JPH07114688A - Method for recognizing vehicle registered number - Google Patents

Abstract

PURPOSE:To provide the method for fast vehicle registered number recognition although geometrical conversion for obtaining a front facing image by correcting geometrical distortion of an image of a license plate is included by solving the problem points that it is difficult to detect the edge of the license plate when the license plate and a vehicle body are in colors of the same system as a result of the geometrical conversion and processes are complicated and require a long time. CONSTITUTION:A coordinate conversion parameter set corresponding to all the coordinates in the range on an image where the license plate is possibly positioned is previously found and stored (SO). an image of a traveling vehicle is taken in (S2) corresponding to vehicle detection (S1), and the license plate part is extracted (S3); and a position in the image of the license place part is specified (S4), coordinate conversion is performed by using the coordinate conversion parameter set made to correspond to the position of the license plate part to obtain the front facing image of the license plate part (S6), and respective characters are recognized (S7) from the front facing image.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vehicle registration number recognition method, and more particularly to a vehicle registration number recognition method including a geometric transformation process for correcting geometric distortion of an input image.

[0002]

2. Description of the Related Art A device for recognizing a vehicle registration number by image-processing a video signal from a video camera is used. For example, a vehicle registration number recognition device for an entering vehicle as shown in FIG. 5 is used for automation even at a gate of a parking lot.

In FIG. 5, reference numeral 10 denotes a vehicle registration number recognition device installed diagonally forward of the vehicle travel path 5 at the parking lot gate. The vehicle registration number recognition device 10 includes a video camera 1, an image processing unit 11 that performs image processing on an output video signal of the video camera 1 to recognize a vehicle registration number, an illuminator 3, an illumination controller 4, and a vehicle head sensor. Vehicle runway 5
A vehicle head sensor is constituted by a light projector 2A installed on one of the roadside portions, and a light receiver 2B installed opposite to the light projector 2A with the vehicle traveling path 5 in between.

The vehicle head sensor (light receiving side) 2B is an image processing unit.
It is connected to 11 and detects that the passing vehicle 7 blocks the irradiation light from the floodlight 2A, and captures the foreground including the license plate 6 of the passing vehicle. At this time, the illuminator 3 and the illumination controller 4 emit strobe light in synchronization with the output of the on-vehicle sensor light receiver 2B, and a clear image can be obtained indoors, outdoors, and day and night.

In this type of image processing apparatus, generally, a video signal including a vehicle obtained by an image input unit is stored in an image storage unit as a still image, and then a license plate extraction unit extracts a license plate corresponding portion to extract characters. The cutout portion extracts the character equivalent portions, and the character recognition portion recognizes the characters corresponding to these character equivalent portions.

Various devices for extracting such a license plate have been proposed, for example, the applicant of the present invention, Japanese Patent Application No. 4-1377873 (image processing device). A license plate extraction method has also been proposed, for example, by the present applicant as Japanese Patent Application Laid-Open No. 5-143888 (Japanese Patent Application No. 3-332552).
Regarding the method of recognizing the vehicle registration number, for example, Japanese Patent Application No. 4-137874 (a method of recognizing numbers etc.) by the applicant
I have a suggestion.

In such a vehicle registration number recognizing device, the installation position of the image input unit (video camera) is generally restricted. For example, in the case of FIG. 5, it is installed on the roadside, and the image is obtained obliquely from the front. The processed image is the target of image processing. An example of an image of the license plate thus obtained is shown in FIG.

By the way, in applications such as recognition of a vehicle registration number, the binary pattern of the character equivalent portion of the license plate is compared with a character pattern (dictionary) stored in the apparatus to determine which character each. It has become. For this reason, the character having geometric distortion as shown in FIG. 4A does not match the character pattern dictionary, so that it cannot be recognized or erroneously recognized, and the reliability of the device is deteriorated.

In order to deal with this, arithmetic processing is performed based on an image in which geometric distortion has occurred, and after the geometrical conversion to a front-facing image (an image corresponding to the case of being imaged from the front: see FIG. 4B), Image processing to improve recognition accuracy. Pseudo affine transformation is well-known as the above-mentioned method of geometric transformation. For example, each coordinate transformation parameter (a, b, c, d, e, f, g, h) is calculated from four points at the four corners of the license plate frame. This coordinate transformation parameter set (a, b, c, d, e, f, g, h
) Is used to perform geometric transformation of each coordinate. By the conversion, for example, the coordinates (X, Y) are obtained corresponding to the coordinates (x, y) (see FIG. 2).

The following method is known as a method of geometrically converting four points. (1) Hough transform is applied to the edge detected vehicle image, the boundary line of the license plate is detected from the extracted straight line group,
The vertex coordinates of the four corners of the license plate are obtained, and an image of the license plate viewed from the front is created by coordinate conversion.
References regarding this method include, for example, “Cui, Yasuiin, Nakajima;“ License plate area extraction method using adaptive parameter plane area limited Hough transform ”: IEICE Transactions '89 / 4 Vol. J72-D-II N
o. 4, pp597-604 ”.

