AU2020102152A4

AU2020102152A4 - Algorithm and Device for License Plate Number Recognition Based on Line Drawing

Info

Publication number: AU2020102152A4
Application number: AU2020102152A
Authority: AU
Inventors: Yingquan Peng; Jiayu Song; Peng Wang
Original assignee: Lanzhou University
Current assignee: Lanzhou University
Priority date: 2019-09-06
Filing date: 2020-09-06
Publication date: 2020-10-15
Anticipated expiration: 2028-09-06
Also published as: CN110533003B; CN110533003A

Abstract

ALGORITHM AND DEVICE FOR LICENSE PLATE NUMBER RECOGNITION BASED ON LINE DRAWING The present invention relates to image data processing, and in particular, to an algorithm and a device for number recognition. This algorithm is used for license plate number recognition. An algorithm and a device for license plate number recognition based on line drawing include the following steps: acquiring a license plate image; performing grayscale and binarization operations on the acquired license plate image; locating a license plate to obtain location data; dividing, based on the location data, each number into valid areas 1, 2, 3, and 4 from top to bottom in the second frame of image; setting contrast values for the four valid areas, counting the number of Is in each row of image data in area 1, area 2, area 3, and area 4, taking a maximum or minimum count value as an eigenvalue, and comparing the eigenvalue with the contrast value for the area to recognize a number. The present invention counts four rows of image data, with faster speed and higher accuracy. Compared with the method of selecting only one row of data, the present invention covers a larger range and is more reliable. 1 / 1 Image processing module First D M Third Display FIFO FIFO Video acquisition FPGA module Preproce Algorithm Second. ssing - -- implementation FIFO module module FIG. 1 AArea I Area 2 Area 3 ) Area 4 FIG. 2

Description

1/1

Image processing module

First D M Third Display FIFO FIFO

Video acquisition FPGA module Preproce Algorithm Second. ssing - -- implementation FIFO module module

FIG. 1

AArea I

Area 2 Area 3

) Area 4

FIG. 2

ALGORITHM AND DEVICE FOR LICENSE PLATE NUMBER RECOGNITION BASED ON LINE DRAWING TECHNICAL FIELD

The present invention relates to image data processing, and in particular, to an algorithm and a device for number recognition. This algorithm is used for license plate number recognition.

BACKGROUND

Xilinx launched the world's first FPGA in 1984. In less than 40 years, FPGA has replaced ASIC chips in some areas due to its short design cycle, programmability, fast calculation speed, low power consumption, and low cost. In recent years, with the development of the image processing technology, artificial intelligence, and big data, the FPGA-based data processing system has developed rapidly. Currently, Xilinx and Intel have dominated the FPGA market. The recognition system and algorithm in the present invention are based on Intel's Quartus II and DE1_SOC development board.

The concept of Optical Character Recognition (OCR) was proposed by the German scientist Tansheck in 1929. With the development of the society and image processing technology, the recognition algorithm for printed numbers such as license plate numbers becomes mature and has a promising prospect. At present, there are three main types of number recognition algorithms: template matching; various neural network algorithms, including convolutional neural networks and spiking neural networks; and character feature extraction. The template matching method pre-stores the template, and then compares the acquired data with the template to obtain the result. The existing template matching method has low anti-interference ability, poor accuracy, and narrow application range, which cannot meet most recognition requirements. The neural network algorithm achieves high recognition, but it requires a large amount of training data to obtain accurate parameters, which greatly increases the hardware cost. The feature extraction method is generally used for printed number recognition. Because the printed number has distinctive features, this algorithm has high recognition accuracy. The most commonly used character feature extraction method is line drawing. The line drawing method extracts specific rows and columns of image data, calculates the changes of the data on these rows and columns (0-1 rising edge and 1-0 falling edge), and then checks the positions of the changed row data and column data to obtain the result. This feature extraction method is undoubtedly the most convenient and efficient method for the hardware system to recognize printed numbers.

