JP6890427B2 - Car number recognition device - Google Patents

Description

The present invention relates to a vehicle number recognition device that photographs a vehicle with a TV camera and recognizes the characters on the license plate from the image.

Car number recognition has been popular for more than 20 years, and many devices are installed in various places. Until now, there were only four types of license plates: green letters on a white background, white letters on a green background, black letters on a yellow background, and yellow letters on a black background, but after 2017, design plates will be distributed for the Rugby World Cup. It is planned. In addition, design plates will be distributed for the 2020 Tokyo Olympics.

Many patents have been filed for car number recognition devices, but the corresponding patents for the design plates that will be provided in the future have not yet been published. Patent Document 1 is a method of cutting out a license plate with high accuracy, and there is no description about symbol correction. Patent Document 2 is a license plate recognition device (also known as a car number recognition device) for multiple lanes, and there is no description of symbol correction in this document either. There is Patent Document 3 regarding general character recognition. This conventional technique is a coping method when a character cannot be extracted neatly due to color shift or color blur of a character reading device such as OCR, and the color component or the brightness component of the outline of the character is set to the color of the center of the character. It replaces the component and the brightness component.

Japanese Unexamined Patent Publication No. 2008-108063 Japanese Unexamined Patent Publication No. 2016-957763 Japanese Unexamined Patent Publication No. 2008-269509

In the design plate, in the conventional character detection and character recognition methods, character recognition may be erroneous due to the influence of the design. In addition, in the design plate for the Rugby World Cup, the plate for business use has white letters on the conventional green background and only the plate frame on the white background turns green, making it impossible to judge for private use or business use by the conventional method.

When the pattern drawn on the license plate is in contact with the characters on the license plate, the target area for character recognition becomes an area containing noise, and there is a problem of erroneous recognition. In addition, there are two types of license plates, one for private use and the other for business use. However, if only the frame of the plate for business use is green, there is a problem that it cannot be determined that the conventional green background is for business use and the white background is for private use.

Therefore, in the present invention, it is possible to recognize the symbol plate with high accuracy and maintain high accuracy recognition.

In order to solve the above problems, the vehicle number recognition device according to the present invention has an imaging unit that captures a number plate, a plate detection unit that detects a region of the license plate, and a serial number in the region of the license plate. The sequence number detection unit to be detected, the peripheral position calculation unit that calculates the position of the character area around the sequence number from the size of the sequence number, and the color information of the character area of the detection result of the sequence number detection unit are detected. , The character color determination unit that determines the character color, the background color determination unit that determines the background color from the result of the character color determination unit, and the area obtained from the position calculated by the peripheral position calculation unit are the color images of the plate area. It was cut out after passing through a noise color detection unit that extracts color components other than the color of characters and a background, a color correction processing unit that corrects a region of the color component extracted by the noise color detection unit, and the color correction processing unit. It has a character recognition unit that recognizes characters using image data in the character area.

According to the present invention, it is possible to provide a vehicle number recognition device that recognizes characters on a design plate with high accuracy and maintains high accuracy recognition.

It is a block diagram explaining the 1st Embodiment of this invention. It is a figure explaining the internal structure of the symbol correction processing of this invention. It is a figure explaining the flow of the character color determination processing. It is a figure explaining the processing example in the H, S, I image of the character color judgment. It is a figure explaining the color of the license plate of a Japanese car. It is a figure explaining the flow of determining the color of the license plate of a Japanese car. It is a figure explaining the process of color correction using the use code as an example. It is a figure explaining the process which replaces the luminance information with the ambient information. It is a figure which shows the correction example of the use code area. Similarly, it is a correction example of the usage code area. It is a block diagram explaining the 2nd Example of this invention. It is a figure explaining the process of correcting other than a character color and a background color from the whole license plate. It is a figure explaining the determination process of a temporary number. It is a figure explaining the determination process of a delivery number. It is a block diagram explaining the 3rd Example of this invention. It is a figure explaining the flow of the symbol correction processing in a shading image. It is a figure which shows the symbol correction result in a shading image. It is a figure explaining the business number determination processing flow of the symbol number for Rugby World Cup. It is a figure explaining the whole flow of the vehicle number recognition processing in the case of having a business plate determination processing unit. It is a figure explaining the determination process of the above-mentioned business number. It is a figure explaining the determination process of the business number. It is a figure explaining the whole flow of the conventional car number recognition process. It is a figure explaining the flow of character recognition. It is a figure explaining the structure of the conventional car number recognition device. It is a figure explaining the type of the number of a Japanese car.

