Number Plate Recognition System
By
KJ Tsiri
A thesis submitted in partial fulfillment of the requirements for the degree of
Bsc(Hons) Computer Science
University of the Western Cape
2013
Date: October 29, 2018
University of the Western Cape
Abstract
Number plate recognition system to grant access
ByKJ Tsiri
Supervisory:Adiel Ismail
Department of Computer Science
Abstract
South African automatic number plate recognition (ANPR) system using artificial neural network. The system of recognition contains of three main major parts- number plate detection, plate character segmentation and South African character recognition. They are several processes that have to be done in the process of number plate detection. The are many problems and complications that arise under the number plate detection such as vehicle in motion, complex background, distance changes, unclearness of the number plate, during day and night etc. For solving this kind of the problem we will apply edge analysis method. By the use of Horizontal and Vertical projections, the number plate segments will be able to be identified fully. South African Number plate contains of letters and numbers and number plates are segmented into characters. The feature extraction method is applied to extract the necessary characters needed for the number plate to be recognized successfully and after this method the characters are recognized.
Table of Contents
Table of Contents
List of figures
Acknowledgments
Glossary
Introduction
Chapter 1
user Requirement Document
Background
User’s view of the problem
What is expected from the software?
What is not expected from the software solution?
Chapter Summary
Chapter 2
Requirements analysis document
Introduction
Designer’s interpretation of the user’s requirements
Breaking down the problem into high level constituent parts
Capture Image
Gray Scale Conversion
Preprocessing
Binarization
Number plate Detection (Find plate region)
Number Plate Extraction
Character Segmentation
Feature Extraction
Artificial Neural Network (Recognition of Characters)
Output
Requirements for the system
System Requirements
Identify existing solutions
Devise ways to test the solution
Chapter summary
Appendices
bibliography
List of figures
NumberPage
[Click and insert List of Figures]
Acknowledgments
Firstly I would like to thank God for giving me the chance to still live and see many years. I want to thank all the guidance that I got from Mr. Ismail; your help into this project is noted. Everyone who played good role into this project just wants to say thank you.
Glossary
Segmentation: This is the process were characters are being segmented or divided into more manageable task that could easily be worked on better that the full image.
Grayscale: This the process were the 24bit image color is being reduced or converted to 8bit color that could be used and accepted by the number plate system.
Image Enhancement: This is the process of increasing the contrast on the image, if the image is unclear this method can be used to increase or make it better to read or be able to see the image with better quality.
Introduction
Number plate recognition is a process where vehicles are identified or recognized using their license plate or number plate. This system involves few processes or steps in the process of identifying the vehicles using their number plate.
The first thing that will take place will be the sensor or vehicle detector that will be so helpful in activating the digital camera so that it can start taking the image of the vehicle for recognition purposes. After the image of the vehicle has been taken, the first process is finding the number plate regionon the image captured.
Find the plate region involves looking or detecting the plate region on the image been captured and the method that follows is plate extraction that simply means the plate that have been identified is now been extracted out of the image and from this point we only going to start focusing on the plate image that has been extracted. Character segmentation follows were simply slicing of the plate image is taking place, this simply means the image plate will be divided into smaller manageable images and from this point will be focusing on the single image of each character.
Feature extraction follows were characters will be extracted from the single images from character segmentation method. After this method comes character recognition were characters will be clearly recognized from this method and after this method the vehicle will be recognized or identified using its license or number plate and as for uwc university the system at this point will grant or deny access into the university.
1
Chapter 1
user Requirement Document
Background
Traffic management is struggling to identify vehicles whose owners have violated traffic laws e.g. speed limit on the road is one of them. The other problem that’s faced is the problem of stolen vehicles and also the problem of computing the parking cost or fee. Uwc as our main focus is failing to keep track of cars or vehicles entering and living uwc premises, their main concern being students can easily steal one another’s vehicle and they won’t be able to realize in time or the current system they using would not alert them in time.They use the student card system.
