TW584811B - Automatic structured plate image recognition system - Google Patents

Abstract

An automatic structured plate image recognition system comprises front sensing unit, rear sensing unit, front CCD camera, rear CCD camera and frame grabber and recognizes character in image after grabbing dynamic image. Based on the image data acquired from the front CCD camera and rear CCD camera, proceed plate positioning; divide character string of plate into isolated plate characters; further recognize each isolated character in sequence; finally generate a set of complete recognized plates as the basis of plate number of the vehicle and take each passing vehicle information as one corresponding passed vehicle information for vehicle management.

Description

584811 1 poor jr: V; A7 B7 PA0101? 0.REX-3 / 1 (Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention (/) [Technical Field] The present invention relates to an image-structured license plate automatic Identification system, which can be applied to public facilities such as roads, bridges, parking lots, institutions, specific places, etc. 5 Vehicle control, and can be used for manual toll vehicle control or smart transportation facility vehicle license plates for public facilities such as roads, bridges, and specific places For the purpose of identity control. [Prior technology] 10 At present, the charging mechanism of roads, bridges, parking lots and other facilities is used. Most of its image law enforcement methods are continuous recording of lane images with video cameras, and access to relevant videos when necessary. Bring the daytime scene at the time of the incident, or access the images with traditional cameras. None of the above image enforcement methods can achieve real-time vehicle control functions, and most of the vehicles in the images require manual interpretation to determine The number of the vehicle causes a waste of manpower. However, if the license plate recognition is performed by automatic license plate image recognition, The recognition method is to cut the license plate image of the image, and then binarize after the word cutting is completed, and then send the relevant characters to the image acquisition training classifier such as neural network to make the final recognition result of the license plate recognition. The result 20 is used as the final output. When the above methods are often affected by poor image quality or the license plate itself is fouled, more noise will be introduced due to image cutting and binarization, resulting in subsequent recognition quality cannot be improved and the original In addition, the current automatic license plate identification vehicle control system mechanism is only a single image source. If this single image source cannot be successfully taken, the paper standard is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) (Please read the notes on the back before filling this page)

584811 Α7 Β7 PA010130.REX-4 / 16 Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention) If the image conforms to the automatic identification of the license plate, the effect of the license plate image recognition is completely affected by the single image quality at that time Identification results of the license plate produced. The correlation between the image of the front and rear license plates when taking images is not used. Please refer to the Republic of China Patent Bulletin No. 123259 "Method and Device for Automatic Recognition of Vehicle License Plates" and the Handout Information of Robot and Industrial Automation Technology 2 of Fengjia University. The method of cutting out individual characters of the license plate string is to use the license plate range. The average gray value of the pixels in each row is its gray projection, and then the difference waveform of the gray projection is used as the word cutting method. When there are impurities in the license plate or the license plate is skewed, the peak of the gray projection difference The trough will become very unstable, which will result in the mistake of cutting characters. Please refer to the Republic of China Patent Bulletin No. 123259 "Automatic License Plate Recognition Method and Device" and Fengjia University's Robotics and Industrial Automation Technology 2 Lecture Materials. The way to identify characters is to use binary images to find the characters first. The upper, lower, left, and right boundaries of the character are divided into 5 * 5 blocks, and the Pixel in each block is checked to calculate the proportion of black dots, and then compared with the standard sample to obtain the identification result; when in When there are impurities in the character range, it is easy to cause misidentification. It can be seen that there are still many shortcomings in the above-mentioned customary methods and technologies. They are not a good designer and need to be improved. 20 In view of the various shortcomings derived from the above-mentioned conventional methods and technologies, the inventor of this case was eager to improve and innovate. After years of painstaking and meticulous research, he finally successfully developed this image-structured automatic license plate recognition system. -4- This paper size applies to Chinese National Standard (CNS) A4 (210 X 297 mm) (Please read the precautions on the back before filling this page)