Prior to the description of the present invention, the outline of the conventional number recognition (Japanese Patent Laid-Open No. 6-215293) will be described. FIG. 22 shows the processing flow. In FIG. 22, since the vertical edge is strong in the vicinity of the plate of the car, the vertical edge is detected, this is used as a binary image, and the binary image is formed to obtain a plate shape candidate. In addition to this, various methods are open to the public. That is, 200 for creating a plate candidate image from an input image, 201 for cutting out a plate candidate, 202 for binarizing a plate area with an optimum threshold value (at this point, it can be known whether it is for private use or business use), and each object. 203 to cut out only the serial number by checking the size and the sequence, 204 to find the slope of the serial number and correct it so that the characters are upright, 205 to binarize again, and the serial number. It is composed of 206 to be detected and 207 to determine whether it is a medium-sized plate or a large-sized plate from the size of the characters of the serial number and calculate the surrounding character positions accordingly.
Once each region is obtained, an image of each region is cut out and recognized by a classifier (neural network, SVM, random forest, etc.). This example is just an example, and various other license plate recognition methods have been published.

FIG. 23 shows an example of recognizing the classification number (vehicle type code). Convert so that the character color is always white (the background is black) 210, set the first digit area 211, set the second digit area 212, and set the third digit area 213. If the area can be set, it is normalized to a preset size 214, and characters are recognized using this image data. Character recognition includes neural networks, SVM (Support Vector Machine), random forest, and deep learning, which has been attracting attention recently. In this recognition process, there is also a method of binarizing a predetermined area and recognizing it with a binary image, but in the present invention, a method of recognizing using a shade image is taken as an example.

FIG. 24 shows a hardware configuration that realizes such processing. The license plate recognition device 10 is provided with a plurality of image memories 2 for storing images captured from a camera 1 (color camera, monochrome camera, also known as an imaging unit), processes images in the image memory 2, and plate detection unit. 3 detects the plate, and the serial number detection unit 4 detects the serial number. The peripheral position calculation unit 5 determines whether it is a medium-sized plate or a large-sized plate from the size of the serial number, and calculates the position based on the coordinates of the character area around the serial number (Japanese plates have medium-sized and large-sized plates). , It is possible to calculate the area coordinates of the Land Transport Office code (Kanji), vehicle type code (small number), and usage code (Hiragana) according to the position of the serial number). The character recognition unit 6 recognizes characters using the image data of the cut out character area. The CPU 7 controls the overall flow. The communication control 8 transmits the recognition result, receives the parameter, and the like.

As shown in FIG. 25, Japanese plates have only four colors, one for private use of registered automobiles, one for commercial use, one for private use of light automobiles, and one for commercial use. When recognizing by shading, if the character color is darker than the background, it can be judged as private use, and if it is bright, it can be judged as business use.

With such a conventional configuration, when a symbol plate to be distributed in the future becomes a recognition target, the color component of the symbol becomes noise.

Hereinafter, examples of the present invention will be described with reference to FIG. The same numbers as in the conventional configuration of FIG. 22 have the same function. In the present invention, as shown in FIG. 1, a symbol correction processing unit 9 for correcting a symbol is provided by using the output of the serial number detection unit 4. As shown in FIG. 2, the symbol correction processing unit 9 includes a character color determination unit 20, a background color determination unit 21, a noise color detection unit 215, and a noise color correction processing unit 22.