(2) Using the character arrangement rule of the license plate, after detecting the lower right number that is always present in the license plate, a peripheral distribution in the vertical direction is created to obtain the inclination of the license plate character, Estimate the vertex coordinates of the four corners of the plate and create an image of the license plate viewed from the front by coordinate conversion. References regarding this method include, for example, “Kawashima, Tsuji, Suzuki;“ Vehicle number recognition of gate-passing vehicles ”: Traffic Engineering Vol. 23 No. 4, pp15-
24 1988 ”.

Each of the above-mentioned conventional methods has the following problems. In the method (1), it is difficult to detect the edge when the color of the license plate and the color of the vehicle body are similar colors. In addition, it takes processing time. The method (2) requires complicated processing and takes time.

[0013]

The present invention has been made in view of the above circumstances, and it is difficult to detect an edge when the color of the license plate and the vehicle body are similar colors, which the conventional geometric conversion method has. Or, a geometric conversion method that solves the problem of complicated and time-consuming processing is adopted, and a method of recognizing the vehicle registration number by obtaining a facing image in which the geometric distortion of the license plate image is corrected is proposed. The purpose is to do.

[0014]

In order to achieve the above object, in the vehicle registration number recognition method of the present invention, a vehicle running path is imaged by a video camera, and the coordinates of all the range on the image where the license plate can be located are respectively. The corresponding coordinate transformation parameter set is calculated and stored, the image including the foreground or the foreground of the traveling vehicle is captured in response to vehicle detection, the license plate portion is extracted, and the position of the license plate portion in the image is extracted. Is specified, coordinate conversion is performed using the coordinate conversion parameter set associated with the position of the license plate portion, a face-to-face image of the license plate portion is obtained, and each character is recognized from the face-to-face image.

[0015]

The image including the license plate is taken in, first, the license plate portion is extracted and, for example, the position of the center position in the image is specified. Next, coordinate conversion is performed using the coordinate conversion parameter set that has been obtained and stored in advance corresponding to the obtained position, and the facing image of the license plate portion is obtained. The coordinate transformation parameter set need only be called. Subsequently, a predetermined character recognition process such as cutting out a character area and recognizing each character is performed from the obtained facing image.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In the method of the present invention, in most vehicle registration number recognition, the inclination of the license plate with respect to the vehicle body due to mounting and the vehicle attitude (angle with respect to the camera) at the time of imaging
Takes advantage of this by noting that it is almost constant. As long as the above conditions are satisfied, there is a correlation between the license plate distortion on the image caused by geometrical reasons and the position coordinates of the license plate character area.

Therefore, a large number of coordinate conversion parameter sets corresponding to the individual coordinates are prepared in advance for all the coordinates of the range where the license plate can exist on the image, and the position of the character plate character area of the input image ( For example, the center coordinate of the character area) is obtained, the corresponding appropriate coordinate conversion parameter set is selected, and the coordinate conversion is performed by calculation using this coordinate conversion parameter set. ). Pay attention to the details of each input image,
Since it is not necessary to obtain each parameter, high speed conversion processing is possible.

The present invention will be described below based on examples. FIG. 1 is a flowchart showing an embodiment of a vehicle registration number recognition method of the present invention. The vehicle registration number recognition device to which the method of the embodiment is applied may be the same as that shown in FIG. 5, and only the control process of each part is a control process suitable for the present invention.

In the method of the embodiment, prior to the vehicle registration number recognition processing for each detected vehicle, a coordinate conversion parameter set (an, bn, cn, dn) suitable for all coordinates of the range on the image where the license plate can be located. , en, fn, gn, hn) is calculated and associated with coordinates (position on the screen) and stored (FIG. 1: S
0).

Actually, it is very difficult to obtain the coordinate conversion parameter set for all the coordinates, and a large amount of memory is required to store the coordinate conversion parameter set. Therefore, the embodiment is schematically shown in FIG. As described above, the entire area of the image is divided into 16 areas (R1 to R16), and an average coordinate conversion parameter set is obtained for one area (for example, R1). a1, b1, c
1, d1, e1, f1, g1, h1) are associated with all the coordinates in the region R1.

Each coordinate conversion parameter set (an, bn, cn, d
n, en, fn, gn, hn) may be obtained by checking the image after actually installing the device at the site, or theoretically from the predicted tilt of the license plate or the installation position of the video camera. May be calculated.

During an actual vehicle registration number recognition operation, a light beam is constantly emitted from the light projector 2A toward the light receiver 2B (see FIG. 5). The light receiver 2B sends out a signal different from that at the time of light reception (normal state) in response to the light from the light projector 2A being blocked by the entering vehicle. This change (rise or fall) is recognized by the image processing unit 11 as a vehicle head detection signal, and it can be seen that the vehicle 7 has entered a fixed position (vehicle detection: S1).