User’s view of the problem
In order for me to get to know how user’s feel or how they view the problem, I managed to interview few people around that owns or drives cars and also the people that are affected by the problem all the time or most of the time, I found out that most of them have equal or similar view. I also used a questionnaire in my interview in order for me to get accurate results. Firstly users have a problem when they lost their vehicles or their vehicles are stolen, they complaining that the procedure to follow up for helping the traffic management is time consuming and at the end of the day they end up not getting the vehicle back.The other one is faced by traffic management where they have to identify vehicles that their owners have violated traffic laws; speed limit is on road is also a problem faced. The otherproblem is computing the packing fee after the users has used the parking lot.
What is expected from the software?
The system is expected to detect vehicles or cars approaching and compare the number plate with the stolen ones including checking if the car is in good condition like has not being given many fines lately. From the parking lot side the system is expected to keep track of the times the vehicle entering and living the environment, the system will get the amount of hours spend by the user using the packing so the system could calculate proper fee. This system will only start processing after the sensor detected that the vehicle approached or need attention and after that the digital camera will be activated as to start capturing images. For uwc university the system has to check if the vehicle have being registered with uwc that is being driven by the student, for checking it will require the database of where the student’s information is stored and if all correct it will grant access to the university else will deny access meaning the particular student would not be allowed to enter the uwc environment.
What is not expected from the software solution?
The system is not allowed to recognize two vehicles at the same time; it should deal with one car at a time. Again cannot detect a vehicle 10m away and if the car that’s being detected does not contain the front number plate or if the number plate is not detected at all from the vehicle image and if the number plate is broken the system would not identify that particular vehicle.
Chapter Summary
In this chapter a problem with the existing system is being identified and analyzed in a way that users gave their own requirements that they think are good for the system. . In the next chapter the user’s requirements are analyzed in terms of what is possible with the technologies available and the time allocated for the project.
Chapter 2
Requirements analysis document
Introduction
In the previous chapter the user’s requirements and the problems of how the number plate recognition could be used to change a lot of things, were identified. In this chapter the requirements will be analyzed and solutions to the problems will be posed.
Designer’s interpretation of the user’s requirements
In the previous chapter the problem and the user’s requirements for the system was briefly described. The requirements are now being revisited and possible solutions are for each problem provided.
Requirements
- Stolen vehicles.
- Compute parking cost.
- Granting access to uwc primacies.
- Identify owner violated traffic laws.
- Speed limit on road
Solutions
A solution to all of the problems or requirements listed above will be by using the Number plate recognition system. Number plate recognition system can also play a good role or can be helpful on stolen cars by matching the current license plate to the one that’s stolen, by doing so if they match then you found the stolen car or vehicle. For computing the parking cost, number plate recognition system can be useful here as well since it will recognize the vehicle entering and living the environment and they will calculate the difference in between for calculating the hours or total cost. This system can be helpful in finding the vehicles that violates traffic laws for example vehicles that go over limit, this number plate system can be used as to identify the vehicle so they can apply fines to it. This system can be helpful for traffic management as to identify the vehicle easily. This system could be very useful to the Uwc University in granting access, it will not be manipulated like their current system being used where it only accept the student card and does not even check if the person trying to grant access is the right person or is a student, all it does if student card is placed it grand access. The carelessly of this system could lead to problem of student stealing each other’s vehicles and still it will not be able to identify automatically all what will happen the vehicle will be repotted and securities have to deal with the matter themselves. But the proposed system that is number plate recognition will help solve all the problems listed about uwc. This system number plate recognition for uwc will identify the vehicle fully before granting access and will make sure know one lives uwc with someone’s vehicle.
Breaking down the problem into high level constituent parts
Capture Image
Here the digital camera is been used to capture the image. The image been taken is then passed to the processes for identification or recognition purposes. The sensor or vehicle detector is been used as to detect a vehicle approaching and activate the digital camera as to stand ready or stay ready to capture the image.
Gray Scale Conversion
From the 24-bit color value of each pixel (I, j) the R, G and B components are separated and the 8-bit gray value is provided. This reduces the color from the original image to gray. And the image is shown on figure 1.
Gray scale image
Figure 1 gray scale image.