The character color determination unit 20 processes using the binary image of the serial field number detected by the serial number detection unit 4 of FIG. The processing flow in the character determination unit 20 is shown in FIG. The processing flow of FIG. 3 includes a process 30 of converting the input color camera image data into H, S, and I images. The H, S, and I images represent Hue angle, Saturation, and Intensity, respectively. Similar effects can be obtained by using R, G, and B images, but this will be described by using H, S, and I images in the present invention. H, S, I images are also called H, S, V images, and V represents Value (intensity).

The H, S, I image is subjected to AND processing with the binary image of the serial number, respectively, to create an H, S, I image of the serial number region31. The process 32 of obtaining the frequency distribution (histogram) of the H, S, I images ANDed by the serial number is included. Further, it includes a process 33 for setting a threshold value determined for each color and a process 34 for determining a character color according to the threshold value.

Processing flow A specific example of FIG. 3 is shown in FIG. A serial number is detected for the plate area image 40, and AND processing 41 is performed on the binary image and the H, S, I images. An example of the H image is 42, and a histogram of the character area is displayed. Similarly, the example of the S image is 43, and the example of the I image is 44. Color determination is performed using these three histograms. In the example of FIG. 4, since the vehicle is for business use (the background is darker than the characters), it is sufficient if the histogram can be used to determine whether the characters are yellow or white.

When the character color determination threshold value is white, the threshold value of the H image is 24 to 105, and the threshold value of the S image is 0 to 90. In the case of green, the threshold value of the H image is 10 to 126, and the threshold value of the S image is 0 to 170. In the case of yellow, the threshold value of the H image is 50 to 150, and the threshold value of the S image is 0 to 224. In the case of black, the threshold value of the H image is 25 to 150, and the threshold value of the S image is 0 to 175. Looking at the distribution of the H and S images in FIG. 4, the peak of the H image is about 130 and the peak of the S image is about 110, and since it is known that it is for business use, it can be judged that the character color is yellow.

Note that these threshold values are examples and may be changed according to the target image. The H image is a circular exponent expressed from 0 to 360 degrees, but since an 8-bit image is used in image processing, it is converted to a value of 0 to 255 (it also changes depending on where the starting point of 0 degrees is determined). To do). Further, in the S image, the value of 0 to 1 is converted to 0 to 255. Since the I image is luminance information, it is 0 to 255. Further, the above threshold value is changed depending on the color sensitivity of the camera, which is an example.

Here, there are only four types of license plate colors for Japanese cars. When this is divided into plate size, private use, business use, registered car, and light car, as shown in Fig. 5, in the case of a medium-sized private car, since it is a registered car or a light car, the character color is green or black. .. Its background color is white or yellow. In the case of medium-sized vehicles for business use, the text color is white or yellow for registered vehicles or light vehicles. Its background color is green or black.

On the other hand, in the case of a large plate, if it is a private car, the character color is only green, and if it is for business use, it is white. That is, in the case of a large size, the character color can be specified if it is known whether it is for private use or for business use.

When this is shown in FIG. 6, it is determined whether it is large or medium size and 50 (this process can be executed by the serial number detection unit 4 in FIG. 1), and if it is large, it is determined whether it is for private use or business use and 51 (this process). The serial number detection unit 4 in Fig. 1 can determine whether the character color is brighter or darker than the background), white characters for business use, and green characters for private use. In the case of a medium-sized plate, it is determined whether it is for private use or business use 52, in the case of business use, the character color is determined 53, and it is determined whether it is white characters or yellow characters. In the case of private use, the character color is black or green, but if the character color can be determined from the H, S, I images, it is possible to know whether the character is green or black. When it is difficult to distinguish between the green and black colors in the H.S.I image, the background color is determined by the background color determination unit 21 54. As a result, if it is white, it is a private car of the registered car. If it is yellow, it will be a private car of a light car. By executing the above processing, the character color determination and the background color determination can be performed.

