Implementing an improved intelligent licence plate detection system using image processing and pattern recognition algorithms

Abstract

In recent years, there had been a substantial increase in demand of an efficient implementation of Automated License plate recognition (ALPR) system in countries all over the world with the system been commercially developed, deployed and used in cities as a tool for mass security surveillance. An implementation of ALPR system involves three processes: location of the license plate region; segmentation with extraction of the plate characters; and recognition of the extracted plate characters. This thesis research focuses on the first process in implementing ALPR, which is location of the license plate characters. In this research, an algorithm is proposed that adopt the concept of Histogram of Oriented Gradients (HOG) into developing a custom algorithm for the feature extraction of a license plate portion from an image by taking into consideration the structure of the plate. Artificial Neural Network (ANN) was employed to classify and assign similarity scores to any portion selected and Genetic Algorithm (GA) samples, selects and evaluates portions using the similarity scores at which it finally select the most likely portion as the plates. Three major contributions were made by this research: the successful use of HOG in feature extraction of a plate; development of a proposed structured HOG for plates; and the successful use of the combination of GA, HOG and ANN algorithms in plate detection process.The ANN was trained with Turkish manually cropped license plates. The plate detection was tested on a database of car images containing Turkish plates and on a database of car images with Non-Turkish plates with varying plate sizes, various environmental conditions and a lot low quality plate images. 1526 plates out of 1572 plates in the car images (97.03 %) where successfully detected for the car images with Turkish license plates and 601 plates out of 655 plates in the car images (almost 92%) where successfully detected for the car images with Non-Turkish license plates.

In recent years, there had been a substantial increase in demand of an efficient implementation of Automated License plate recognition (ALPR) system in countries all over the world with the system been commercially developed, deployed and used in cities as a tool for mass security surveillance. An implementation of ALPR system involves three processes: location of the license plate region; segmentation with extraction of the plate characters; and recognition of the extracted plate characters. This thesis research focuses on the first process in implementing ALPR, which is location of the license plate characters. In this research, an algorithm is proposed that adopt the concept of Histogram of Oriented Gradients (HOG) into developing a custom algorithm for the feature extraction of a license plate portion from an image by taking into consideration the structure of the plate. Artificial Neural Network (ANN) was employed to classify and assign similarity scores to any portion selected and Genetic Algorithm (GA) samples, selects and evaluates portions using the similarity scores at which it finally select the most likely portion as the plates. Three major contributions were made by this research: the successful use of HOG in feature extraction of a plate; development of a proposed structured HOG for plates; and the successful use of the combination of GA, HOG and ANN algorithms in plate detection process.The ANN was trained with Turkish manually cropped license plates. The plate detection was tested on a database of car images containing Turkish plates and on a database of car images with Non-Turkish plates with varying plate sizes, various environmental conditions and a lot low quality plate images. 1526 plates out of 1572 plates in the car images (97.03 %) where successfully detected for the car images with Turkish license plates and 601 plates out of 655 plates in the car images (almost 92%) where successfully detected for the car images with Non-Turkish license plates.