Preprocessing
The major cause of failure in detecting the number plate from the vehicle is low quality of image. The preprocessing algorithm helps in improving the quality of the image or the plate image being inputted to the system.
This involves:
- Resize
- This involves resizing the image that is been taken for optimization purposes.
- Grayscale
Thresholding
This the quick way to convert the grayscale image into binary image. Here the grayscale image pixels are being separated or arranged. The grayscale image contains of pixels 0 to 255. So 0-127 is being initialized to 1 and 128-255 to 0 and at the end that’s why we say it is the quickest way to convert the image to binary image using the grayscale.
Binarization
It is difficult to work on an image that contains different kind of colors. It is simple to work with an image that contains two colors. Here the grayscale image is been converted into black and white (1 and 0) were black is for background and white is for text. Black is initialized to 1 and white to 0.
Number plate Detection (Find plate region)
This simply means that the number plate region needs to be identified or localized from the image been capture. After the image have been binarized then would be very easy in detecting the plate region since we only focusing on two colors black and white. The number plate would be easily identified from this point.
Here is another approach in detecting plate region and we use the Sobel operator to get edge information for the plate detection. In order to detect number plate from enhanced image we use a match filter method that will be able to remove most of the clutter part in image from the region of interest at the first stage. To be able to detect number plate or its place of region we first have to consider calculating the edge density and after that Gaussian kernel is being used for smoothing.
Match filter
This is the mixture of Gaussian functions emphasizes the constancy of intensity values within plate region along horizontal directions. Expected edge density along vertical directions at the middle of region are maxima and it decreases when approaching to plane borders. This function properly models the low edge densities above and below the car plate.
Number Plate Extraction
This is the process of extracting the plate from the image/ vehicle image and we going to use the morphological operation method in order to be able to extract the boundary of the number plate. This refers to a broad set of image processing operations that process images based on shapes. To find the boundary of the number plate region morphological closing is applied on edge image and a rectangular structuring element (SE) with 3×10 pixels is used. Morphological operation is used only to the candidate region to find plate boundary and to detect true region from the modified
Candidates, simple geometrical features such as shape, aspect ratio and size of region are used.
Character Segmentation
In license plate recognition or number plate recognition, when all license plate images are extracted from an image, one has to find all characters in the license plate and recognize them. Character segmentation is applied to the license plate in order to outline the individual characters. It affects greatly the accuracy of the recognition. If the contours of the characters are inaccurate, it might lead to errors in the recognition stage or even to failed recognition stage. This method use prior knowledge to help in character segmentation and prior knowledge includes information such as the size of the license plate or number plate, the size of the characters, the size of the interval between the characters and lastly the number of all characters within a license plate.
Character segmentation is based on projection method vertical and horizontal projection. Vertical and horizontal projections are used to perform the vertical and horizontal segmentation. Horizontal segmentation eliminates the unnecessary top and bottom parts of the license plate image. Vertical segmentation helps in dividing each character on the license plate image.
Feature Extraction
This plays an important role in the recognition system and effectively reduces the number of computations, this faster the recognition process. This uses the projection analysis method that is used to locate the features that have been analyzed from the vertical and horizontal segmentation. This also involves extracting the characters from the segmented images and this character are our features the main features that we wanted to extract out of the number plate in the first place. From this method on will be focusing on single character images that will be easily used for the recognition purposes.
Artificial Neural Network (Recognition of Characters)
The final and the most important stage of number plate recognition is character recognition. This involves recognizing the character images that have been extracted from the license plate image. There are many methods involved on character recognition but will be using neural network to recognize the characters on the plate image. Neural network contains of pattern mapping. Pattern mapping is a traditional method and before recognizing characters patterns need to be created. There are several types of patterns: images of each character, different sizes, different angles and projections of each character. Neural networks involve taking input computing calculations and giving the output. This used the segmented images as its input and it checks the output or valid output by training the network with different characters. Since South Africa contains only of numbers 0-9 and letters A-Z and all in all is 36 characters so the network will be trained through 36 characters in checking the errors and recognizing the characters.