The large / medium size determination can be made when the character is detected by the serial number detection unit 4 in FIG. 1, and the judgment for private use and business use can also be made by the serial number detection unit 4 (the entire plate at the time of character detection). By finding the histogram of, it is possible to make a judgment for private use or business use depending on whether the frequency is in the dark side or the bright side. Even if the judgment is wrong, the serial number cannot be detected. Creates an inverted image and detects the sequence number, so if the sequence number can be detected, the plate size (large and medium size) and plate type (for private use and business use) can be obtained).

With the above processing, the character color determination can be executed and the background color can be specified.

In the area obtained by the peripheral position calculation unit 5, the noise color correction unit 22 of FIG. 2 detects the character color and the noise color other than the background color obtained by the character color determination unit 20 and the background color determination unit 21 in the noise color detection unit 215. It is a process of detecting with and replacing the H, S, I image of the area with the surrounding H, S, I image. At this time, the peripheral position calculation unit 5 in FIG. 1 calculates the area of the classification number (lowercase number), the land transportation branch code (kanji), and the usage code (hiragana) from the coordinates of the serial number. The noise color correction processing unit 22 corrects the area only in the above area (the entire plate image area may be corrected).

FIG. 7 shows an example of correction of the usage code area. Since this image 60 is a registered vehicle of a private car (the background is brighter than the characters and the background is not yellow), the background is green characters and white background, so colors other than green and white are detected and corrected. Histogram (density frequency distribution) of the entire H image of the entire usage code area is obtained 61, and the density part other than the character color (green) and background color (white) is detected by this frequency distribution 64, and this density value is used as the background H. Convert to the representative value (peak density or average density) of the image. Similarly, a histogram of the S image of the entire usage code is obtained 62, and a portion having a density other than the character color and the background color is detected 65 by this frequency distribution and converted into a representative value of the S image. The density conversion may be performed by using the density conversion process of image processing (a process of preparing an input density table and an output density table and converting the density according to the tables). For the H image of FIG. 7, the input and output densities may be the same for about 0 to 90 and about 170 to 255, and the output density of about 90 to 170 may be set as the representative density.

I image 63 is replaced with the peripheral density. As a process of replacing with the peripheral concentration, for example, there is a process shown in FIG. FIG. 8 shows the flow of the replacement process for the peripheral density of the I image. When the original image 80 is horizontally executed by the MAX filter (processing of replacing the maximum value of several pixels with the density of the central pixel) N times, the luminance information of the character area and the noise component can be replaced with other luminance information 81. On the other hand, by performing AND processing 83 between the exclusion area image 82 (the area other than the character color and the background color) in which the noise area of the original image is detected and the image 81 of the MAX filter, a density image to be corrected can be obtained 84 (the density replaced by white). Indicates. Black is the area not to be replaced). Further, an inverted image of the exclusion region 82 is created 85, AND processing of this image and the original image 80 is executed 86, and the density of the region other than the exclusion of the original image is detected 87. Then, the corrected image 89 can be created by performing the OR process 88 on the filled density image 84 and the unfilled image 87. In 89 of FIG. 8, the correction density seems to be emphasized, but the density is actually close to the background of the characters.

The corrected result is shown in FIG. A little light blue remains, but the pink color component has been completely removed 90. When this is schematically shown, the correction is as shown in FIG. When the original image 100 is corrected, an image like 101 is obtained. This enables character recognition without any problem. An example of conversion of the usage code is shown, but the land transportation branch code and classification number are also executed in the same process.