The traditional line drawing method draws horizontal and vertical lines, where the horizontal line represents the row data of the image, and the vertical line represents the column data of the image. At least two rows and one column of data of the image need to be extracted. The result is obtained by determining the data changes and checking the position changes. However, the image data is transmitted in rows, and it is troublesome to extract the row data and column data at the same time. The image data needs to be cached, which affects the recognition speed. In addition, the extraction is greatly affected by noise, and the image area of each number must be accurately segmented and located.

SUMMARY

In order to solve the above problems, the present invention proposes an algorithm and a device for license plate number recognition based on line drawing.

The present invention has the following technical solution.

An algorithm for license plate number recognition based on line drawing includes the following steps: acquiring a license plate image; and preprocessing the acquired license plate image.

The specific process of image processing is as follows: (1) Image preprocessing: Perform grayscale and binarization operations on the acquired license plate image to obtain binary image data; and convert the image into a black and white image with only data "0" and "1", where an area of numbers is white, and the blue background of a license plate is black. (2) Image recognition: 1) Locate the license plate to obtain location data. Obtain the position of the license plate in the entire binary image data and the width of the license plate from the first frame of image data. 2) Divide the area of each number based on the location data in the second frame of image. 3) Divide each number into four valid areas from top to bottom for data recognition: area 1, area 2, area 3, and area 4; set contrast values for the four valid areas; count the number of 1s in each row of image data in area 1, area 2, area 3, and area 4, take a maximum or minimum count value as an eigenvalue, and compare the eigenvalue with the contrast value for the area to recognize the number.

Preferably, in areas 1, 3, and 4, the maximum count values are selected as the eigenvalues and recorded as cnt_1, cnt_3, and cnt_4; and in area 2, the minimum count value is selected as the eigenvalue and recorded as cnt_2.

More preferably, the specific process of dividing each number into four valid areas from top to bottom for data recognition is as follows: For area 1: count the number of 1s in each row of image data; take a maximum value as the eigenvalue and record it as cnt_1; set a contrast value layer_1; compare cnt_1 with layer_1, and if cnt_1 > layer_1, record an instruction value 00, or if cnt_1 < layer_1, record an instruction value 11. For area 2: count the number of 1s in each row of image data; take a minimum value as the eigenvalue and record it as cnt_2; set a contrast value layer_2; compare cnt_2 with layer_2, and if cnt_2 < layer_2, record an instruction value 1, or if cnt_2 > layer_2, record an instruction value 0; and update the instruction value, where if the instruction value contains a unique value, a number corresponding to the instruction value is recognized, that is, the number 4 is recognized. For area 3: group unrecognized numbers with the same instruction value, group 1: numbers 0, 8, and 9, all corresponding to an instruction value 000; group 2: numbers 1 and 6, both corresponding to an instruction value 111; group 3: numbers 2, 3, 5, and 7, all corresponding to an instruction value 001; set three contrast values for the groups, which are recorded as layer_3_0, layer_3_1, and layer_3_2; count the number of 1s in each row of image data; take a maximum value as the eigenvalue and record it as cnt_3; identify 1, 2, and 7 from other numbers based on layer_3_0, and identify the number 6; if cnt_3 < layer_3_0, record 11; if layer_3_0 < cnt_3 < layer_3_1, record 10; if layer_3_1 < cnt_3 < layer_3_2, record 00; if cnt_3 > layer_3_2, record 01, where 0, 1, 3, 4, 5, and 6 have been recognized, and 2, 7, 8, and 9 remain unrecognized. For area 4: count the number of 1s in each row of image data; take a maximum value as the eigenvalue and record it as cnt_4; set a contrast value layer_4; compare cnt_4 with layer_4, and if cnt_4 > layer_4, record an instruction value 0, or if cnt_4 < layer_4, record an instruction value 1, to identify 2. 7, 8, and 9.

Alternatively, preferably, the specific implementation step of locating the license plate to obtain location data is: when the first frame of image data arrives, counting the number of Os in each row of data, where "0" represents the black part of the image; and when the number of Os in a row is greater than a specified value, determining that this row is a starting position of the license plate, and that the number of Os in this row of image data is the width of the license plate.

