introduction
This article refers to the address: http://
The research field of intelligent transportation systems is very broad, and the focus of each region is different. For example, the electronic toll collection system is the specific performance of ITS in the field of highway tolls. It can solve the “bottleneck†restriction of toll stations and better alleviate the problems of traffic congestion, queue waiting and environmental pollution at toll stations. In order to meet these needs, it is necessary to introduce automatic license plate recognition technology in the intelligent traffic Management System .
A car license is the clearest, most accurate and unique sign of a vehicle. The Vehicle License Plate Recognition (VLPR) system is a special computer vision system that automatically captures the dynamic data of the vehicle, effectively determines and extracts the image data of the license plate, and accurately identifies the vehicle license plate in real time. The character on it.
1 Implementation process of automatic license plate recognition system
A complete automatic license plate recognition system is mainly divided into image acquisition, image processing, license plate location, character segmentation, character recognition and other parts. Figure 1 shows the main working flow chart of an automatic license plate recognition system.
1.1 Image Acquisition and Processing
At present, image acquisition mainly uses a dedicated camera to connect the image acquisition card, or directly connected to a portable notebook for real-time image acquisition, and at the same time converts the analog signal into a digital signal. Image processing is mainly to enhance, recover, transform and other processes of the captured image, the purpose is to highlight the main features of the license plate, in order to better extract the license plate area.
1.2 License plate positioning
From the perspective of human vision, according to the characteristics of the character target area of ​​the license plate, based on the binarized image, the corresponding positioning features can be extracted. In essence, this is a problem of finding the optimal positioning parameter in the parameter space, which needs to be implemented by the optimization method. Generally, the projected area of ​​the edge image can be calculated, the peak and valley points are searched, the license plate position is roughly determined, the aspect ratio in the connected domain is calculated, the connected domain that is not in the range of the domain value is eliminated, and the license plate area is finally obtained. License plate location is a key and difficult point in the automatic identification system of vehicle license plates. Noise, complex background and other interferences in the actual image will increase the difficulty of positioning. The division of the vehicle license plate is a process of finding the area that best fits the license plate.
The license plate detection and positioning method includes image preprocessing, rough positioning of the vehicle license plate, and precise positioning of the vehicle license plate. Figure 2 shows the license plate location flow chart of the system.
The function of the image preprocessing part is to distinguish the information that has become an electrical signal, and at the same time remove the noise such as stains and blanks in the signal, and remove some non-essential signals according to certain criteria, and then the size, position and stroke thickness of the text. Etc. Normalize, and finally simplify the complexity of the judgment part.
The rough positioning part of the license plate will give a number of candidate license plate areas to be further judged and identified. If the number of candidate areas is zero, it means that the image does not contain the license plate, and the next step is not needed.
The precise positioning of the license plate is to classify the candidate area of ​​the license plate to determine which one is the real license plate area and give the coordinates of the license plate area.
1.3 Character segmentation of vehicle license plates
The character segmentation of the vehicle license plate divides the positioned license plate area into a plurality of sub-areas, each of which contains one character. The task of character segmentation is to separate each character in a multi-line or multi-character image from the entire image into a single character.
For general character recognition, the recognition process is to extract the feature describing the character from the input character pattern of the character to be recognized (sample), and then determine the mode category to which the sample belongs according to certain criteria. Therefore, character description, feature extraction and selection, classification decision, etc. are the three basic links of character recognition.
.
2 Identification of license plate characters
Character recognition is the most important part of the license plate recognition system. This part needs to identify and process the results obtained by image acquisition, image processing, license plate location and character segmentation to finally obtain the characters of the vehicle license plate.
The vehicle license plate character recognition method discussed in this paper is divided into character image preprocessing, feature extraction, classifier design and so on.
2.1 Character Image Preprocessing
Character image preprocessing is the processing of the input character image to make it into a specific standard form, making subsequent feature extraction and character recognition easier. Its main functions have two points: one is to eliminate the noise in the image, to correct the image break or stick phenomenon; the other is to make the transformed image relatively stable and easy to identify through various linear and nonlinear normalization methods.