According to this embodiment, by cutting out a serial number, specifying a character color and a background color, and correcting other colors, there is an effect that even a license plate of a pattern can be recognized with high accuracy. Further, as a by-product of this embodiment, the color detection of the serial number and the color determination of the background color facilitate the determination of the light vehicle. That is, when the character color is black and the background area has a lot of yellow, it can be determined as a private light vehicle, and when the character color is yellow and the background area has a lot of black, it can be determined as a business light vehicle. Although it is difficult to judge a mini vehicle in a monochrome image (it can be judged by the combination of the usage code and the classification number, but if the classification number of the old car is 2 digits, it may not be possible to judge by this combination). Using a color image has the effect of making it possible to determine a light vehicle.

In the first embodiment, the process has been described on the premise that the symbol does not affect the cutting out of the serial number, but in reality, it is assumed that the symbol comes in the vicinity of the serial number. In this case, in the first embodiment, the detection of the serial number fails. If the detection of the serial number fails, the character color cannot be determined, and the pattern color cannot be corrected. Therefore, the second symbol correction processing unit 11 is provided by using the information of the plate detection unit 3 of FIG.

After the plate area is cut out by the second symbol correction processing unit 11 (result of the plate detection unit 3 in FIG. 1), Japanese cars basically have only green, black, white, and yellow color plates. , Green, black, white, and yellow are detected, and the H, S, and I images in that area are corrected. At this time, if the color component of the character color is corrected, the character detection is affected, so only the color component that is far from the character color and the background color is removed. If detailed color component removal as in Example 1 is not necessary and can be roughly corrected, there is no problem in detecting the sequence number because the sequence number is a large character (when a color very close to the character is near the character). It is not used because it makes character recognition difficult even for humans).

The flow of processing will be described with reference to FIG. The actual image is a color image, but since the color image cannot be displayed in the specification, it is represented by shading. A color image is converted into an H, S, I image 120 with respect to the plate region detected by the plate detection unit 3 of FIG. 11, and color components other than green, black, white, and yellow are detected 121.

This region is corrected by the representative value of the H image (representative value of the entire plate region) for the H image, the representative value of the S image (representative value of the entire plate region) for the S image, and the representative value of the I image (average) for the I image. The corrected image 122 can be created by replacing the peak value with a density brighter than the brightness), and the serial number can be detected. The representative values of H and S can be obtained by the average concentration or the peak concentration. This treatment treats the entire area of the plate.

According to this embodiment, even if there is a symbol in the vicinity of the serial number, the serial number can be detected correctly by correcting the color component.

In the case of general documents, it is possible to detect other than the character color and background color by setting the character color and background color in advance, and to correct the background color other than the character by the same correction process as above. ..

Further, as shown in FIG. 13, the red component was detected using the H, S, and I images of the image 130 of the entire plate (the shaded area that looks black is the red image), and 131 (the region slightly brighter than the background). This image is binarized and molded 132 (it can be molded by repeating expansion and contraction of the detected red component region in the plate region several times). By correcting the color components H, S, and I in this region in the same manner as in FIG. 12, an image like 133 can be obtained. As a result, numbers can be recognized even with temporary license plates (hereinafter referred to as temporary numbers). When the shape of this red component is slanted (a tentative number if it is a straight line that rises to the right in a straight line approximation of a binary image. Or, compare the horizontal coordinate position of the upper right of the binary image with the horizontal coordinate of the lower left, and lower left. If is small, it can be determined as a temporary number) temporary number. In addition, the character arrangement around the numbers is different between the normal plate and the temporary number. In FIG. 13, there is a place name on the left and a place name on the lower right, and in the case of a temporary number, it is possible to recognize the place name by recognizing the position (however, the place name on the normal plate is horizontal). Therefore, it is necessary to learn the classifier).

Further, as shown in FIG. 14, an image 141 detected in red from the plate area image 140 is created, and an image 142 obtained by binarizing the image 142 is created. (Hereinafter referred to as the forwarding number).

According to this embodiment, by detecting the red component and checking its shape, it is possible to determine whether it is a temporary number, a forwarding number, or a normal number, and by correcting the color component in the red region. , The serial number can be detected even with a temporary number, and the characters of the serial number can be recognized.

Although Examples 1 and 2 utilize correction in a color image, it is difficult to remove a color close to a character color (for example, green). Therefore, the shading correction process is used as the preprocessing for character recognition.

Specifically, as shown in FIG. 15, a third symbol correction processing unit 12 that corrects the shading of the area to be recognized from the information of the peripheral position calculation unit 5 is provided. The third symbol correction processing unit is composed of a region determination unit and a density correction unit.

After the character recognition area is obtained, the area determination unit executes a process of replacing the density value of the area with a density of 0 when the density is equal to or less than a predetermined density (a process of deleting a noise-like density). This density value is changed by the density correction unit to be recognized, and if it can be recognized, the flow ends. FIG. 16 shows a processing flow in the third symbol correction processing unit 12. 160 to loop 3 times. When the loop is the first time, the density below the reference density value (for example, 1/3 of the average density of the area) is converted to 0, the character is recognized by the character recognition unit 164, and the result is recognized by the recognition result storage unit 165. Store in. In the second loop, a concentration less than twice the reference concentration value is converted to a concentration of 0, and 162 is processed in the same manner. The loop 3 times converts the concentration less than 3 times the reference concentration to 0 163.

When the loop ends, the recognition result adoption section compares the recognition result obtained for each loop with the certainty (evaluation value), and adopts the recognition result with the highest certainty (after character recognition, the certainty is equal to or higher than the predetermined value). Then you can exit the loop and end).

By recognizing while removing the density of the noise component in this way, as shown in FIG. 17, since there is noise at first with respect to the original image 170, the character recognition area is large 171 and cannot be recognized, but the density value is deleted. As a result, the character recognition area is set in the appropriate area, so that the correct answer can be obtained 172. The point of interest of this method is that the color component of the pattern is darker than the character density when expressed in shades. If there is a pattern with a density brighter than the character color, it is difficult to correct it.

According to this embodiment, even if there is a color pattern that cannot be completely corrected by the color correction processing, it can be recognized while deleting the noise image by the shading image correction processing, so that a high recognition rate can be obtained. The number of loop processes and the magnification of the density determined as noise may be changed as appropriate.

In the design license plate for Rugby World Cup, the background of the business plate is not green, the background is the same white as the private car, and only the edge drawn on the plate is green. In this case, it would be wrong to use the judgment of whether it is for private use or business use in FIG. Therefore, it is necessary to change the judgment for business. When it is determined that the flow of FIG. 18 is for private use, the business determination processing unit 180 of the rugby symbol is added. That is, the specific processing flow is as shown in FIG. The business judgment processing unit 180 determines whether it is for private use or business use, stores the result in the CPU via the bus, and transmits the result together with the plate recognition result in the communication unit. As shown in FIG. 20, in the business judgment process 180, 190 for detecting the serial number of the plate and medium / large size can be determined at the time of detecting the serial number, so that the rectangular coordinates of the plate frame can be determined from the character pitch of the serial number and the like. From 191 which can be calculated, a region is set with a width of several pixels inside the rectangular coordinates of this plate frame, and the color component of this region is obtained. If the color of this area is close to green, it is judged to be a business number. Alternatively, the ratio of the calculated area of the plate frame to the area of the white background is obtained, and if the white background is equal to or less than a predetermined value, it can be determined to be a business number. A more detailed version of the flow of FIG. 20 is shown in FIG. In FIG. 21, the serial number detection 201 is executed by the serial number detection unit, and the plate frame position calculation 202 and thereafter are processed by the commercial plate determination unit of FIG.

According to this embodiment, even a special business plate of a design plate for the Rugby World Cup can be judged for private use or business use.

