JP2006018339A - License plate recognition device and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a license plate recognition device capable of recognizing number plates for conventional two incoming and outgoing license plate recognition devices by one device in alternate passage roads whose ongoing directions are switched. <P>SOLUTION: This license plate recognition device is provided with a camera 1 for license number recognition for photographing the image of a vehicle and an image processor 2 having an image processor 13 for processing an image photographed by the camera 1 for license number recognition. A template 46 is calculated from an inter-image projection difference between two frames photographed by the camera 1 for license number recognition by the image processor 2, and a region matched with the template 46 is retrieved from the inter-image projection difference of the two frames after the lapse of a time, and a moving vector 50 is calculated from the region matched with the template 46, and the front license plate recognition processing and the back license plate recognition processing are switched. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a license plate recognition apparatus using a camera installed on a road or the like, and more particularly a license plate recognition apparatus suitable for recognizing a license plate of a vehicle even when the traveling direction is changed on an alternately passing road. About.

  Systems that automatically recognize license plates, so-called vehicle number recognition devices, are installed in various locations. This vehicle number recognition device detects the number plate of the same number at a plurality of points, measures the travel time of the vehicle that has traveled between the points, and displays it on the display panel, and matches the number of the number plate registered in advance It is used to detect vehicles such as stolen vehicles and criminal vehicles.

  As such a vehicle number recognition device, there is one using a neural network as described in [Patent Document 1].

  By the way, in order to efficiently operate the road, there is an alternate road (also referred to as a reversible lane) that switches between up and down in order to change the number of up lanes and the number of down lanes according to time zones. Since the recognition algorithm of the front license plate of the vehicle and the license plate of the rear surface of the vehicle are different in the recognition of the vehicle number on such an alternate road, conventionally two vehicle number recognition devices for up and down use are provided on the alternate road. When installed, when the direction of travel changes, two vehicle number recognition devices are switched in accordance with a switching instruction from the user to recognize the license plate of the vehicle.

JP-A-6-215293

  The above-mentioned conventional technology has a problem that since the vehicle number recognition device cannot be switched unless the user gives a switching instruction each time the traveling direction changes, proper license plate recognition cannot be performed unless the user gives a switching instruction. there were.

  An object of the present invention is to provide a license plate recognition apparatus and method that can automatically recognize a license plate on an alternate road by detecting the traveling direction of the vehicle with a single vehicle number recognition device.

  Another object of the present invention is to provide a license plate recognition apparatus and method that can be installed without distinguishing between vehicle number recognition on the front license plate and vehicle number recognition on the rear license plate.

  In order to achieve the above object, a license plate recognition apparatus and method according to the present invention captures a plurality of frames of an image of a vehicle with a vehicle number recognition camera, and performs a second process between two frame image projections captured by the image processing apparatus. A template is calculated from one image projection difference, a second image projection difference between the image projections of two frames after the elapse of time is calculated from the two frames, and the calculated second image projection difference and the pattern matching of the template are calculated. Thus, a region that matches the template is searched, a movement vector is calculated from the region that matches the template, and the front license plate recognition processing and the rear license plate recognition processing are switched.

  According to the present invention, the up and down license plate recognition processing can be realized by one license plate recognition processing device. Moreover, it can install without distinguishing the vehicle number recognition of a front license plate or the vehicle number recognition of a rear license plate.

  An embodiment of the present invention will be described with reference to FIGS. FIG. 1 is a configuration diagram of a license plate recognition device, FIG. 2 is a configuration diagram of an image processing device, and FIG. 3 is a flowchart for automatically switching between recognition processing of a front license plate and a rear license plate.