2.2 Feature Extraction
The main purpose of feature extraction is to extract the essential features used to distinguish different categories from the original data. Because the applicability of different features is different, the feature characteristics extracted from characters with different effects are not the same. Therefore, it is difficult to adapt to the recognition of license plate characters affected by various conditions with a single feature. In addition, since the meanings expressed by different dimensions of different features are not the same, and the weights may be quite different, if the direct combination method is adopted, the features with larger weights will be dominant, and the weights will be ignored. Small features. To solve this problem, the feature vector normalization method or the weighting method can be used to combine the two features by weighting methods, so as to achieve the purpose of combining the two features.
2.3 classifier design
The classifier is to classify the identified objects into a certain category in a feature space. The basic approach is to determine a certain decision rule based on the sample training set so that the error rate caused by classifying the identified object according to the decision rule is minimized or minimized.
When using neural network as a classifier, it is necessary to have a certain training sample, and the number of samples should not be too small. However, in the experimental environment of this paper, there are fewer Chinese characters and English samples, and even some Chinese characters have only one English or Several samples, so can not guarantee the degree of training of the neural network. Therefore, this paper uses the template matching method. The template matching method is actually a distance classifier that uses multiple standard samples. The average sample method can usually be used to calculate the sample mean as a standard sample for each category, then calculate the distance between the sample to be identified and the standard sample, and finally select the standard sample with the smallest distance as the sample category to be identified.
The commonly used distance criteria are as follows;
(1) Minkowsky distance
This distance is a generalized representation of several distances:
(2) "City block" distance
That is, the block distance, which is a correction to the Manhattan distance, plus weights. which is:
(3) Euclidean distance
That is, the Euclidean distance is a special case where the Minkowsky distance is at λ=2, and the advantage is that each point is continuously different:
(4) Mahalanobis distance
That is, the Mahalanobis distance, which takes note of the statistical properties of the sample, and excludes the correlation effects between the samples. It can be expressed as:
This design uses the Euclidean distance. Because the Euclidean distance can only be calculated
This can reduce the calculation time.
3 Conclusion
The license plate recognition method mentioned in this paper has a good recognition effect, and can be improved for the phenomenon of leakage and recognition errors. The image brightness can also be analyzed during preprocessing, for images that are too bright or too dark. Different binarization strategies are adopted; the backtracking method can also be used to verify the accuracy of license plate location and character segmentation according to the result of character recognition; the character recognition part can increase the number of training samples of the character template, and the neural network is used as the classifier. Can improve the accuracy of character recognition.
Switch Mode Power Supply include Plug Type Power Supply,Desktop Type Power Adapter,Network Power Supply and 4 Spliters DC Power Supply etc. Switching power supply exclusive Patent : 12v dc power supply Multi-function power plug, it is changeable plug.Plastic housing use ultrasonic process, it is No need any screw.100% Enough Power, High efficiency, all above 80%.
Features:
Wide input range: 100-240V, 50/60Hz.
Exclusive Patent : Multi-function power plug,changeable plug.
Protections:short circuit, over current, overload,over voltage.
Being available for direct plug-in type,desktop type, wall-mounted type.
Customized Ac Plug , EU,UK,USA,AU Plug is available
100% full load burn-in test
Output voltage between 12VDC and 24VAC that can satisfy the demands for various types of power
Adopts No hole design, Plastic housing use ultrasonic process, it is No need any screw.
3 years warranty
Product application:
Application to CCTV System, Access control System, Alarm System.
Switch Mode Power Supply
Switch Mode Power Supply,Switch Mode Power Supply Design,Switch Mode Power Supply Schematic,Switch Mode Power Supply Advantages
Dongguan Xiaoerduo Electronics Co., Ltd. , https://www.steadysmps.com