1 Camera 2 Image memory 3 Plate detection unit 4 Serial number detection unit 5 Peripheral position calculation unit 6 Character recognition unit 9 Symbol correction processing unit 11 Second symbol correction processing unit 12 Third symbol correction processing unit

Claims (8)

An imaging unit that shoots license plates and
A plate detection unit that detects the license plate region and
A serial number detection unit that detects serial numbers in the license plate region, and
A peripheral position calculation unit that calculates the position of the character area around the serial number from the size of the serial number, and
The symbol correction processing unit for performing symbol correction using the output of the sequence number detection unit and the peripheral position calculation output unit,
A vehicle number recognition device having a character recognition unit that recognizes characters using image data of a character area cut out after passing through the design correction processing unit.
The symbol correction processing unit detects the color information of the character area of the detection result of the serial number detection unit, and determines the character color, and the character color determination unit.
A background color determination unit that determines the background color from the results of the character color determination unit, and
For the area obtained from the position calculated by the peripheral position calculation unit, a noise color detection unit that extracts color components other than the character and background colors in the color image of the plate area, and
Have a, a color correction processing unit for correcting the region of the extracted color component in the noise color detection unit,
In the color correction processing unit, the color component in the plate region replaces the region other than the serial number with the average value of the color component or the color component having the highest frequency of the color component, and the luminance component in the plate region is other than the serial number. vehicle number recognition device average value or frequency of the color component is characterized Rukoto replaced highest luminance value regions of. The vehicle number recognition device according to claim 1.
The color correction processing unit is a vehicle number recognition device characterized in that it performs a process of propagating brightness information in a region other than the serial number to replace the brightness component. The vehicle number recognition device according to claim 1 or 2.
The imaging unit is a color camera, and the noise color detecting unit detects a color component other than black, green, white, yellow, or a plurality of colors of the entire plate region detected by the plate detecting unit. Vehicle number recognition device. The vehicle number recognition device according to claim 1.
A vehicle number recognition device including a second symbol correction processing unit that performs symbol correction after cutting out the plate region by the plate detection unit. The vehicle number recognition device according to claim 4.
The imaging unit is a color camera.
The second symbol correction processing unit includes a red detection processing unit that detects a red component and a red detection processing unit.
A red shape determination unit that determines the shape of the red region,
A vehicle number recognition device characterized by having a color correction processing unit that corrects a color component in a red region when the red shape is a diagonal line. The vehicle number recognition device according to claim 1 or 4.
A recognition result storage unit that stores the character recognition result of the character recognition unit,
A recognition result adoption unit that compares the character recognition result with a recognition evaluation value that evaluates the character recognition degree and adopts the recognition result having a high character recognition degree.
It is provided with a third symbol correction processing unit that corrects the shading of the area recognized from the information of the peripheral position calculation unit.
The third symbol correction processing unit executes density conversion to brighten the characters and darken the background of the image in the area obtained by the peripheral position calculation unit, and performs density conversion to obtain a start point of density and a density when recognizable characters are present. The area determination unit for finding the end point and
A vehicle number recognition device characterized by having a density correction unit that gradually changes the minimum value of a density that can recognize characters. The vehicle number recognition device according to claim 1 or 4.
Equipped with a business judgment processing unit that judges that it is a business plate
The business determination processing unit includes a plate position calculation unit that calculates the coordinates of the plate frame determined from the position of the serial number, and the plate position calculation unit.
A vehicle number recognition device comprising a business determination processing unit that sets an area inside the plate frame and determines that the area is a business plate when the color component of the area is green. The vehicle number recognition device according to claim 1 or 4.
Equipped with a business judgment processing unit that judges that it is a business plate
The business determination processing unit includes a plate position calculation unit that calculates the coordinates of the plate frame determined from the position of the serial number, and the plate position calculation unit.
Having a business determination unit that determines that the plate is a business plate when the ratio of the plate frame area obtained by the plate position calculation unit to the area of the white area in the plate frame is smaller than a predetermined ratio. A featured vehicle number recognition device.

Images