1 Introduction
This article refers to the address: http://
With the rapid development of the national economy, more and more highways, urban roads and parking lots are being built, and the requirements for traffic control and safety management are also increasing. The intelligent transportation system has become a research field in the world of transportation. Frontier topics. Based on this development of license plate recognition lpr (license plate recognition)
The system is an important part of the intelligent transportation system and plays an important role in traffic management. The lpr system is mainly composed of three parts: license plate location, character segmentation, and character recognition. The success of license plate location directly affects the accuracy of license plate recognition and license plate recognition. The main license plate location methods: license plate location method based on gray image [1], license plate location method based on wavelet transform [3], morphological based license plate location method [4], neural network based license plate location method [7] , based on the support vector machine license plate positioning method [8] and so on. Although these algorithms have better recognition under certain conditions, the influence of some disturbance factors such as weather, background, license plate wear and image tilt can not fully meet the requirements of practical applications, and further research is necessary.
2 license plate positioning method
2.1 License plate positioning method based on gray image
A grayscale digital image is an image with only one sample color per pixel. Such images are usually displayed as grayscales from black to white. In order to facilitate the positioning of the license plate, the image is converted into a binary image, that is, an image of only two colors of black and white.
This method is applied to the license plate as follows: the license plate license has a relatively large contrast between the character and the background, and the horizontal gray scale corresponding to the license plate area changes frequently; and the license plate is generally hung on or near the bumper of the car, and is close to the image. In the lower part, the interference is generally less. According to the above characteristics, a first-order difference operation close to the horizontal direction is used to highlight an area where the gradation changes frequently. The formula for the first-order difference operation is:
From the fact that the license plate is a rectangle, we can judge that its corresponding horizontal projection is similar to the shape of the license plate. It is a relatively independent rectangular area. From the horizontal projection, the license plate position can be seen to correspond to the bottom of the figure. To the first larger peak, the license plate projection value region roughly corresponds to the projection value region contained between the valleys of the upper and lower neighborhoods of the above-mentioned peak, and the two valleys are substantially symmetrical with respect to the peak, the peak and the trough. The rate of change is large. The most important thing in this process is to determine which peak to choose, if the height between the two troughs of this peak is greater than a certain set value, and the width between the two troughs is roughly equal to the height of the license plate, It is determined that the area determined by it is the horizontal position of the license plate. For the vertical positioning algorithm of the license plate: In general, the background color of the license plate and the color of the character have a large contrast, and the change is relatively frequent in a relatively small range, and the vertical direction of the license plate is determined by this feature. This method has a good positioning for high quality images, but there are too many backgrounds in front of the car and the license plate in the image, which may lead to wrong license plate positioning.
2.2 License plate location method based on wavelet transform
Wavelet transform is an important branch of applied mathematics developed on the basis of Fourier analysis in the mid-to-late 1980s. Compared with the Fourier transform, the wavelet transform has strong local and frequency domain local analysis capabilities. The multi-scale refinement analysis of the signal is performed by the operation functions such as stretching and translation, and the narrow (empty) window is taken at the high frequency, and the wide (empty) window is taken at the low frequency, which overcomes the Fourier analysis. At the same time, the limitations of good local characteristics are obtained in the time domain and the frequency domain. In recent years, this method has been widely used in many engineering fields. The core of wavelet analysis is multi-resolution decomposition, and its good time-frequency characteristics make it an ideal tool for studying fine structures. In the aspect of license plate image processing, after high-pass filtering based on wavelet transform, the horizontal, vertical, æ’‡, and æº strokes of the license plate area become very simple and clear, and the noise interference in the picture is filtered out by the median filtering method in the traditional algorithm. In comparison, the wavelet transform enhances the pertinence of the processing, enhances useful information while reducing noise, and facilitates obtaining detailed information of the license plate area in subsequent programs. Ma Yongyi and Song Wei of Nanjing Aerospace Airlines applied wavelet transform directly to license plate location, and proposed a method for directly denoising grayscale images.
The method firstly designs a filter based on wavelet transform, then attenuates the low-frequency part llx after performing x-layer wavelet transform on the picture, and then performs inverse-wave transform to achieve high-pass filtering, filtering out the unevenness of the picture due to uneven illumination. The noise caused by the factors interferes with the license plate area. In the traditional license plate location algorithm, the image is first binarized. For the original grayscale image, there is a clear difference in brightness caused by uneven illumination. The brightness of the left half of the car is darker than the brightness of the right half of the building, and there is also a partial lack of light in the right half. In the case of direct averaging of the average grayscale threshold, it is easy to cause the license plate area to be unrecognizable due to the threshold selection being too high. If the threshold is deliberately lowered in the program, a lot of noise will be introduced, and the second is lost. The meaning of the use of value. If local threshold binarization is used, on the one hand, the workload for calculating the threshold is increased, the processing time is increased, and new boundary noise interference may be caused due to the division of the region. After using high-pass filtering based on wavelet transform, the picture effect is very ideal, which not only completely eliminates the influence of uneven illumination, but also makes the license plate area more prominent, which greatly improves the accuracy of positioning and searching. However, the reconstructed signal during noise reduction preprocessing will lose the original time domain characteristics.