  As shown in FIG. 1, the license plate recognition device is a vehicle for recognizing a vehicle number for photographing a vehicle traveling on an alternate road 5 (also referred to as a reversible lane 5) that changes the number of up and down lanes according to time zones. 1, an image processing apparatus 2 that inputs an image captured by a car number recognition camera 1 and performs image processing, and a central processing unit 3 that is connected to the image processing apparatus 2 via a network 4. Although FIG. 1 shows an example in which the vehicle number recognition camera 1 is provided on the alternating traffic road 5, the vehicle number recognition camera 1 is also installed on the road 6 where the vehicle travels in one direction. The field of view of the vehicle number recognition camera 1 varies depending on the type of camera, but generally the area for character recognition is about 2 to 3 meters wide and about 2 to 3 meters high, and depending on the time zone, the direction of up and down The vehicle is installed so that it can photograph the alternate traffic road 5 for changing the number of lanes in the direction, and the front image of the vehicle is photographed as an image 21 shown in FIG. 4, and the rear image of the vehicle is an image shown in FIG. 22 is taken. Usually, as shown in FIG. 4 and FIG.

  As shown in FIG. 2, the image processing apparatus 2 includes an image processing unit 10 and a control unit 16, and the image processing unit 10 and the control unit 16 are connected by a bus 20. The image processing unit 10 receives an image from the vehicle number recognition camera 1, stores an analog signal into a digital signal converted by the A / D converter 11 and A / D converter 11, and performs image processing. An image memory 12 for storing the image processed by the processor 13, an image processing processor 13 for processing the image stored in the image memory 12, and an image stored in the image memory 12 and processed by the image processing processor 13 are required. The D / A converter 14 converts a digital signal into an analog signal for display in response to the signal, and the monitor 15 displays the image signal converted by the D / A converter 14.

  The control unit 16 includes a CPU 17 that controls the image processing unit 10 and the communication unit, a RAM 18 that stores programs, and a communication unit 19 that transmits and receives data to and from the central processing unit 3 via the network 4. In the central processing unit 3, the image processing result transmitted via the communication unit 19 and the network 4 is accumulated, and the travel time of the vehicle traveling between the points is measured. The detected vehicle is detected and a stolen vehicle or a criminal vehicle is detected.

  Recognition processing and switching between the front license plate and the rear license plate of the license plate recognition apparatus configured as described above are performed as shown in the flowchart of FIG.

  In step 31, each area is set. As shown in FIG. 6, when the horizontal axis of the input image 40 from the vehicle number recognition camera 1 is the X axis, the vertical axis is the Y axis, and the size of the input image 40 is X0, Y0, Y0 on the Y axis. A switching determination area 41 between the front license plate and the rear license plate and a pattern matching search area 42 are set around / 2. Also, a template area to be described later is set.

  In step 32, an inter-frame projection difference feature calculation process is performed. As shown in FIG. 7, the first frame image projection 43 and the second frame image projection 44 taken by the car number recognition camera 1 are calculated, and the first frame image projection 43 and second frame image projection are calculated. 26 is calculated. An area having a large difference in the density value distribution of the image projection is calculated from the projection difference 45 between the frames, and the area is registered as the template 46.

  In step 33, pattern matching is performed using the template 46. In the pattern matching, as shown in FIG. 8, a projection difference 48 between the image projection 44 of the second frame photographed by the vehicle number recognition camera 1 and the image projection 47 of the third frame is calculated. The calculated projection difference 48 and the projection difference of the template 46 are compared, and an area 49 where the projections overlap and the difference between the two projection differences is the smallest is searched. The region 49 is called a matched region.

  Although the first, second, and third frame image projections have been described, these frames can be taken in time series. For example, the first and third frames can be used to register a template, and the fifth and seventh frames can be used for pattern matching.

When the matched area is searched, the movement vector is calculated and added in step 34. As shown in FIG. 9, a specific position of the registered template 46, for example, a start point position that is the uppermost part of the template 46 and a position corresponding to the specific position of the registered template 46 in the matched region 49 are calculated. . From the specific position of the region 49 that matches the specific position of the calculated template 46, the movement direction and the movement distance are calculated to calculate the movement vector 50. If the downward direction is positive in FIG. 9 and the matched region 49 is below the template 46, the movement vector 50 is positive and the vehicle is judged to be passing forward. .

  If the specific position of the matched region 49 is above the specific position of the template 46, the movement vector 50 is negative, and it is determined that the vehicle is shooting backward, that is, the rear side of the vehicle.

