Intelligent transportation system is a hot research field, which has received more and more attention. License plate recognition system (LPR) is an important application of computer vision and pattern recognition technology in the field of intelligent transportation, including license plate positioning, license plate character segmentation, and character recognition. The license plate positioning is a key step in the entire system.
The current license plate positioning methods mainly include:
(1) Based on the method of Hough transform, it is analyzed that the license plate has an obvious rectangular border, and the location boundary is detected by Hough transform.
(2) Based on the edge detection method, the rich feature of the license plate character edge is used, combined with mathematical morphology or area growth method to achieve license plate positioning.
(3) Based on the neural network method, the neural network is trained using the color or texture features of the image, and then the trained classifier is used to classify each pixel of the image, and then the classification results are synthesized to obtain accurate positioning of the license plate. However, due to uneven lighting, pollution and other factors, the boundary of the license plate area may not be obvious or there may be multiple interference areas, thereby increasing the difficulty of accurate positioning.
To improve the accuracy of license plate positioning, the information provided by him should be fully utilized to highlight the license plate area and suppress the non- license plate area. The license plate area is rich in texture, looking for a classifier with good performance, highlighting this texture feature, which distinguishes it from other areas. Support Vector Machine (SVM) is just such a classification learning mechanism, built on the Structural Risk MinimizaTIon (SRM) criterion, and has been used in text recognition, face recognition, texture classification and other pattern recognition The field has succeeded.
In this paper, the SVM mechanism is used to automatically locate the license plate area. First, each training image is divided into several N & TImes; N-sized image sub-blocks, and each block is labeled as two types of license plate and non- license plate areas, and the features of the sub-block image are extracted Vector training SVM classifier; then use the classifier to classify each pixel in the test image, and finally achieve the positioning of the license plate area through post-processing combined with the prior knowledge of the license plate.
2 SVM principle
SVM constructs the optimal hyperplane based on the SRM criterion to maximize the interval between each type of data while keeping the classification error as small as possible. Cover theorem states that a complex pattern recognition classification problem is more easily linearly separable in low-dimensional space than in low-dimensional space. In fact, SVM realizes the idea that the vector x is mapped to a high-dimensional feature space through some kind of pre-selected nonlinear mapping, and then the optimal classification hyperplane is constructed in this space.
For two types of pattern classification problems, in the case of non-linear separability, through a non-linear transformation φ: x → φ (x), map the given pattern data to a high-dimensional feature space, and then construct a classification hyperplane, representing For decision-making:
Considering that both types of samples should have a certain distance from the decision surface, the decision surface should satisfy the inequality constraints:
There is no hyperplane that fully satisfies equation (2). Considering that there are some samples that cannot be correctly classified by the decision surface, the relaxation variable ξi (≥0) is introduced, and the constraint condition (2) becomes:
There are more than one hyperplane that meets the requirements. Finding the optimal hyperplane can be attributed to the quadratic programming problem:
Among them, C is called a penalty factor, and C can be used to make a compromise between the generalization ability and the error rate of the classifier. Using the Lagrangian function to solve the dual form of the optimization problem (4), the maximizing function:
Solve equation (5) to get ai, substitute equation (7) to determine ω, and the classification function can be expressed as:
3 SVM positioning license plate area
Accurate positioning of the license plate area is a non-linear separable pattern classification problem.
Keywords: Vector machine license plate positioning
3.1 Feature extraction
Using the SVM's own structure can achieve effective feature extraction, choose to directly extract pixel gray features. The pixels of the image are not isolated, there is a correlation between each other, which reflects a texture. The gray value of some specific pixels can be extracted as the feature of the entire image, while reducing the amount of calculation. First cut each image into several N & TImes; N sub-blocks, and then mark each sub-block as two types: license plate area (+1) and non-license plate area (-1), and then use the word "m" shown in Figure 1 The model model extracts the pixel gray value (the shadow in the figure is the pixel to be extracted). In this way, the feature dimension of each sub-picture is reduced from N & TImes; N to 4N-3, which improves the training and classification speed.
Description
✿Service Life
Mechanical ≥ 1000000cycles
Electrical ≥ 50000cycles(Degree of protection againstDepend on model No.
✿Approvals
UL CUL VDE TUV CE DEMKO CQC
✿ Material
Made of high quality plastic and metal, rust resistance and corrosion resistance, durable enough for you to hanging items, So the product appearance is exquisite, perfect workmanship.
✿ Rating
You can meet the different RATING daily needs. Please pay attention to the model of the switch before purchase, to ensure that you purchase the same as you need.
✿ Vairous Sizes
Actuator Action is momentary and Actuator Type is long straight hinge lever. Switch Body Size as shown in the picture.So different sizes can meet all your daily different needs.
✿ Wide Application
Home appliance: micro oven, electric cooker, washing machine, electric heater, warmer, water fountain and so on.
Commercial appliance: Vending machine, electric toy, electric tools, duplicating machine and so on.
Machinery: Transport machinery, printing machinery, textile machinery and so on.
Basic Micro Switch,Micro Power Switch,Waterproof Micro Switch,Micro Switch Limit Switch
Ningbo Jialin Electronics Co.,Ltd , https://www.donghai-switch.com