A device for license plate number recognition based on line drawing uses the foregoing method for license plate number recognition based on line drawing, and includes a video acquisition module, an FPGA image processing module, and a display module that are electrically connected in sequence, where the video acquisition module is an OV5640 camera that acquires 1024*720 images; the display module is a display; and the FPGA image processing module includes two branches connected to the video acquisition module, where one branch includes a first FIFO, an SDRAM, and a third FIFO connected in sequence; the other branch includes a preprocessing module, a second FIFO, an algorithm implementation module, and an LED connected in sequence; the first FIFO is also connected to the preprocessing module, and the third FIFO is connected to the display.

The present invention has the following technical effects: 1. In this algorithm, four horizontal lines are drawn, that is, four rows of image data are analyzed, which can be quickly completed during the data transmission. 2. This algorithm counts the number of 1s in four rows of image data instead of determining the positions of the rising edge and falling edge. This can reduce the impact of noise on the accuracy of the algorithm. 3. This algorithm counts each row of data in four areas, and then selects the maximum or minimum values as the four eigenvalues. Compared with the algorithm of selecting only one row of data, this algorithm covers a larger range and is more reliable.

BRIEF DESCRIPTION OF DRAWINGS

FIG. 1 is a structural diagram of a system according to the present invention; and

FIG. 2 is a schematic diagram of an algorithm according to the present invention.

DETAILED DESCRIPTION

Example 1

An algorithm for license plate number recognition based on line drawing includes the following steps: (1) acquiring a license plate image; and (2) preprocessing the acquired license plate image. The specific process is as follows: Image preprocessing: Perform grayscale and binarization operations on the acquired license plate image to obtain binary image data; and convert the image into a black and white image with only data "0" and "1", where an area of numbers is white, and the blue background of a license plate is black. Image recognition: 1) Locate the license plate to obtain location data. Obtain the position of the license plate in the entire binary image data and the width of the license plate from the first frame of image data. 2) Divide the area of each number based on the location data in the second frame of image. 3) Divide each number into four valid areas from top to bottom for data recognition: area 1, area 2, area 3, and area 4; set contrast values for the four valid areas; count the number of 1s in each row of image data in area 1, area 2, area 3, and area 4, take a maximum or minimum count value as an eigenvalue, and compare the eigenvalue with the contrast value for the area to recognize the number.

Example 2

An algorithm for license plate number recognition based on line drawing includes the following steps: (1) Acquire a license plate image through an OV5640 camera (video acquisition module), where the image size is 1024*720, and the data format is RGB565. (2) Preprocess the acquired license plate image. The specific process is as follows: Image preprocessing: Perform grayscale and binarization operations on the acquired license plate image to obtain binary image data; take an R component of the image data as the grayscale data, set a threshold, and then determine whether the R component exceeds this threshold; and obtain new image data "1" if the R component of the image data is greater than the threshold, or obtain "0" if the R component is less than the threshold. In this way, the entire image is converted into a black and white image with only data "0" and "1". For the license plate, an area of numbers is white, and the blue background is black. (3) Calculate the position of the license plate in the entire image and the width of the license plate from the first frame of image data. The specific process is as follows: When the first frame of image data arrives, count the number of Os in each row of data, where "0" represents the black part of the image; and when the number of Os in a row is greater than a specified value, determine that this row is a starting position of the license plate, and that the number of Os in this row of image data is the width of the license plate. (4) Divide the area of each number in the second frame of image. The specific process is as follows: Divide each number into four valid areas from top to bottom for data recognition: area 1, area 2, area 3, and area 4. (5) Recognize the data in each area sequentially. The specific process is as follows: For area 1: count the number of 1s in each row of image data, that is, in the white part of the image; after counting on all the rows in area 1 is completed, take a maximum value as the eigenvalue and record it as cnt_1; set a contrast value layer_1; compare cnt_1 with layer_1, and if cnt_1 > layer_1, record an instruction value 00, or if cnt_1 < layer_1, record an instruction value 11.

The cnt_1 values of the numbers 1, 4, and 6 are obviously smaller than the cnt_1 values of other numbers. The following table lists the first two digits of each instruction value for determining each number.

Number 0 1 2 3 4 5 6 7 8 9 Instruction 00 11 00 00 11 00 11 00 00 00

For area 2: count the number of is in each row of image data; take a minimum value as the eigenvalue and record it as cnt_2; set a contrast value layer_2; compare cnt_2 with layer_2, and if cnt_2 < layer_2, record an instruction value 1, or if cnt_2 > layer_2, record an instruction value 0; and update the instruction value, as listed in the following table.

Number 0 1 2 3 4 5 6 7 8 9 Instruction 000 111 001 001 110 001 111 001 000 000

The number corresponding to a unique instruction value is recognized. For example, the instruction for determining the number 4 is 110, and the number 4 is recognized.

For area 3: group unrecognized numbers with the same instruction value, group 1: numbers 0, 8, and 9, all corresponding to an instruction value 000; group 2: numbers 1 and 6, both corresponding to an instruction value 111; group 3: numbers 2, 3, 5, and 7, all corresponding to an instruction value 001.

Set three contrast values for the groups, which are marked as layer_3_0, layer_3_1, and layer_3_2; count the number of s in each row of image data; take a maximum value as the eigenvalue and record it as cnt_3.

There is a great difference between the cnt_3 values of numbers 1 and 6 in group 2, between the cnt_3 values of numbers 2 and 7, and between the cnt_3 values of numbers 3 and 5, which is close to the number 1. Therefore, select layer_3_0 to identify the numbers 1, 2, and 7 from other numbers, and identify the number 6, and if cnt_3 < layer_3_0, record 11.

The cnt_3 value of the number 0 in group 1 is less than the cnt_3 values of the numbers 8 and 9. Record 10 if layer_3_0 < cnt_3 < layer_3_1.

The cnt_3 value of the number 3 is obviously less than the cnt_3 value of the number 5. Then, use layer_3_2 to distinguish 3 from 5, and record 00 if layer_3_1 < cnt_3 < layer_3_2, or record 01 if cnt_3 > layer_3_2. Because the determining is made by group, some numbers may satisfy multiple conditions. Therefore, all possible instructions are listed to ensure accuracy, as listed in the following table.

Number 0 1 2 3 4 5 6 7 8 9

Instruction 00010 11111 00111 00100/ 110xx 00101 11101 00111 00000 00000

00110 / / /

00010 11100 00001 00001

According to the above table, 0, 1, 3, 4, 5, and 6 have been recognized, and 2, 7, 8, and 9 remain unrecognized.

For area 4: count the number of is in each row of image data; take a maximum value as the eigenvalue and record it as cnt_4; set a contrast value layer_4; compare cnt_4 with layer_4. The cnt_4 values of numbers 2 and 8 are obviously greater than the cnt_4 values of numbers 7 and 9. Record 0 if cnt_4 > layer_4, or record 1 if cnt_4 < layer_4, to obtain the following determining instructions.

Number 0 1 2 3 4 5 6 7 8 9

Instruction 00010 11111 00111 00100 110xx 00101 11101 00111 00000 00000

n x x 0 x x x x 1 0 1

00010 00110 11100 00001 00001

x x x 0 1

Each number corresponds to a different instruction, that is, all numbers can be recognized.

Example 3

Claims

What is claimed is:

1. An algorithm for license plate number recognition based on line drawing, comprising: acquiring a license plate image; and preprocessing the acquired license plate image, wherein the specific process of image processing is as follows: (1) image preprocessing: performing grayscale and binarization operations on the acquired license plate image to obtain binary image data; and converting the image into a black and white image with only data "0" and "1",wherein an area of numbers is white, and the blue background of a license plate is black; (2) image recognition: 1) locating the license plate to obtain location data; obtaining the position of the license plate in the entire binary image data and the width of the license plate from the first frame of image data; 2) dividing the area of each number based on the location data in the second frame of image; 3) dividing each number into four valid areas from top to bottom for data recognition: area 1, area 2, area 3, and area 4; setting contrast values for the four valid areas; counting the number of 1s in each row of image data in area 1, area 2, area 3, and area 4, taking a maximum or minimum count value as an eigenvalue, and comparing the eigenvalue with the contrast value for the area to recognize the number.

