5] the authors used a dataset consisting of only 46 images and reported an overall system accuracy of 70%. In [6], the accuracy was 89% for a dataset of 100 images. In [20], an improved accuracy of 87% was reported for a dataset of 187,200 images that was used to train a deep neural network.…”

“…The main three components of the multi-stage method are detection, segmentation and recognition. These components, which were of particular interest in many studies including those of the Jordanian license plates [5,6], are described as follows.…”

“…Character segmentation relies on different techniques. The most common technique of character segmentation is the connected component analysis (CCA), as reported in [6,10,[13][14][15][16]. CCA technique depends on the connectivity of the character region (pixels), especially for letters and numbers.…”

“…CCA technique depends on the connectivity of the character region (pixels), especially for letters and numbers. Other techniques include the utilization of the morphological and histogram [3][4][5][6][7] and radon transformation [9]. The latter technique is concerned with the angle of inclination and partial vertical projection to extract and divide the characters.…”