* Ding · 11 1 ^ 1 n —.1 -n H II · -I II · ϋ «ϋ Hi In n I 584811

LW an Invention Description (> / _PA010130.rex-5/16 10 15 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 20 The purpose of this gross month is to provide a road, bridge, stop ==! Specific place The image-structured license plate from _ ^^ is used as a mechanism for vehicle access control. Private sale = Mingji-the purpose is to provide-a type of image-structured license plate from = _: first 'to omit the work of image binarization in image recognition, and Both the same quality license plate and the defaced license plate can achieve the effect of difficult to increase the Γ. Another of the present invention-g & post + # dynamic identification line, Lee W license plate 1 raised f4 image structured license plate The identification data integration mechanism is more effective for vehicle control. "Technical content" The image structure that can achieve the above-mentioned purpose of the invention is to 'identify and measure the self-image in the image after capturing the dynamic image according to the camera. Induction into the front camera =: On the face, Ya sends the sensor information to the camera to take a single front vehicle image; the rear camera sensor enters the vehicle in the rear camera shooting area and transmits the sensor message ㈣ =: ;! Backlight CC D camera capture-detail of single back photo; = take the scene data, edit the positioning of license plate M ^ Based on: card: set for image segmentation, the license plate string is divided into unique ::::: Recognition 2 County _ Zhuanyuan analysis of the structure of the material passing through the material t The paper size thorn f ϋ ® Home Standard (CNS) A4 specification ⑵0

297 mm) (Please read the notes on the back before filling out this page)

584811 7. 23 V. Description of the Invention (PA010130.REX-6 / 16 [Brief Description of the Drawings] Please refer to the following detailed description of a preferred embodiment of the present invention and the accompanying drawings, for further understanding of the present invention The technical content and its effects; 5 The drawings related to this embodiment are: Figure A is a flow block diagram of an image structured automatic license plate recognition system of the present invention; Figure B is a schematic view of an image structured automatic license plate recognition system of the present invention; 10 Head 2 is an example of the structural analysis of the image structured automatic license plate recognition system, the font and RCH; Figure 3 is the structure analysis of the imaged structured license plate automatic recognition system in four directions;. «彡 像 结构 式 转 _ 识 线The structure is a 15-square pixel array; and Figure 5 is the sum of the self-hearted pixel values of the image structure license plate. Fu Dao Ji [Major Symbols] 1 image data 2 license plate positioning 3 character cutting 31 character area cutting The 32-character cut paper size applies the Zhongguanjia standard _ (CNS) A4 specification (210 X F9_7 male ^-584811 S2., 7. 2, PA010130.REX-7 / 1 (A7 B7 V. Description of the invention (ο 4 Character recognition. 5 loses Characters of the license plate [preferred embodiment] Please refer to FIG. 1A and FIG. 1B, which are block diagrams and schematic diagrams of the image-structured automatic license plate recognition system provided by the present invention. A single front image of the vehicle, a single back image, and automatic identification of the vehicle license plate image. The identification method uses the structure analysis method to identify the license plate number, which is divided into three steps: Step 1: Locate the license plate 2 10 When the camera captures the image After the data 1, the vehicle image is processed by the horizontal gradient method. Because the license plate string has the characteristic of horizontal change, a signal that changes sharply at the position of the license plate. Therefore, after the gradient processing is performed on the entire vehicle image, The block scan looks for areas with sharp changes and concentration. In order to reduce the error, scan the vicinity of the areas that have been found. If you find 15 areas with real changes, use this area as the license plate position, and the license plate positioning 2 is completed. Step 2 : Character cutting 3, this step is divided into the following three small steps; A. character area cutting 31, in order Shadow's image cutting error is reduced, first make the image gradation clear with histogram equalization. B. Then use the combination of projection and contour tracing to find the character area, and slide the window (five or more points). Find the boundary candidate The paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) (Please read the precautions on the back before filling this page)