2. The algorithm for license plate number recognition based on line drawing according to claim 1, comprising: in areas 1, 3, and 4, selecting the maximum count values as the eigenvalues, and recording them as cnt_1, cnt_3, and cnt_4; and in area 2, selecting the minimum count value as the eigenvalue, and recording it as cnt_2.

3. The algorithm for license plate number recognition based on line drawing according to claim 2, wherein the specific process of dividing each number into four valid areas from top to bottom for data recognition is as follows: for area 1: counting the number of 1s in each row of image data; taking a maximum value as the eigenvalue and recording it as cnt_1; setting a contrast value layer_1; comparing cnt_1 with layer_1; and if cnt_1 > layer_1, recording an instruction value 00, or if cnt_1 < layer_1, recording an instruction value 11; for area 2: counting the number of 1s in each row of image data; taking a minimum value as the eigenvalue and recording it as cnt_2; setting a contrast value layer_2; comparing cnt_2 with layer_2, and if cnt_2 < layer_2, recording an instruction value 1, or if cnt_2 > layer_2, recording an instruction value 0; and updating the instruction value, wherein if the instruction value contains a unique value, a number corresponding to the instruction value is recognized, that is, the number 4 is recognized; for area 3: grouping unrecognized numbers with the same instruction value, group 1: numbers 0, 8, and 9, all corresponding to an instruction value 000; group 2: numbers 1 and 6, both corresponding to an instruction value 111; group 3: numbers 2, 3, 5, and 7, all corresponding to an instruction value 001; setting three contrast values for the groups, which are recorded as layer_3_0, layer_3_1, and layer_3_2; counting the number of 1s in each row of image data; taking a maximum value as the eigenvalue and recording it as cnt_3; identifying 1, 2, and 7 from other numbers based on layer_3_0, and identifying the number 6; if cnt_3 < layer_3_0, recording 11; if layer_3_0 < cnt_3 < layer_3_1, recording 10; if layer_3_1 < cnt_3 < layer_3_2, recording 00; if cnt_3 > layer_3_2, recording 01, wherein 0, 1, 3, 4, 5, and 6 have been recognized, and 2, 7, 8, and 9 remain unrecognized; for area 4: counting the number of 1s in each row of image data; taking a maximum value as the eigenvalue and recording it as cnt_4; setting a contrast value layer_4; comparing cnt_4 with layer_4; and if cnt_4 > layer_4, recording an instruction value 0, or if cnt_4 < layer_4, recording an instruction value 1, to identify 2. 7, 8, and 9.

4. The algorithm for license plate number recognition based on line drawing according to claim 1, wherein the specific implementation step of locating the license plate to obtain location data is: when the first frame of image data arrives, counting the number of Os in each row of data, wherein "0"represents the black part of the image; and when the number of Os in a row is greater than a specified value, determining that this row is a starting position of the license plate, and that the number of Os in this row of image data is the width of the license plate.

5. A device for license plate number recognition based on line drawing, wherein the device uses the method for license plate number recognition based on line drawing according to claim 1, 2, 3, or 4, and comprises a video acquisition module, a field-programmable gate array (FPGA) image processing module, and a display module that are electrically connected in sequence, wherein the video acquisition module is an OV5640 camera that acquires 1024*720 images; the display module is a display; and the FPGA image processing module comprises two branches connected to the video acquisition module, wherein one branch comprises a first FIFO, an SDRAM, and a third FIFO connected in sequence; the other branch comprises a preprocessing module, a second FIFO, an algorithm implementation module, and an LED connected in sequence; the first FIFO is also connected to the preprocessing module, and the third FIFO is connected to the display.