2.3 Morphology-based license plate location method
Mathematical morphology is composed of a set of morphological algebraic operators. Its basic operations are four:
Expanding, corroding, opening and closing, they each have their own characteristics in binary images and grayscale images. Based on these basic operations, various mathematical morphology practical algorithms can be deduced and combined to analyze and process image shapes and structures, including image segmentation, feature extraction, boundary detection, image filtering, image enhancement and recovery. The mathematical morphology method uses a "probe" called a structural element to collect image information. As the probe moves continuously through the image, the relationship between the various parts of the image can be examined to understand the structural features of the image. Lu Yaqin and Yan Lingchao of Shanghai University proposed a morphological-based license plate location method based on morphology.
The specific steps of the method are as follows: First, the original image is binarized, and then the binarized image is filtered by using a 4×1 structural element to remove the noise of the image. The 16×9 structural element is used to expand the denoised image, and then the same size structural element is used for the corrosion operation to make the area where the license plate is located form a connected area. Then mark all the connected areas in the image, and calculate the circumscribed rectangle of each connected area, and use the prior knowledge of the license plate to determine the position of the license plate. The above-mentioned morphological processing based license plate positioning method can better determine the position of the license plate in the image, and the algorithm is simple and real-time. However, the use of structural elements that are too large or too small can not form a closed connected area of ​​the license plate area. Therefore the selection of structural elements is very important.
2.4 Vehicle license plate location method based on neural network
Bp neural network is one of the most in-depth and widely used models in artificial neural networks. It is an application model of bp (back propagation) learning algorithm in multi-layer feedforward networks.
The essence of the bp network is the multi-layer perceptron (mlp). For the commonly used three-layer (including input layer, hidden layer and output layer) networks, the first layer belongs to the input layer and accepts input vectors. The second layer belongs to the hidden layer, which is used for memory, increasing the adjustable parameters of the network to make the network output more accurate; the third layer belongs to the output layer and outputs the network result. The nodes between adjacent layers belong to a full connection, and the nodes between the same layers are not connected. In theory, any three-layer bp network can fit any nonlinear function as long as the nodes of the hidden layer are increased to a certain extent. The bp algorithm consists of forward propagation and back propagation. Forward propagation is the input signal passing from the input layer through the hidden layer to the output layer. If the output layer gets the desired output, the algorithm ends; otherwise, it goes to the back propagation. Backpropagation is to calculate the error signal (the difference between the sample output and the network output) in the reverse direction of the original connection path. The gradient descent method adjusts the weight of each layer of neurons to reduce the error signal until the error reaches the desired error.
This method mainly uses neural networks for some advantages of pattern recognition: first, it allows less understanding of the problem; second, it can achieve more complex partitions in the feature space. The vehicle license plate location method based on neural network is mainly divided into the following steps. Firstly, the bp network is trained, the license plate image in various cases is selected, and converted into a grayscale image, and the whole gray value of each image is used as a group of the network. Input vector, if it is a license car image, set its network output to high (0.9), otherwise low (0.1), repeat training until the desired effect is achieved. The actual image is then preprocessed: grayscale, histogram, filter filtering, enhancing the image and removing image noise. Finally, an m×n sliding window (m×n is set according to the aspect ratio of the actual license plate) is used to traverse the preprocessed image pixel by pixel. The data of the sub-image in the window is normalized and sent to the input end of the neural network. As an input vector, if the neural network output is high, it can be judged that there is a license plate at the position of the sliding window, otherwise there is no license plate. The use of neural networks for vehicle license plate location has the advantage of making full use of the neural network's adaptability, but this method also has the disadvantage that it takes some time to train the network. How to shorten the training time of the neural network, that is, to improve its convergence speed is a difficult point of research.
2.5 Positioning method based on support vector machine
Support vector machine (svm) is a new type of machine learning method proposed by vap2nik et al. in the early 1990s. It is mainly used to solve the pattern recognition problem under the condition of limited samples. This method can achieve good classification and promotion ability with few training samples. Due to the texture of the license plate area, looking for a good performance classifier highlights this texture feature and distinguishes it from other areas. Support vector machine (svm) is such a classification learning mechanism.
So for the research of license plate location, a positioning method based on support vector machine is proposed. Firstly, the image is divided into sub-blocks of n×n size, the gray-scale features of each sub-block are proposed, and the svm classifier is trained. Then, the trained classifier is used to classify the license plate sub-block and the non-licensing sub-block, and then use the mathematical form. Learning filtering and region merging; finally using the projection method to locate the license plate area. The experimental results show that the method can locate the license plate area better, but it is difficult to implement large-scale training samples due to the svm algorithm, and it is difficult to solve the multi-classification problem with svm. If this problem can be solved, the license plate location will be more accurate.
3 Conclusion
Based on the existing license plate recognition system, this paper gives a comprehensive overview of the domestic license plate location methods in recent years. License plate location is the key to license plate recognition. The above-mentioned license plate location method is very good under ideal conditions, but due to the complexity of the license plate background and the diversity of the license plate features, researching a more practical license plate location method is the next step. If we combine the characteristics of some of the above methods and gather the strengths of each individual, it is possible to study a better method of license plate location.
The multi-speed air supply mode can bring you a comfortable and cool breeze in various scenarios. The 180° wide-angle surrounds the healthy and refreshing natural wind, and the soft wind blows on the face, returning to the natural and refreshing. The fuselage is equipped with a lanyard, so you can take the wind with you anytime anywhere.
Neck Fan,Small Fan,Portable Mini Fan,Hanging Neck Fan
Guangdong Aiyimi Electronic Technology Co., Ltd. , https://www.nbminifan.com