--Order -------- I Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs and Consumer Cooperatives 584811 92 · 7 · 23 V. Description of Invention (Heart) A7 B7 _PA01Q130, REX-fl / 1fi Employees of the Intellectual Property Bureau of the Ministry of Economic Affairs The consumer cooperative prints the points and records the position brothers, and then continues to the starting point of the next line, in which a sobel operation is performed on each point in the sliding window, and the one with the largest operation value is the boundary candidate point. First do a horizontal slide (from left to right, record all candidate points, the smallest is set to the left boundary. Then from 5 to left, record all candidate points, the largest is set to the right boundary.), And then do a vertical slide, from the top Next, record all candidate points 'the largest one' as the upper boundary. From bottom to top, all candidate points are recorded, and the smallest is set as the lower boundary. C. Character cutting 32, similar to the steps, using the 10 way of sliding window, horizontal and vertical sliding 'take the average value of all points in the window as the critical value, when the value of a point is greater than the critical value, the weight is strengthened , Sub-record position, # the number of points on the same axis, this point is the object point 'jump down-the line starts from the beginning, after all the lines have been scanned, the upper and lower boundaries of each word are found, and every 15 sub-scans will be found with a vertical scan Left and right borders. Step 3: Character recognition 4, the steps are: A. The processed image is an image of each character that has been cut out. B. Keep the gray level of the image, and do not perform the binarization first (see Figure 20). C. Scan the image in four directions (see Figure 3). D. Set up an array of five _ as a group (see E. Scan the image of D, the scanning method is as follows: 取得 First obtain the pixeHi of the background color and characters in the image. The method is shown in Figure -8 _ This Paper size applies Chinese National Standard (CNS) A4 specification (210 X 297 cm) " " " " " ---- 一 · ____ (Please read the precautions on the back before filling this page) Order- ---------. 584811 A7 PA010130.REX-9 / 1 (B7 V. Invention Description (7) (Please read the notes on the back before filling this page) ROI (Region of interest) , Take the maximum and minimum Pixel values in the ROI, and take the one-fifth of the maximum and minimum values as the Threshold of this image. (2) Take the horizontal scanning of No. 3 as an example, five The pixel array is scanned from 5 left to right. When the foremost pixel 1 encounters smaller than Threshold (that is, there is an object), record the position of the point and continue to move to the right by one space. (3) If the movement exceeds The three 袼 pixei values are smaller than each other, it is judged that this is the font itself. 10 ⑷If it is found that the pixel value is less than three cells during the movement, the pixel value is greater than Th. In the case of reshold, it is judged to be noise, shave it off, and re-record a position where the pixel value is smaller than Threshold. (5) When it is judged to be a font, the original pixel value is changed to 1 when it is moved, and it is judged to be non- In the font, change the pixel value to · 0 (see Figure 5), and then add them all up. (6) Do the same in the other three directions. F_ font structure to judge the consumption of employees of the Intellectual Property Bureau of the Ministry of Economic Affairs Cooperative printed (1) According to the coding in these four directions, the larger number represents the length of the object in that direction, and the characteristics of the day of the pen can be judged. 20 (2) If there are two or more directions, in a certain one When the position codes are very large, it means that there is a meeting point. By comparing the font characteristics of ⑴ and (2) with the font database that has been established beforehand, the fonts can be compared to output the license plate. Character 5. This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 11 84 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention (g) 10 Fruit 15 20

PA010130.REX-1 [Features and effects] When the image structure type automatic license plate technology provided by the present invention is compared with each other, it has the following advantages ... People "Value car combination ^ nt_rae_n Miscellaneous in gray scale cutting and characters The cutting is not affected by the roaring shirt on the license plate, so the process of eliminating noise is omitted. 2. The present invention not only omits the work of image binarization in image recognition, but also makes it difficult to cut the license plate character f. It is treated in the same way, so that the effect of improving the conversion rate can be achieved for different quality transfers and defaced license plates. 3. The present invention combines the mechanism of image structure analysis and font comparison, which will not be caused by the difference in the camera image capturing position. The size of the license plate is not fixed, the characters can be recognized, and the system has the ability to learn and update. 4. The present invention uses Histogram Equalization to first distinguish the characters in the license plate string. Then use the combination of projection and coffee tracing to find the character area. C0m0ur tradng can find the upper and lower contours of the license plate characters. When there are impurities in the license plate or the license plate is skewed, although Contour tracing cannot pass through characters, but it can find the relative minimum value of contour. It can also be used as a point cutoff for s0bd operation in local areas. This method can improve the conventional technology when it encounters impurities or the license plate is skewed. Disadvantages 5. The present invention does not use a binary character image in identifying characters, but directly processes graylevel grayscale images, first scanning the grayscale character images in four directions and scanning at the same time Stroke End of Record Character • 10- ..------- 1T ---- (Please read the notes on the back before filling this page)

n n n 1 I Box size of this paper applies to China National Standard (CNS) A4 (210 X 297 mm) 584811

On the one hand, it also filters unreasonable strokes (impurities) so as not to affect the structure of the character itself. After scanning, we get the stroke special symptoms of the character. Then we use these stroke characteristics and the standard character's The stroke characteristics are compared, and the recognition result transmitted to the delta element is carried out, and the method of the present invention does not produce different results for the size of the characters. The above detailed description is a specific description of a feasible embodiment of the present invention, but this embodiment is not intended to limit the patent scope of the present invention. Any equivalent implementation or change that does not depart from the technical spirit of the present invention should be included in Within the scope of the patent in this case. 〇 In summary, this case is not only technically innovative, but also enhances the above-mentioned multiple effects compared with conventional items. It should have fully met the requirements of statutory invention patents that are novel and progressive, and applied in accordance with the law. Approval of this invention patent application to encourage inventions to the utmost. ----—----- Order -----— (Please read the precautions on the back before filling out this page) The Intellectual Property Bureau of the Ministry of Economic Affairs's Consumer Cooperatives printed a modest piece of paper i 12 X 10 (2 Standard 4 i) A 5 N (C Standard

Claims (1)

584811 无 无-Α8 Β8 C8 m 6. Application for Patent Fan Pu (please read the note on the back of the threshold first and then fill out this page) 1. An image-structured automatic license plate recognition system, which will list each passing vehicle The front image, single back image, and automatic identification of the vehicle license plate image. The identification method is based on structural analysis to identify the five license plate numbers, which is divided into three steps: Step 1: Locate the license plate. After the camera captures the image data The vehicle image is processed using the horizontal gradient method. Because the license plate string has the characteristic of horizontal change, a signal that changes drastically will be generated at the position of the license plate. Therefore, after the gradient processing is performed on the entire vehicle image, it is searched in 10 blocks. In order to reduce the error, we will scan the vicinity of the found area to find the area that is really changing. Then use this area as the position of the license plate and locate the license plate. Step 2: Character cutting, this The steps are divided into the following three small steps: 15 A. Character area cutting, in order to make the image cutting error Less, first use histogram equalization to make the image gradation clear; printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economics. B. Then use the combination of projection and contour tracing to find the character area, and use the sliding window to find the boundary candidate points. And record the position, change to the starting point of the next line and continue, in which a Sobel operation is performed on each point in the 20 sliding window, with the maximum calculated value as the boundary candidate point; horizontal sliding first, then vertical sliding, Record all candidate points from top to bottom, the largest one will be set as the upper boundary, and then all the candidate points will be recorded from the bottom to the top, and the smallest one will be set as the lower boundary; This paper is in the standard _ 尺度 家 # 率 (CNS) Α4 Specifications (2⑽ × 297mm) 584811 A8 B8 C8 m --RAQ404 Sand AMD " ... 抡 / 46 VI. Application for patent scope (please read the note on the back to fill out this page} C. Character cutting, and step B Similarly, the sliding window method is also used to slide horizontally and vertically. The average value of all points in the window is used as the critical value. When the value of a certain point exceeds the critical value, the weight is strengthened, and Record the position. When the number of points is the same, this point is 5 points. Jump to the next line and start from the beginning. After all the lines have been scanned, find the upper and lower boundaries of each word. Vertical scanning will find each word. Step 3: Character recognition, the steps are: A. The processed image is the image of each word that has been cut out; 10 B. Keep the gray level of the image, do not binarize first; C · Right This image is scanned in four directions; D. Set up an array of five pixels as a group; E. Scan the image of D; F. Determine the font structure; 15 Generate font characteristics through the above three steps , Use the font characteristics to compare with the font database, and output the license plate characters after comparison. Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs and the Industrial and Consumer Cooperatives 2. The image structure type automatic license plate identification system described in item 1 of the scope of patent application, where the scanning method described in item E of step 3 is: (1) Obtain first The pixel values of the background color and the characters in the image are divided into 20 in the ROI. Do not take the maximum and minimum pixel values, and use one-fifth of the maximum and minimum values as the Threshold of this image; (2) five Each pixel array is scanned from the start scanning position to the end scanning position. When the front pixel 1 encounters a value smaller than Threshold, record the position of the point and continue to move one grid position to the next scanning position. The paper wave scale is applicable to China National rate (CNS) A4 specification (21®X: 2r7 mm> 584811 A8 38 C8 D8 DArvmmAMn VI. Patent application scope (3) If the pixel value is smaller than Threshold for three consecutive movements, it is judged to be a font Itself; (please _ read the note f on the back before filling in this page) ⑷If you find that the pixel value is less than three divisions larger than Threshold during the movement, it is judged to be noise, shave it off, and record it again 5 pixels whose values are smaller than Threshold; (5) When the font is judged, the original pixel value is changed to 1 when the font is judged, and the pixel value is changed to 0 when the font is judged to be non-font, and then all are added up. 3. The image structure type automatic license plate recognition system described in item 1 of the scope of patent application, wherein the judgment method described in item F of step 3 is: (1) According to the coding in these four directions, the larger number represents The length of the object in this direction can determine the characteristics of the character stroke; (2) If there are two directions or more, the code at a certain position is large, it means that there is a meeting point here. The image structure type automatic license plate recognition system is based on the structure analysis of each vehicle's single front image, single back image, and vehicle license plate image. The identification method is based on structural analysis. There are three steps to identify the license plate number: Step 1: Locate the license plate. After the camera captures the image data, use the horizontal gradient method to process the vehicle image. The license plate string has the characteristic of horizontal change, so there will be a drastic change in the position of the license plate. Therefore, after the gradient processing is performed on the entire vehicle image, the block scan is used to find the area with sharp changes and concentration. In order to reduce the error, Scan around the area you have found to find the real change script paper size, using the Chinese standard (CNS) A4 size (21®X297 mm) 584811 84 VI. Patent application difficulties D. Set up an array of five pixels; E. Scan the image of D; (Please read the note f on the back side before writing this page) (1) Get the image first The pixel values of the background color and the characters, respectively, take the maximum and minimum pixel values in the ROI 'as one fifth of the maximum and minimum values as the Threshold of this image; (2) five pixel arrays Scan from the start scan position to the end scan position. When the foremost pixel 1 encounters a value smaller than Threshold, record the position of the point and continue to move one grid to the next scan position; ⑶ If the movement is more than three pixels in a row If the value is smaller than Threshold, it is judged that it is the font itself. (4) If there is a situation where the pixel value is less than three grids larger than Threshold during the movement, it is judged to be noise, shave it off, and record it again. A position where the pixel value is smaller than Threshold; (5) When it is judged that it is a font, the original pixel value is changed to 1 when it is judged to be a side, and when it is judged that it is not a font, the pixel value is changed to 0, and then all are added up; F. Font structure judgment; (1) According to this The encoding of the direction. The larger number represents the length of the object in that direction, and the characteristics of the stroke can be judged. Printed by the Industrial and Commercial Consumer Cooperative of the A Bureau of the Ministry of Economic Affairs (2) If there are two or more directions, When the codes of a position are all 20, it means that there is a meeting point; through the above three steps, the font characteristics are generated, the font characteristics are compared with the font database, and the license plate after the comparison is output. Characters. This paper uses the Chinese national standard # 丰 （CNS） A4 (21 @ > < Sand 7mm)