  The movement vector calculated in this way is associated with the vehicle passing time, transmitted to the central processing unit 3 via the network 4, and stored. FIG. 10 shows an example when the movement vector addition process is performed, with the horizontal axis representing time and the vertical axis representing the accumulated value of the movement vectors.

  Reference numerals 56 and 55 in FIG. 10 are a positive switching determination threshold value and a negative switching determination threshold value for performing switching determination between the front license plate and the rear license plate, respectively. In the example shown in FIG. 10, the accumulated value of the movement vector is reset to 0 at time 0 and the forward vehicle continuously passes, so the accumulated value increases in a positive region. The horizontal portion 51 means that the vehicle is not passing through the portion where the cumulative value does not change. If the cumulative value reaches the positive switching determination threshold value 52 during the rear license plate recognition process, it means that a forward-facing vehicle has passed, and the cumulative value is reset by switching to the front license plate recognition process. . After that, since the forward-looking vehicle passes, the accumulated value increases in the positive area and exceeds the positive switching determination threshold value 56. In this case, processing for not adding the movement vector is performed, and the accumulated value is switched to the positive value. The determination threshold 56 is a constant 53. As described above, when the cumulative value in the same direction exceeds the switching determination threshold after switching to the front license plate recognition process or the rear license plate recognition process, the movement vector is not added and the cumulative value of the movement vector is positive. The transition is made within the range of the switching determination threshold 56 and the negative switching determination threshold 55.

  When a backward-facing vehicle passes, the cumulative value decreases, and when the negative switching determination threshold 55 is reached 54, it is determined that the backward-facing vehicle is passing and is switched to the rear license plate recognition process and accumulated. The value is reset.

  In step 35, as described above, it is determined whether the current license plate recognition process is the front license plate recognition process or the rear license plate recognition process, and it is determined whether the positive switch determination threshold value 56 has been reached or the negative switch determination threshold value 55 has been reached. To perform recognition direction switching determination. If it is determined that the recognition direction switching determination is forward, front license plate recognition processing is performed in step 36, and if the recognition direction switching determination is determined rearward, rear license plate recognition processing is performed in step 37. Here, the method of using the accumulated value for the recognition direction switching determination has been described. However, when the vehicle passes in the same direction continuously for a unit time, the recognition direction switching determination in that direction may be performed. .

The license plate recognition process is performed as follows. Actually, since the character area cannot be accurately specified due to the bending or bending of the plate, first, character recognition processing from the grayscale image is performed. In the character recognition processing, (1) background difference processing, (2) detailed coordinate extraction of the character area,
(3) Gradation feature extraction and (4) recognition are performed in this order.

  (1) In the background difference processing, for example, when the character of the processing target image is darker than the background, the local maximum value filter is executed several times so that it can cope with the brightness change of the plate and the color change of the plate. Similarly, a local minimum value filter process is executed several times to create a background image, and a processing target image is subtracted from the background image to create a background difference image in which only character line image information is extracted.

  (2) In the detailed coordinate extraction of the character area, if the background difference image is recognized as it is, misrecognition is likely to occur due to character misalignment. Calculate the cumulative projection distribution of each density, determine the coordinates where this frequency exceeds a predetermined threshold for each direction, determine the coordinates of the character area, and use the background difference image for detailed coordinates of the kanji area for recognition Is calculated.

  (3) In the gradation feature extraction, since the character area is small, all the density data of the detailed area is used as the feature quantity, or m × n feature quantities are extracted from the detailed area image (mosaicization), etc. After the detailed coordinates of the area are obtained as an area, a feature amount for recognition is obtained.

  (4) In recognition, if density feature value data is obtained, learning and recognition processing are performed in a neural network. Thereby, the land transportation branch office code, the vehicle type code, and the use code are recognized from the serial number in the photographed plate and the coordinates of the extracted serial number.

  According to the present embodiment, the up and down license plate recognition processing can be realized by one license plate recognition processing device. In addition, the same effect can be obtained even when vehicles pass in opposite directions, such as a parking lot entrance / exit, as well as alternating roads.

  Further, in this embodiment, since the front license plate and the rear license plate can be recognized by one license plate recognition processing device, the vehicle number recognition of the front license plate or the vehicle of the rear license plate when installed on a general road. It can be installed without distinguishing the number recognition. In this case, the front license plate recognition and the rear license plate recognition are automatically switched even on roads that have alternated on one side due to temporarily overrunning vehicles such as overtaking vehicles from the opposite lane, emergency vehicles, or traffic restrictions. We can recognize the license plate.

  Further, according to the present embodiment, an area having a large density difference distribution among the projection differences between the two frames is used as a template, and an area matching the template is searched from the projection difference between the two frames photographed thereafter. Since the movement vector is calculated, it is possible to recognize the license plate of the vehicle that has entered the license plate within the field of view of the camera regardless of the traveling direction of the vehicle.

  FIG. 11 shows another embodiment of the present invention, in which the number of the front license plate and the rear license plate is determined in a plurality of lanes such as a camera having all lanes of the up lane and the down lane, and two lanes of the up lane. It is a license plate recognition processing device configured to be able to perform recognition.

  With this configuration, since the camera field of view is all lanes as shown in FIG. 11, it is possible to determine the vehicles in the up lane and the vehicles in the down lane and recognize the license plate of each lane. In addition, it can be installed without distinguishing the vehicle number recognition of the front license plate or the vehicle number recognition of the rear license plate, and the license plate recognition processing device can be easily installed.

It is a block diagram of the license plate recognition processing apparatus which is one Example of this invention. It is a block diagram of the image processing apparatus of a present Example. It is a flowchart which performs automatic switching of the recognition process of the front number plate and back surface number plate of a present Example. It is a figure which shows the visual field of the camera at the time of front license plate recognition. It is a figure which shows the visual field of the camera at the time of rear surface license plate recognition. It is a figure which shows the example of a setting of a switching determination area | region and a template search area | region. It is a figure explaining the characteristic calculation process of the projection difference between 2 frames. It is the figure which showed the pattern matching process. It is the figure which showed the calculation process of the movement vector. It is the figure which showed the change of the cumulative value of a movement vector. It is a block diagram of the license plate recognition processing apparatus which is another Example of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Camera for vehicle number recognition, 2 ... Image processing apparatus, 3 ... Central processing unit, 4 ... Network, 5 ... Alternating road, 6 ... Road, 10 ... Image processing part, 11 ... A / D converter, 12 ... Image memory, 13... Image processor, 14... D / A converter, 15.

Claims (5)

  A vehicle number recognition camera that captures an image of a vehicle, and an image processing device that includes an image processing processor that performs image processing on an image captured by the vehicle number recognition camera, the vehicle number recognition camera using the image processing device The template is calculated from the first image projection difference between the image projections of the two frames photographed in Step 2, and the region matching the template is searched from the second image projection difference between the image projections of the two frames after the elapse of time. A license plate recognition device that calculates a movement vector from an area matched with the template and switches between a front license plate recognition process and a rear license plate recognition process.   A vehicle number recognition camera captures a plurality of frames of an image of a vehicle, calculates a template from a first image projection difference between two frame image projections captured by an image processing device, and after a lapse of time from the two frames A second image projection difference between the image projections of two frames is calculated, a region matching the template is searched by pattern matching between the calculated second image projection difference and the template, and the region matching the template is searched. A license plate recognition method that calculates a movement vector and switches between front license plate recognition processing and rear license plate recognition processing.   The license plate recognition apparatus according to claim 1, wherein a region having a large distribution difference in image projection density value of the first image projection difference is used as a template.   The license plate recognition device according to claim 1, wherein the vehicle number recognition camera is installed on an alternate road. 3. The license plate recognition method according to claim 2, wherein switching between the front license plate recognition process and the rear license plate recognition process is determined based on a cumulative value of the movement vector and a threshold value, or based on a direction of the movement vector per unit time.

Images