The image processing unit 11 correspondingly illuminates by the illuminator 3 in the near infrared region, and synchronously captures the foreground of the vehicle as a still image including the license plate 6 from the video signal (S2). Next, the license plate corresponding part in the still image is extracted (S3). For the extraction, an appropriate method such as the method described in JP-A-5-143888 (license plate extraction method) mentioned above may be used.

Next, it is specified which part in the image the extracted license plate equivalent part is located. In the embodiment, the coordinates (xn, yn) of the central position of the license plate, which can be easily and stably obtained, are calculated and used as the (representative) position of the license plate (S4).

A coordinate transformation parameter set (an, bn, c) for pseudo affine transformation associated with an area including the license plate position (xn, yn) obtained in the above process.
n, dn, en, fn, gn, hn) is called from the storage table (S
5) Using these, coordinate conversion is performed by calculation of pseudo affine transformation (S6).

In FIG. 3, if the license plate portion is located at the position indicated by the symbol P1, the central coordinates (x
Region R1 in which this center coordinate is included in the calculation of (1, y1)
Coordinate conversion parameter set (a1, b1, c1, d1, e
1, f1, g1, h1) is called from the storage table, and by the pseudo affine transformation, the facing image P ′ of the license plate
Is required.

Similarly, when the license plate portion (center coordinates) is located at the position indicated by reference sign P16, the coordinate conversion parameter set (a16, a16, corresponding to the region R16 including the center coordinates (x16, y16) is included. b16, c16, d16, e16, f1
6, g16, h16) is adopted and pseudo affine transformation is performed,
The facing image P'of the license plate is determined.

Front image P'of the obtained license plate
Each character area is cut out from and each character is recognized (S
7). The vehicle registration number may be recognized by a method according to the above-mentioned Japanese Patent Application No. 4-137874 (a method for recognizing numbers etc.).

In the embodiment, the image is divided into 16 areas (R1 to R1).
Although it is divided into R16), it is further divided to obtain the coordinate conversion parameter set for each area,
The accuracy of geometric conversion may be increased.

In addition to the affine transformation, the present invention can be applied to a vehicle registration number recognition method using the same kind of geometric transformation using parameters, and a coordinate transformation parameter set required for geometric transformation is displayed in an image. A plurality of sets are obtained in advance and stored in association with the positions of the license plate portions, and a predetermined geometric transformation is performed using an appropriate corresponding coordinate transformation parameter set stored at each actual vehicle detection. Should be done. This eliminates the need for determining the coordinate conversion parameter set for each event, which is generally complicated and therefore requires a long processing time, and enables sufficiently high speed vehicle registration number recognition while including geometric conversion.

[0031]

As described above in detail, according to the vehicle registration number recognition method of the present invention, the vehicle traveling path is imaged by the video camera, and the coordinate conversion parameters corresponding to all the coordinates of the range on the image where the license plate can be located. The set is sought and stored, the image including the foreground or the background of the traveling vehicle is captured in response to the vehicle detection, the license plate portion is extracted,
The position of the license plate in the image is specified, coordinate conversion is performed using the coordinate conversion parameter set associated with the position of the license plate to obtain the front image of the license plate, and from the front image. Since each character is recognized, the vehicle registration number recognition is significantly faster than the conventional method even though it includes the geometric conversion process that corrects the geometric distortion of the license plate image to obtain the front-facing image. There is. There is also an effect that the license plate portion can be appropriately extracted and the vehicle registration number can be surely recognized even when the license plate and the vehicle body whose edges are difficult to detect have the same color.

[Brief description of drawings]

FIG. 1 is a flowchart showing an embodiment of a vehicle registration number recognition method of the present invention.

FIG. 2 is a diagram illustrating an example of geometric transformation according to the present invention.

FIG. 3 is a diagram illustrating an example of geometric transformation according to the embodiment of the present invention.

FIG. 4 is an example of an image including a license plate portion according to the present invention.

FIG. 5 is a plan view illustrating an arrangement of a vehicle registration number recognition device according to the present invention.

[Explanation of symbols]

 5 ... Vehicle running path, 6 ... License plate, 1 ... Video camera, 11 ... Image processing unit.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location G08G 1/04 D 7531-3H H04N 7/18 K

Claims (1)

[Claims] 1. A vehicle traveling path is imaged by a video camera, and a coordinate conversion parameter set corresponding to each coordinate of all the range on the image where the license plate can be located is obtained and stored (S0), vehicle detection ( Corresponding to (S1), the image including the foreground or the foreground of the traveling vehicle is captured (S2), the license plate part is extracted (S3), and the position of the license plate part in the image is specified (S4). A process of performing coordinate conversion using a coordinate conversion parameter set associated with the position of the plate portion to obtain a facing image of the license plate portion (S6), and recognizing each character from the facing image (S7). A vehicle registration number recognition method comprising: