WO2018018788A1 - 一种基于图像识别的计量表抄表装置及其方法 - Google Patents

一种基于图像识别的计量表抄表装置及其方法 Download PDF

Info

Publication number
WO2018018788A1
WO2018018788A1 PCT/CN2016/105763 CN2016105763W WO2018018788A1 WO 2018018788 A1 WO2018018788 A1 WO 2018018788A1 CN 2016105763 W CN2016105763 W CN 2016105763W WO 2018018788 A1 WO2018018788 A1 WO 2018018788A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
character
region
meter reading
recognition
Prior art date
Application number
PCT/CN2016/105763
Other languages
English (en)
French (fr)
Inventor
刘相莹
刘何来
乐方升
崔涛
Original Assignee
深圳友讯达科技股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳友讯达科技股份有限公司 filed Critical 深圳友讯达科技股份有限公司
Publication of WO2018018788A1 publication Critical patent/WO2018018788A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition
    • G06V30/14Image acquisition
    • G06V30/146Aligning or centring of the image pick-up or image-field
    • G06V30/147Determination of region of interest
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements
    • G06V20/60Type of objects
    • G06V20/62Text, e.g. of license plates, overlay texts or captions on TV images
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/29Graphical models, e.g. Bayesian networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/40Image enhancement or restoration using histogram techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/70Denoising; Smoothing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/90Dynamic range modification of images or parts thereof
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/24Aligning, centring, orientation detection or correction of the image
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/30Noise filtering
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition
    • G06V30/14Image acquisition
    • G06V30/148Segmentation of character regions
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition
    • G06V30/16Image preprocessing
    • G06V30/162Quantising the image signal
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition
    • G06V30/16Image preprocessing
    • G06V30/164Noise filtering
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition
    • G06V30/18Extraction of features or characteristics of the image
    • G06V30/18086Extraction of features or characteristics of the image by performing operations within image blocks or by using histograms
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition
    • G06V30/19Recognition using electronic means
    • G06V30/196Recognition using electronic means using sequential comparisons of the image signals with a plurality of references
    • G06V30/1983Syntactic or structural pattern recognition, e.g. symbolic string recognition
    • G06V30/1988Graph matching
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10004Still image; Photographic image
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00Indexing scheme relating to image or video recognition or understanding
    • G06V2201/02Recognising information on displays, dials, clocks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition

Definitions

  • the invention relates to the field of meter reading, in particular to a meter reading device based on image recognition and a method thereof.
  • the Chinese invention patent CN200410066507 the patent name is: a remote wireless meter reading method, which uses a photographing device to take the degree information on the electricity meter to obtain an original image representing the degree information.
  • the original image is converted into a corresponding gray image and a binary image; the preprocessing portion performs image denoising, tilt correction, spurious removal, and filtering processing on the binary image.
  • character segmentation is performed to obtain a single image
  • feature extraction is performed on a single image
  • character information is recognized by the Euclidean method, and finally the degree information is obtained.
  • China invention patent CN201420159056 the patent name is: a gas meter reading system based on the public service platform mobile terminal, which uses the public service platform terminal and the server connected to the payment institution to realize the meter reading function.
  • the reading recognition module locates the region where the number in the image information is located, and then performs a single digital segmentation process on the display data, and then identifies the segmented single digit, and finally confirms or corrects the single digit through the data manual confirmation and correction module. correct.
  • the pre-processing part of the patent CN200410066507 denoises the binary image, tilts the correction, removes the stray point filtering and performs character segmentation.
  • This processing method is not clean or photographed due to the gas meter. Incompletely clear images caused by unclean heads, light interference, etc., do not meet the segmentation requirements.
  • the Euclidean distance recognition algorithm utilizes the features of the image, the recognition algorithm is simple, but the feature vector class label is discriminated by the distance between the feature vectors, which is not based on statistical prior knowledge.
  • the patent CN201420159056 locates the region where the number in the image information is located, then performs a single digital segmentation process on the display data, and then identifies the segmented single digit, and finally confirms or corrects the single digit through the data manual confirmation and correction module. Due to the mechanical rotation of the roller, the carry-in mechanism of the reading, half-character appears, and the patent does not meet the half-character identification requirements.
  • an apparatus for meter reading based on image recognition comprising:
  • the image acquisition unit is configured to take a picture of the meter wheel number and collect the scanned image data of the reading image
  • a data processing unit for processing the scanned image data of the reading to obtain a digital image of the reading
  • a data storage unit for pre-storing a digital model of the meter
  • An image recognition unit configured to compare the read digital image with a pre-stored digital model of the meter to obtain identification data
  • a digital display unit configured to display the obtained identification data after comparison
  • the image acquisition unit obtains the scan image data of the meter by collecting the wheel number of the meter, and the collected scan data is processed by the data processing unit, and the processed data is compared with the digital model of the meter of the data storage unit by the image recognition unit. , get the identification data, and finally pass the number
  • the word display unit displays the identification data.
  • the data storage unit (400) includes a full character digital model and a half character digital model.
  • the image recognition unit (300) includes:
  • a pre-processing module for performing denoising and contrast enhancement processing on the digital image of the reading
  • a character domain region of interest detecting module for determining a region of interest of the character domain
  • An image binarization module configured to perform binarization processing on the grayscale image of the region of interest of the character domain
  • An interference removal module configured to remove interference points and interference blocks in the image of the region of interest in the binarized character domain
  • a character segmentation module configured to segment a plurality of character regions from the binarized image of the optimized region of interest region
  • a character image normalization module for uniformly normalizing each character region into an image of a specified size
  • a character recognition module for comparing and identifying a single character region content with a meter digital model stored in the data storage unit (400).
  • a meter reading-based meter reading method includes the following steps:
  • the reading digital image is denoised in the step S2, and a Gaussian filtering algorithm is adopted, and the specific formula is:
  • f(i,j) is the original grayscale image
  • g(i,j) is the filtered image
  • i,j are the longitudinal and lateral coordinates of the image
  • h(u,v) is the Gaussian kernel, as shown in the following equation :
  • the method for improving the contrast of the reading image in the step S2 is to adopt the histogram equalization method.
  • the step S3 includes the following steps:
  • S31 Obtain a vertical gradient image of the image after S2 processing, sequentially scan each line of the image, count the number of foreground pixels in each row, obtain a horizontal projection of the vertical gradient image, and smooth the same, and determine the smoothed horizontal projection image.
  • the step S4 is based on the image.
  • the foreground portion and the background portion feature determine the threshold value, and the grayscale image of the region of interest of the character domain is binarized, the foreground portion is set to white, and the background portion is set to black.
  • the method of determining the foreground portion and the background portion threshold is the Otsu method or the adaptive threshold method.
  • the interference block removing step S52 in the step S5 specifically includes the following steps:
  • the connected domain corresponding to the area height outside the effective range is an interference block.
  • the step S6 counts the number of foreground pixel points in each column of the character domain image to obtain a vertical projection image, performs column segmentation on the vertical projection image, and determines a wheel number single character. The location of the area.
  • the step S8 includes the following steps:
  • the method of finding the peak and trough can It is stable enough to determine the upper and lower boundaries of the character image, and solves the problem of image segmentation accuracy and stability in the region of interest of the character domain.
  • the histogram distribution of the characteristics of the connected regions is statistically extracted, and the interference block is effectively removed by using the Gaussian distribution characteristic to avoid the under-segmentation or over-segmentation of the character partial region, and the interference point and block problem appearing after the binarization of the character image are solved. Improved segmentation accuracy and more reliable input characters for subsequent recognition models.
  • FIG. 1 is a block diagram showing the structure of a meter reading apparatus based on image recognition according to the present invention
  • FIG. 2 is a diagram showing a structure of an image recognition unit in a meter reading apparatus based on image recognition according to the present invention
  • FIG. 3 is a flow chart showing the flow chart reading method of the meter based on image recognition according to the present invention
  • step 4 is a flow chart of detecting a region of interest of a character domain in step 3 of the meter reading method based on image recognition according to the present invention
  • FIG. 5 is a flow chart of detecting upper and lower boundaries of a character field in step 3 of the meter reading method based on image recognition according to the present invention
  • FIG. 6 is a flow chart of detecting left and right borders of a character field in step 3 of the meter reading method based on image recognition according to the present invention
  • step 7 is a flow chart of the binarized character image interference removal in step 5 of the meter reading-based meter reading method according to the present invention.
  • step 8 is a flow chart of removing a binarized character image interference block in step 5 of the method for meter reading based on image recognition according to the present invention
  • FIG. 9 is a flow chart of single character detection in step 6 of the meter reading method based on image recognition according to the present invention.
  • step 8 is a half-character identification flowchart in step 8 of the meter reading-based meter reading method according to the present invention.
  • FIG. 11 is a schematic diagram of detecting upper and lower boundaries in a horizontal gradient image of a binarized image of a region of interest in a pixel recognition method according to the present invention
  • FIG. 12 is a schematic diagram of detecting left and right boundaries in a vertical gradient image of a binarized image of a region of interest in a pixel recognition method according to the present invention
  • FIG. 13 is a schematic diagram showing the characteristics of a Gaussian distribution curve used for removing interference blocks in a binarized image of a region of interest in the image recognition-based meter reading method according to the present invention.
  • An image recognition-based meter reading device includes an image capturing unit 100.
  • the image capturing unit photographs the wheel number of the meter through the camera device, and scans the reading image obtained by the photographing.
  • the data is transmitted to the data processing unit 200 through the data.
  • the data processing unit 200 processes the read image scan data to obtain a read digital image.
  • the data storage unit 400 is configured to pre-store the meter digital model of the meter; the image recognition unit 300, For performing a series of identification processing on the digital image of the reading, by comparing the digital image of the processed reading with the digital model of the pre-stored meter to obtain the identification data; the digital display unit 500; for displaying the identification data obtained after comparison;
  • the specific process of the meter reading-based meter reading method is as follows: the image capturing unit 100 takes a picture of the wheel number of the meter to obtain the reading image scan data, and the collected reading image scan data is processed by the data processing unit 200 to read the reading data. Digital image recognition by image After processing, the unit 300 compares with the meter digital model of the data storage unit 400 to obtain identification data, and finally displays the identification data on the display screen through the digital display unit 500.
  • the data storage unit 400 includes a complete character digital model and a half-character digital model, the complete character digital model stores the complete number of the meter wheel number, and the half-character digital model stores the meter wheel number.
  • the half-character number is compared with the data by a half-character digital model, which effectively improves the accuracy of half-character recognition.
  • the image recognition unit 300 specifically includes a preprocessing module 310, a character domain region of interest detection module 320, an image binarization module 330, an interference removal module 340, a character segmentation module 350, and a character image normalization module 360. And character recognition module 370.
  • the preprocessing module 310 of the present invention is used for denoising and enhancing the contrast processing of the read image; as shown in FIG. 5, the character region of interest detecting module 320 specifically detects the upper, lower, left and right of the character image in the preprocessed image. The boundary is used to determine the region of interest in the character domain.
  • the upper and lower boundary detections first obtain the vertical gradient image of the preprocessed image, and sequentially scan each row of the vertical gradient image, and count the number of pixels in the front of each row to obtain the vertical gradient image. Project horizontally and smooth it. Since the projection values at the edges of the upper and lower boundaries in the vertical gradient image are the largest, the corresponding two maximum wave peak points are searched from the horizontal projection, and the row coordinates corresponding to the two peak points are respectively the upper boundary of the character region of interest.
  • the lower boundary as shown in Fig. 11, the first maximum wave peak position is the upper boundary top, and the second maximum wave peak position is the lower boundary down.
  • the horizontal gradient image of the preprocessed image is acquired, and each column of the horizontal gradient image is sequentially scanned, and the number of foreground pixels of each column is counted, and the level is acquired.
  • Vertical projection of the gradient image and smoothing it Since the gradient of the flat area on the left side of the vertical projection image is very small, corresponding to the trough portion of the vertical projection, the peak position after finding the trough area on the left side of the vertical projection image is the left boundary of the character area; the right side of the vertical projection image is all characters, so the right boundary That is, the rightmost edge position of the vertically projected image, as shown in FIG. 12, the first peak after the valley region is the left boundary position of the character region.
  • the image binarization module 330 is configured to perform binarization processing on the grayscale image of the region of interest of the character domain, and divide the image into a foreground portion and a background portion, the meter wheel number is a white on a black background, and the foreground portion is set to a white color.
  • the background portion is set to black.
  • the interference removal module 340 is configured to remove interference points and interference blocks in the binarized image. As shown in FIG. 8 , specifically, by performing connectivity analysis on the binarized image, the area and height of the connected region are calculated, and the histograms of the area and height of the connected region are respectively obtained, and the interference block is removed by using the Gaussian distribution characteristic. In the Gaussian distribution curve shown in Fig. 13, the mean value is in the middle, and the probability that the data is distributed on both sides is small. Most of the data is located at [c-1.5 ⁇ sig, c+1.5 ⁇ sig], so the area outside the interval is high. The corresponding connected domain is the interference area.
  • the histogram distribution of the connected region features in the above uses the Gaussian distribution characteristic to effectively remove the interference block, and avoids the partial division or over-segmentation of the character region, and solves the problem that the character image is binarized, the interference point, and the block is removed. Improves segmentation accuracy and provides more reliable input characters for subsequent recognition models.
  • the character segmentation module 350 is configured to divide the character region.
  • the character segmentation module 350 uses the projection method to perform column projection on the image, sequentially separates a single character from left to right and determines the regional position thereof, and uses the method of connected region analysis to determine Whether there is a half-character, if there is a half-character, calculate the height of two half-characters, and use the half-character corresponding to the maximum height as the half-character to be recognized.
  • the half-character and the full-character discriminating in the present invention are respectively identified by using different models, thereby improving the accuracy of half-character recognition, and using different recognition models for the complete and half-character situations, effectively improving the accuracy of half-character recognition.
  • the character image normalization module 360 is configured to normalize the individual character regions into images of a specified size; the character recognition module 370 is configured to use the single character region content (ie, comparison data) with the meter numbers stored in the data storage unit 400. The model is compared and identified.
  • the method for meter reading based on image recognition according to the invention comprises the following steps:
  • the denoising algorithm for reading images in step S2 may adopt a Gaussian filtering algorithm, and the specific formula is:
  • f(i,j) is the original grayscale image
  • g(i,j) is the filtered image
  • i,j are the longitudinal and lateral coordinates of the image
  • h(u,v) is the Gaussian kernel, as shown in the following equation :
  • is the standard deviation
  • the Gaussian kernel is a 5 ⁇ 5 template, and the grayscale image is filtered, and then the image enhancement effect is processed on the reading image based on the histogram equalization method.
  • step S3 Determining the region of interest of the character specifically includes the following steps:
  • the upper boundary, the lower boundary, the left boundary, and the right boundary of the character domain can be determined through steps S31 and S32, and the rectangular frame of the character region of interest is determined by the boundary.
  • step S4 the threshold is determined according to the foreground part and the background part feature of the image, and the image is divided into a foreground part and a background part, the meter reading is white on a black background, the foreground part is set to white, the background part is set to black, and the foreground part is determined.
  • the method of the background partial threshold may be the Otsu method or the adaptive threshold method.
  • the specific steps of removing the interference point and the interference block in the binarized image in step S5 to obtain an optimized image include the following steps, as shown in FIG. 7:
  • the interference point is generally a relatively isolated foreground pixel area in the image, and scanning the foreground pixel and its eight neighborhoods, if the foreground pixel of the foreground pixel is in the foreground of the image block If there are too few pixels, the pixel is set as the background pixel.
  • step S52 Removing the interference block of the binarized image by using the connected region analysis method, based on the result of step S51.
  • the foreground pixel of the foreground portion is marked as a plurality of connected regions, and the area and height of each connected region are calculated, and the area and height histograms are respectively counted. Since the character regions have similar areas and heights, The heterogeneous points on either side of the histogram distribution curve are removed by the area and height histogram surface distribution.
  • the heterogeneous points may be determined to be outside the range of [c-1.5 ⁇ sig, c+1.5 ⁇ sig] or determined according to statistical empirical values, and the blocks corresponding to the heterogeneous points are interference blocks. (as shown in Figure 8)
  • Step S6 includes the following steps as shown in FIG. 9:
  • S61 Scan the columns of the optimized image in sequence, obtain a vertical projection of the optimized image, and identify a character region therein. (as shown in block S61 in Fig. 9)
  • S62 Determine an identification character in the character area. Find the two peaks in the vertical projection from left to right in the vertical projection. The positions corresponding to the two peaks are determined as the left and right boundaries of the single character, and the peaks are searched from left to right in turn until all the individual characters are separated. The position of the recognized character of all character regions in the character field is determined. (as shown in block S62 in Fig. 9)
  • step S8 a single character to be recognized is compared with the digital data stored in the data storage unit to obtain corresponding identification data, as shown in FIG. Including the following steps:

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Character Input (AREA)
  • Character Discrimination (AREA)

Abstract

一种基于图像识别的计量表抄表装置及对应的方法,装置包括用于采集读数图像的图像采集单元(100)、数据处理单元(200)、数据存储单元(400)、图像识别单元(300)和数字显示单元(500)。通过梯度图像的平滑后水平、垂直投影曲线,查找波峰、波谷的方法,能够稳定地确定字符图像的上下、左右边界;通过统计连通区域特征的直方图分布,利用高斯分布特性,有效地去除了干扰块,避免字符的部分区域被欠分割或过分割;通过判别的半字符、全字符,分别使用不同的模型进行识别,提高了半字符识别准确率。

Description

一种基于图像识别的计量表抄表装置及其方法 技术领域
本发明涉及抄表领域,尤其是一种基于图像识别的计量表抄表装置及其方法。
背景技术
随着物联网技术的发展与成熟,使得跨产业,跨领域技术和业务融合成为现实。智能仪表发展迅速,具有远程自动抄表功能的智能电子表正逐步进入千家万户。基于拍照的抄表系统具有平时不工作,安装方便,使用寿命长,抗干扰能力强等优势,必将成为今后远传抄表系统的发展方向。
中国发明专利CN200410066507,专利名称为:一种远程无线抄表方法,其利用拍照装置拍摄电度表上的度数信息,以获得表示度数信息的原始图像。将该原始图像转化为相应的灰度图像和二值图像;预处理部分对二值图像进行图像去噪、倾斜校正、去除杂散像素以及滤波处理。然后进行字符分割,获得单个图像,对单个图像进行特征提取,采用欧几里得的方法识别字符信息,最终获得度数信息。
中国发明专利CN201420159056,专利名称为:一种基于公众服务平台移动终端的燃气抄表系统,其使用公众服务平台终端与缴费机构相连的服务器实现抄表功能。其中读数识别模块对图像信息中的数字所在区域进行定位,然后对显示数据进行单个数字分割处理,再对分割出的单个数字进行识别,最后通过数据人工确认及修正模块对单个数字进行确认或错误更正。
专利CN200410066507的预处理部分对二值图像进行去噪,倾斜校正,去除杂散点滤波后进行字符分割,这种处理方法对于由于气表表具不干净,或拍照 头不干净,光线干扰等原因造成的不完全清晰的图像不能够满足分割要求。图像中除了杂散点,还有各类干扰块影响字符分割效果。欧几里得距离识别算法虽然利用了图像的特征,识别算法简单,但是通过特征向量间的距离判别特征向量类别标签,没有基于统计先验知识。
专利CN201420159056对图像信息中的数字所在区域进行定位,然后对显示数据进行单个数字分割处理,再对分割出的单个数字进行识别,最后通过数据人工确认及修正模块对单个数字进行确认或错误更正。由于滚轮的机械转动,读数的进位机制,半字符显现时而会出现,该专利没有能满足半字符的识别需求。
发明内容
鉴于上述状况,有必要提供一种稳定可靠、识别准确率高且可识别半字符的基于图像识别的计量表抄表装置及其方法。
为解决上述技术问题,提供一种基于图像识别的计量表抄表装置,包括:
图像采集单元,用于对计量表滚轮数字进行拍照,采集得到读数图像扫描数据;
数据处理单元,用于对读数图像扫描数据进行处理得到读数数字图像;
数据存储单元,用于预先存储计量表数字模型;
图像识别单元,用于将读数数字图像与预先存储的计量表数字模型进行对比,得到识别数据;
数字显示单元;用于显示经对比后的得到的识别数据;
所述图像采集单元通过拍照采集计量表的滚轮数字后得到读数图像扫描数据,采集的扫描数据经数据处理单元进行处理后将处理数据经图像识别单元与数据存储单元的计量表数字模型进行对比后,得到识别数据,最后通过数 字显示单元将识别数据显示出来。
在本发明上述基于图像识别的计量表抄表装置中,所述数据存储单元(400)包括完整字符数字模型和半字符数字模型。
在本发明上述基于图像识别的计量表抄表装置中,所述图像识别单元(300)包括:
预处理模块,用于对读数数字图像进行去噪和对比度提升处理;
字符域感兴趣区域检测模块,用于确定字符域感兴趣区域;
图像二值化模块,用于将所述字符域感兴趣区域的灰度图像进行二值化处理;
干扰去除模块,用于去除二值化后字符域感兴趣区域图像中的干扰点和干扰块;
字符分割模块,用于从优化后的字符域感兴趣区域二值化图像分割出若干个字符区域;
字符图像归一化模块,用于将每个字符区域统一归一化为指定大小的图像;
字符识别模块,用于将单个字符区域内容与数据存储单元(400)中存储的计量表数字模型进行对比识别。
在本发明上述基于图像识别的计量表抄表方法中,一种基于图像识别的计量表抄表方法,包括如下步骤:
S1、对计量表滚轮数字进行拍照,得到读数数字图像;
S2、对读数数字图像进行去噪和对比度提升处理;
S3、确定字符域感兴趣区域;
S4、将所述字符域感兴趣区域灰度图像进行二值化,并将其分为前景部分和背景部分;
S5、对字符域二值化图像执行多次干扰去除算法,去除二值化图像中的干扰 点和干扰块得到优化图像;
S6、将优化图像进行列分割得到若干个字符区域,并判断每个字符区域中待识别字符并确定其在字符域图像的位置;
S7、将每个待识别字符区域图像统一归一化为指定大小的图像;
S8、再将每个待识别字符区域图像,与数据存储单元中存储的计量表数字模型进行对比识别,得到对应的识别数据;
S9、最后将识别数据通过数字显示单元显出出来。
在本发明上述基于图像识别的计量表抄表方法中,所述步骤S2中对读数数字图像进行去噪,采用高斯滤波算法,其具体公式为:
g(i,j)=f(i,j)*h(u,v)
其中f(i,j)为原始灰度图像,g(i,j)为滤波后图像,i,j分别是图像的纵向,横向坐标,h(u,v)是高斯核,如下式所示:
Figure PCTCN2016105763-appb-000001
其中σ是标准方差;所述步骤S2中对读数图像进行提升对比度处理方法为采用直方图均衡化方法。
在本发明上述基于图像识别的计量表抄表方法中,所述步骤S3中包括如下步骤:
S31、获取S2处理后图像的垂直梯度图像,依次扫描图像的每一行,统计每一行前景像素点的个数,获取垂直梯度图像的水平投影,并对其平滑处理,通过平滑后水平投影图像确定图像的上边界和下边界;
S32、获取S2处理后图像的水平梯度图像,依次扫描图像的每一例,统计每一例的前景像素点个数,获取水平梯度图像的垂直投影,并对其平滑处理,通过平滑后的垂直投影图像确定图像的左边界和右边界。
在本发明上述基于图像识别的计量表抄表方法中,所述步骤S4中根据图像 的前景部分和背景部分特征确定阈值,将字符域感兴趣区域灰度图像二值化,前景部分设置为白色,背景部分设置为黑色。
在本发明上述基于图像识别的计量表抄表方法中,确定前景部分和背景部分阈值的方法为大津法或自适应阈值方法。
在本发明上述基于图像识别的计量表抄表方法中,所述步骤S5中干扰块去除步骤S52具体包括如下步骤:
S521、利用连通域分析法获取二值化图像的所有连通区域并标记;
S522、计算各个连通域的面积和高度;
S523、统计所有连通域面积和高度的直方图;
S524、通过直方图分布确定均值和方差;
S525、利用直方图计算的均值和方差确定有效范围的边界阈值;
S526、有效范围之外的面积高度对应的连通域为干扰块。
在本发明上述基于图像识别的计量表抄表方法中,所述步骤S6统计字符域图像的每列前景像素点个数,得到垂直投影图像,对垂直投影图像进行列分割,确定滚轮数字单个字符所在的区域位置。
在本发明上述基于图像识别的计量表抄表方法中,所述步骤S8包括如下步骤:
S81、判断待识别字符区域为半字符或完整字符;
S82、如果是半字符,将待识别的半字符与数据库中半字符数字模型进行对比,得到识别数据;
S83、如果是完整字符,将待识别完整字符与数据库中完整字符数字模型进行对比,得到识别数字。
本发明基于图像识别的计量表抄表装置及其方法的有益效果是:
1、通过梯度图像的平滑后水平,垂直投影曲线,查找波峰,波谷的方法能 够稳定地确定字符图像的上下,左右边界,解决了字符域感兴趣区域图像切分准确度及稳定性问题。
2、统计连通区域特征的直方图分布,利用高斯分布特性,有效地去除了干扰块,避免字符部分区域被欠分割或过分割,解决了字符图像二值化后出现的干扰点,块问题,提高了分割准确度,为后续识别模型提供更可靠的输入字符。
3、判别的半字符,全字符分别使用不同的模型进行识别,提高了半字符识别准确率,针对完整,半字符情形,使用不同的识别模型,有效提高了半字符识别准确率。
附图说明
图1是本发明基于图像识别的计量表抄表装置中结构方框图;
图2是本发明基于图像识别的计量表抄表装置中图像识别单元结构方法图;
图3是本发明基于图像识别的计量表抄表方法流程结构图;
图4是本发明基于图像识别的计量表抄表方法步骤3中字符域感兴趣区域检测流程图;
图5是本发明基于图像识别的计量表抄表方法步骤3中字符域上下边界检测流程图;
图6是本发明基于图像识别的计量表抄表方法步骤3中字符域左右边界检测流程图;
图7是本发明基于图像识别的计量表抄表方法步骤5中二值化字符图像干扰去除流程图;
图8是本发明基于图像识别的计量表抄表其方法步骤5中二值化字符图像干扰块去除流程图;
图9是本发明基于图像识别的计量表抄表方法步骤6中单个字符检测流程 图;
图10是本发明基于图像识别的计量表抄表方法中步骤8中半字符识别流程图;
图11是本发明基于图像识别的计量表抄表方法中字符域感兴趣区域二值化图像的水平梯度图像中检测上下边界示意图;
图12是本发明基于图像识别的计量表抄表方法中字符域感兴趣区域二值化图像的垂直梯度图像中检测左右边界示意图;
图13是本发明基于图像识别的计量表抄表方法中字符域感兴趣区域二值化图像中干扰块去除使用的高斯分布曲线特性示意图。
具体实施方式
下面将结合附图及实施例对本发明的基于图像识别的计量表抄表装置及其方法作进一步的详细说明。
实施例一
本发明实施例的一种基于图像识别的计量表抄表装置,如图1所示,包括图像采集单元100,图像采集单元通过摄像头装置对计量表的滚轮数字进行拍照,拍照得到的读数图像扫描数据通过数据传输至数据处理单元200处,数据处理单元200对读数图像扫描数据进行处理后得到读数数字图像;数据存储单元400,用于预先存储计量表的计量表数字模型;图像识别单元300,用于对读数数字图像进行一系列识别处理,通过将处理后的读数数字图像与预先存储计量表数字模型进行对比,得到识别数据;数字显示单元500;用于显示经对比后得到的识别数据;本发明基于图像识别的计量表抄表方法具体过程为:通过图像采集单元100拍照采集计量表的滚轮数字后得到读数图像扫描数据,采集的读数图像扫描数据经数据处理单元200进行处理后将读数数字图像经图像识别 单元300处理后与数据存储单元400的计量表数字模型进行对比,得到识别数据,最后通过数字显示单元500将识别数据显示在显示屏上。
本发明基于图像识别的计量表抄表装置中,数据存储单元400包括完整字符数字模型和半字符数字模型,完整字符数字模型存储计量表滚轮数字的完整数字,半字符数字模型存储计量表滚轮数字的半字符数字,通过半字符数字模型对比数据,有效提高了半字符识别准确率。
如图2所示,图像识别单元300具体包括预处理模块310,字符域感兴趣区域检测模块320、图像二值化模块330、干扰去除模块340、字符分割模块350、字符图像归一化模块360和字符识别模块370。本发明预处理模块310用于对读数图像进行去噪和提升对比度处理;如图5所示,字符感兴趣区域检测模块320具体为检测预处理后图像中字符图像的上、下、左、右边界来确定字符域感兴趣区域,上、下边界检测首先获取预处理后图像的垂直梯度图像,并依次扫描垂直梯度图像的每一行,统计每一行前景点像素的个数,获取垂直梯度图像的水平投影,并对其平滑处理。由于垂直梯度图像中上、下边界的边缘处投影值最大,从水平投影中寻找对应的两个极大波峰值点,这两个波峰值点对应的行坐标分别是字符感兴趣区域的上边界和下边界,如图11所示,第一个极大波峰值位置为上边界top,第二个极大波峰值位置为下边界down。
当上、下边界极大波峰值确定后,如图6所示,获取预处理后图像的水平梯度图像,并依次扫描水平梯度图像的每列,统计每列的前景像素点的个数,获取水平梯度图像的垂直投影,并对其平滑处理。由于垂直投影图像左边平坦区域梯度非常小,对应垂直投影的波谷部分,在垂直投影图像左边查找波谷区域后的波峰位置,即为字符区域的左边界;垂直投影图像右边全部为字符,因此右边界即为垂直投影图像的最右侧边缘位置,如图12所示,波谷区域后的第一个波峰为字符区域的左边界位置。
图像二值化模块330用于将所述字符域感兴趣区域灰度图像进行二值化处理,将图像分为前景部分和背景部分,计量表滚轮数字为黑底白字,将前景部分设置为白色,背景部分设置为黑色。
干扰去除模块340用于去除二值化图像中的干扰点和干扰块。如果8所示,具体为,通过对二值化图像进行连通区域分析,再计算出连通区域的面积和高度,再分别获得连通区域的面积和高度的直方图,利用高斯分布特性去除干扰块,如图13所示的高斯分布曲线中,均值位于中间,数据分布在两侧的概率很小,大部分数据位于[c-1.5×sig,c+1.5×sig],所以区间外的面积,高度对应的连通域即为干扰区域。上述中的连通区域特征的直方图分布,利用高斯分布特性,有效地去除了干扰块,避免字符的部分区域被欠分割或过分割,解决了字符图像二值化后,干扰点,块的去除,提高了分割准确度,为后续识别模型提供更可靠的输入字符。
字符分割模块350用于分割字符区域,本发明中字符分割模块350利用投影的方法,图像进行列投影,从左到右依次分割出单个字符并确定其区域位置,同时利用连通区域分析的方法判定出是否有半字符,如果有半字符,计算出两个半字符的高度,使用最大高度对应的半字符为待识别半字符。本发明中判别的半字符,全字符分别使用不同的模型进行识别,提高了半字符识别准确率,针对完整,半字符情形,使用不同的识别模型,有效提高了半字符识别准确率
字符图像归一化模块360用于将单独字符区域统一归一化为指定大小的图像;字符识别模块370用于将单个字符区域内容(即对比数据)与数据存储单元400中存储的计量表数字模型进行对比识别。
实施例二
本发明一种基于图像识别的计量表抄表方法,如图3所示,包括如下步骤:
S1、通过摄像头对计量表滚轮数字进行拍照,得到读数图像,执行步骤 S2;
S2、对读数图像进行去噪和提升对比度处理;
S3、确定字符域感兴趣区域;
S4、将所述字符域感兴趣区域进行灰度图像二值化处理后得到字符图像,并将其分为前景部分和背景部分,前景部分设置为白色,背景部分设置为黑色;
S5、对字符图像执行多次干扰去除算法,去除二值化图像中的干扰点和干扰块得到优化图像;
S6、将优化图像进行分割得到若干字符区域,并判断字符区域待识别的单个字符并确定其位置;
S7、将待识别字符区域图像统一归一化为指定大小的图像;
S8、再将单个待识别字符区域图像与数据存储单元中存储的计量表数字模型进行对比识别,得到对应的识别数据;
S9、最后将识别数据通过数字显示单元显出出来。
本发明基于图像识别的计量表抄表方法中,所述步骤S2中对读数图像进行去噪算法可采用高斯滤波算法,其具体公式为:
g(i,j)=f(i,j)*h(u,v)
其中f(i,j)为原始灰度图像,g(i,j)为滤波后图像,i,j分别是图像的纵向,横向坐标,h(u,v)是高斯核,如下式所示:
Figure PCTCN2016105763-appb-000002
其中σ是标准方差;具体步骤中,高斯核为5×5模板,并对灰度图像进行滤波处理,然后基于直方图均衡化的方法对读数图像进行图像增强效果处理。
本发明基于图像识别的计量表抄表方法中,如图4至图6所示,步骤S3 中确定字符感兴趣区域具体包括如下步骤:
S31、获取预处理后图像的垂直梯度图像,依次扫描图像的每一行,统计每一行前景像素点的个数,获取垂直梯度图像的水平投影,并对其平滑处理,在水平投影图像中确定字符域的上边界和下边界。水平投影图像的上下边界边缘处投影值最大,从水平投影中寻找对应的两个极大波峰点,这两个波峰对应的行坐标分别是字符感兴趣区域的上边界和下边界。(如图5所示)
S32、获取预处理后图像的水平梯度图像,依次扫描图像的每一例,统计每一例的前景像素点个数,获取水平梯度图像的垂直投影,并对其平滑处理,在垂直投影图像中确定字符域的左边界和右边界。垂直投影图像的左边平坦区域梯度非常小,对应垂直投影的波谷部分,在垂直投影图像的左边查找波谷区域后的波峰位置,即为字符区域的左边界;由于拍照后的读数图像的右边全部为字符,所以右边界即为水平梯度图像的最右侧边缘。(如图6所示)
通过步骤S31和S32即可确定字符域的上边界、下边界、左边界和右边界,通过边界确定字符感兴趣区域矩形框。
步骤S4中根据图像的前景部分和背景部分特征确定阈值,将图像分为前景部分和背景部分,计量表读数为黑底白字,将前景部分设置为白色,背景部分设置为黑色;确定前景部分和背景部分阈值的方法可以为大津法或自适应阈值方法。
步骤S5中去除二值化处理后图像中的干扰点和干扰块得到优化图像的具体步骤包括如下步骤,如图7所示:
S51、利用邻域分析法去除二值化图像的干扰点,干扰点一般为图像中相对孤立的前景像素区域,扫描前景像素及其八邻域,如果该前景像素的八邻域图像块的前景像素过少,则该像素点置为背景像素点。
S52、利用连通区域分析法去除二值化图像的干扰块,基于步骤S51的结果 后利用连通区域分析,将前景部分的前景像素点标记为多个连通区域,并计算每个连通区域的面积和高度,分别统计面积和高度直方图,由于字符区域具有相似范围的面积和高度,通过面积和高度直方图面分布,去除位于直方图分布曲线两侧的异类点。异类点可确定为[c-1.5·sig,c+1.5·sig]范围之外或依据统计经验值确定参数,这些异类点对应的块即为干扰块。(如图8所示)
步骤S6中包括图9所示的如下步骤:
S61、依次扫描优化后图像的列,获取优化图像的垂直投影,识别其中的字符区域。(如图9中框S61所示)
S62、判断字符区域中的识别字符。在垂直投影图中从左至右依次查找垂直投影中的两个波峰,两个波峰对应的位置确定为单个字符的左右边界,依次从左向右查找波峰,直到所有单个字符被分离出来,即确定了所有字符区域的识别字符在字符域的位置。(如图9中框S62所示)
本发明基于图像识别的计量表抄表方法中,步骤S8中将单个待识别字符,与数据存储单元中存储的计量表数字数据进行对比识别,得到对应的识别数据,如图10所示,具体包括如下步骤:
S81、判别待识别字符是否含有半字符(图10中框S81);
S82、如果是半字符,将半字符输入至半字符数字模型进行对比,得到识别数字(图10中框S82);
S83、如果是完整字符,将完整字符输入到完整数字模型进行对比,得到识别数字(图10中框S83)。
以上所述,仅是本发明的较佳实施例而已,并非对本发明作任何形式上的限制,虽然本发明已以较佳实施例揭露如上,然而并非用以限定本发明,任何熟悉本专业的技术人员,在不脱离本发明技术方案范围内,当可利用上述揭示的技术内容做出些许更动或修饰为等同变化的等效实施例,但凡是未脱离本发 明技术方案内容,依据本发明的技术实质对以上实施例所作的任何简单修改、等同变化与修饰,均仍属于本发明技术方案的范围内。

Claims (11)

  1. 一种基于图像识别的计量表抄表装置,其特征在于,包括:
    图像采集单元(100),用于对计量表滚轮数字进行拍照,采集得到读数图像扫描数据;
    数据处理单元(200),用于对读数图像扫描数据进行处理得到读数数字图像;
    数据存储单元(400),用于预先存储计量表数字模型;
    图像识别单元(300),用于将读数数字图像与预先存储的计量表数字模型进行对比,得到识别数据;
    数字显示单元(500);用于显示经对比后的得到的识别数据;
    所述图像采集单元(100)通过拍照采集计量表的滚轮数字后得到读数图像扫描数据,采集的扫描数据经数据处理单元(200)进行处理后将处理数据经图像识别单元(300)与数据存储单元(400)的计量表数字模型进行对比后,得到识别数据,最后通过数字显示单元(500)将识别数据显示出来。
  2. 如权利要求1所述的基于图像识别的计量表抄表装置,其特征在于,所述数据存储单元(400)包括完整字符数字模型和半字符数字模型。
  3. 如权利要求1所述的基于图像识别的计量表抄表装置,其特征在于,所述图像识别单元(300)包括:
    预处理模块(310),用于对读数数字图像进行去噪和对比度提升处理;
    字符域感兴趣区域检测模块(320),用于在经过预处理模块处理过的区域里确定字符域感兴趣区域;
    图像二值化模块(330),用于将所述字符域感兴趣区域的灰度图像进行二值化处理;
    干扰去除模块(340),用于去除二值化后字符域感兴趣区域图像中的干扰点 和干扰块;
    字符分割模块(350),用于从优化后的字符域感兴趣区域二值化图像分割出若干个字符区域;
    字符图像归一化模块(360),用于将每个字符区域统一归一化为指定大小的图像;
    字符识别模块(370),用于将单个字符区域内容与数据存储单元(400)中存储的计量表数字模型进行对比识别。
  4. 一种基于图像识别的计量表抄表方法,其特征在于,包括如下步骤:
    S1、对计量表滚轮数字进行拍照,得到读数数字图像;
    S2、对读数数字图像进行去噪和对比度提升处理;
    S3、确定字符域感兴趣区域;
    S4、将所述字符域感兴趣区域灰度图像进行二值化,并将其分为前景部分和背景部分;
    S5、对字符域二值化图像执行多次干扰去除算法,去除二值化图像中的干扰点和干扰块得到优化图像;
    S6、将优化图像进行列分割得到若干个字符区域,并判断每个字符区域中待识别字符并确定其在字符域图像的位置;
    S7、将每个待识别字符区域图像统一归一化为指定大小的图像;
    S8、再将每个待识别字符区域图像,与数据存储单元中存储的计量表数字模型进行对比识别,得到对应的识别数据;
    S9、最后将识别数据通过数字显示单元显出出来。
  5. 如权利要求4所述的基于图像识别的计量表抄表方法,其特征在于:所述步骤S2中对读数数字图像进行去噪,采用高斯滤波算法,其具体公式为:
    g(i,j)=f(i,j)*h(u,v)
    其中f(i,j)为原始灰度图像,g(i,j)为滤波后图像,i,j分别是图像的纵向,横向坐标,h(u,v)是高斯核,如下式所示:
    Figure PCTCN2016105763-appb-100001
    其中σ是标准方差;所述步骤S2中对读数数字图像进行提升对比度处理方法为采用直方图均衡化方法。
  6. 如权利要求4所述的基于图像识别的计量表抄表方法,其特征在于:所述步骤S3中包括如下步骤:
    S31、获取S2处理后图像的垂直梯度图像,依次扫描图像的每一行,统计每一行前景像素点的个数,获取垂直梯度图像的水平投影,并对其平滑处理,通过平滑后水平投影图像确定图像的上边界和下边界;
    S32、获取S2处理后图像的水平梯度图像,依次扫描图像的每一例,统计每一例的前景像素点个数,获取水平梯度图像的垂直投影,并对其平滑处理,通过平滑后的垂直投影图像确定图像的左边界和右边界。
  7. 如权利要求6所述的基于图像识别的计量表抄表方法,其特征在于:所述步骤S4中根据图像的前景部分和背景部分特征确定阈值,将字符域感兴趣区域灰度图像二值化,前景部分设置为白色,背景部分设置为黑色。
  8. 如权利要求7所示的基于图像识别的计量表抄表方法,其特征在于:确定前景部分和背景部分阈值的方法为大津法或自适应阈值方法。
  9. 如权利要求8所示的基于图像识别的计量表抄表方法,其特征在于:所述步骤S5中干扰块去除步骤S52具体包括如下步骤:
    S521、利用连通域分析法获取二值化图像的所有连通区域并标记;
    S522、计算各个连通域的面积和高度;
    S523、统计所有连通域面积和高度的直方图;
    S524、通过直方图分布确定均值和方差;
    S525、利用直方图计算的均值和方差确定有效范围的边界阈值;
    S526、有效范围之外的面积高度对应的连通域为干扰块。
  10. 如权利要求9所示的基于图像识别的计量表抄表方法,其特征在于:所述步骤S6统计字符域图像的每列前景像素点个数,得到垂直投影图像,对垂直投影图像进行列分割,确定滚轮数字单个字符所在的区域位置。
  11. 如权利要求10所示的基于图像识别的计量表抄表方法,其特征在于:所述步骤S8包括如下步骤:
    S81、判断待识别字符区域为半字符或完整字符;
    S82、如果是半字符,将待识别的半字符与数据库中半字符数字模型进行对比,得到识别数据;
    S83、如果是完整字符,将待识别完整字符与数据库中完整字符数字模型进行对比,得到识别数字;
PCT/CN2016/105763 2016-07-29 2016-11-14 一种基于图像识别的计量表抄表装置及其方法 WO2018018788A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610614798.8A CN106228159A (zh) 2016-07-29 2016-07-29 一种基于图像识别的计量表抄表装置及其方法
CN201610614798.8 2016-07-29

Publications (1)

Publication Number Publication Date
WO2018018788A1 true WO2018018788A1 (zh) 2018-02-01

Family

ID=57536338

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/105763 WO2018018788A1 (zh) 2016-07-29 2016-11-14 一种基于图像识别的计量表抄表装置及其方法

Country Status (2)

Country Link
CN (1) CN106228159A (zh)
WO (1) WO2018018788A1 (zh)

Cited By (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110060260A (zh) * 2019-04-12 2019-07-26 南京信息工程大学 一种图像处理方法及系统
CN110197164A (zh) * 2019-06-04 2019-09-03 深圳市摩西尔电子有限公司 一种led屏拼接单元编码和编码识别的方法及装置
CN110287967A (zh) * 2019-06-28 2019-09-27 哈尔滨工业大学 一种基于图像的数字及机械表数字识别方法
CN110689017A (zh) * 2018-07-06 2020-01-14 康姆德润达(无锡)测量技术有限公司 一种滤膜id图像识别系统
CN110991437A (zh) * 2019-11-28 2020-04-10 北京嘉楠捷思信息技术有限公司 字符识别方法及其装置、字符识别模型的训练方法及其装置
CN111178395A (zh) * 2019-12-12 2020-05-19 平高集团有限公司 一种隔离开关状态识别方法及装置
CN111242135A (zh) * 2020-03-23 2020-06-05 贵州电网有限责任公司 一种用于变电站二次屏柜的自动接线图生成设备及其操作方法
CN111401381A (zh) * 2020-03-27 2020-07-10 白志青 慢门成像识别仪表数字监控系统及监测方法
CN111539330A (zh) * 2020-04-17 2020-08-14 西安英诺视通信息技术有限公司 一种基于双svm多分类器的变电站数显仪表识别方法
CN111881913A (zh) * 2019-07-05 2020-11-03 深圳数字生命研究院 图像识别方法及装置、存储介质和处理器
CN111914847A (zh) * 2020-07-23 2020-11-10 厦门商集网络科技有限责任公司 一种基于模板匹配的ocr识别方法及其系统
CN112434693A (zh) * 2020-11-03 2021-03-02 辽宁长江智能科技股份有限公司 一种数字式水表识别方法与系统
CN112489071A (zh) * 2020-11-03 2021-03-12 辽宁长江智能科技股份有限公司 一种指针水表识别方法与系统
CN112712465A (zh) * 2020-12-31 2021-04-27 四川长虹网络科技有限责任公司 一种优化拍照式抄表终端通信数据量的方法和系统
CN112785508A (zh) * 2019-11-11 2021-05-11 珠海金山办公软件有限公司 一种电子文档图片去噪的方法和装置
CN113076945A (zh) * 2021-03-17 2021-07-06 华夏芯(北京)通用处理器技术有限公司 一种基于改进ransac的摄像直读抄表仪异常点剔除方法
CN113269194A (zh) * 2021-06-11 2021-08-17 四川长虹网络科技有限责任公司 读数表不完整字符识别方法以及读数表字符识别方法
CN113554022A (zh) * 2021-06-07 2021-10-26 华北电力科学研究院有限责任公司 电力仪器检测试验数据的自动获取方法及装置
CN114067095A (zh) * 2021-11-29 2022-02-18 黄河勘测规划设计研究院有限公司 基于水尺字符检测识别的水位识别方法
CN114519694A (zh) * 2021-12-28 2022-05-20 河南大学 基于深度学习的七段数码管液晶显示屏识别方法及系统
CN114926839A (zh) * 2022-07-22 2022-08-19 富璟科技(深圳)有限公司 基于rpa和ai的图像识别方法及电子设备
CN114998887A (zh) * 2022-08-08 2022-09-02 山东精惠计量检测有限公司 一种电能计量表智能识别方法
CN116645682A (zh) * 2023-07-24 2023-08-25 济南瑞泉电子有限公司 一种水表表盘数字识别方法及系统
CN117095423A (zh) * 2023-10-20 2023-11-21 上海银行股份有限公司 一种银行单据字符的识别方法及装置

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107392068B (zh) * 2017-07-25 2021-01-15 朱宸 一种计量数据识别的方法和装置
CN107452144A (zh) * 2017-08-17 2017-12-08 成都工业学院 自动计费方法与装置
CN107688802A (zh) * 2017-09-29 2018-02-13 深圳市玛塔创想科技有限公司 一种基于图像识别的简易编程方法及装置
CN107730478A (zh) * 2017-10-17 2018-02-23 云南电网有限责任公司电力科学研究院 一种电能计量自动化终端的外形检测方法及装置
CN108171282B (zh) * 2017-12-29 2021-08-31 安徽慧视金瞳科技有限公司 一种黑板笔迹自动合成方法
CN107958253A (zh) * 2018-01-18 2018-04-24 浙江中控技术股份有限公司 一种图像识别的方法和装置
CN108072443A (zh) * 2018-01-26 2018-05-25 上海康斐信息技术有限公司 一种体重检测装置测试方法及系统
CN110263778A (zh) * 2018-03-12 2019-09-20 中移物联网有限公司 一种基于图像识别的抄表方法及装置
CN108597204A (zh) * 2018-05-22 2018-09-28 广州市暨联牧科信息科技有限公司 一种智能抄表系统及其实现方法
CN110969063B (zh) * 2018-09-30 2023-08-18 中移物联网有限公司 抄表方法、抄表设备、远程抄表装置、系统及存储介质
CN109886276B (zh) * 2019-02-18 2023-05-09 福州视驰科技有限公司 一种表盘滚动数字字符的半字判断方法
CN110232382B (zh) * 2019-02-18 2023-04-07 福州视驰科技有限公司 一种表盘滚动数字字符的多位半字跳转判断与识别方法
CN110517281A (zh) * 2019-08-19 2019-11-29 温州大学 一种从高到低扫描一维投影图分割目标的方法
CN112446262A (zh) * 2019-09-02 2021-03-05 深圳中兴网信科技有限公司 文本分析方法、装置、终端和计算机可读存储介质
CN110728687B (zh) * 2019-10-15 2022-08-02 卓尔智联(武汉)研究院有限公司 文件图像分割方法、装置、计算机设备和存储介质
CN111568199B (zh) * 2020-02-28 2023-11-07 佛山市云米电器科技有限公司 接水容器的识别方法、系统及存储介质
CN112818993A (zh) * 2020-03-30 2021-05-18 深圳友讯达科技股份有限公司 一种用于摄像直读抄表器的字轮读数表端识别方法及设备
CN113723396A (zh) * 2020-05-21 2021-11-30 安徽小眯当家信息技术有限公司 一种数码管小数点位置识别方法及装置
CN111639643B (zh) * 2020-05-22 2023-06-27 深圳市赛为智能股份有限公司 字符识别方法、装置、计算机设备及存储介质
CN111814795A (zh) * 2020-06-05 2020-10-23 北京嘉楠捷思信息技术有限公司 字符分割方法、装置及计算机可读存储介质
CN111914717B (zh) * 2020-07-24 2024-06-21 安徽华速达电子科技有限公司 一种基于抄表数据智能识别的数据录入方法及装置
CN112017129A (zh) * 2020-08-28 2020-12-01 湖南尚珂伊针纺有限公司 一种高效率袜子数字模型生产装置
CN113486892B (zh) * 2021-07-02 2023-11-28 东北大学 基于智能手机图像识别的生产信息采集方法及系统
CN113343988A (zh) * 2021-07-06 2021-09-03 北矿机电科技有限责任公司 摇床矿带识别方法、系统及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101000652A (zh) * 2006-12-31 2007-07-18 沈阳工业大学 流量计费表数字远传图像自动识别方法及数字远传抄表系统
CN101673338A (zh) * 2009-10-09 2010-03-17 南京树声科技有限公司 基于多角度投影的模糊车牌识别方法
CN101877050A (zh) * 2009-11-10 2010-11-03 青岛海信网络科技股份有限公司 一种车牌字符的自动提取方法
CN102236788A (zh) * 2010-04-20 2011-11-09 荣科科技股份有限公司 电力电能表图像自动识别方法
US20130050498A1 (en) * 2011-08-31 2013-02-28 Next Future, LLC Analog utility meter reading

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101079094A (zh) * 2007-04-30 2007-11-28 中国科学院合肥物质科学研究院 用于远程自动抄表系统的计量表读数识别装置
CN101806601A (zh) * 2010-03-31 2010-08-18 昆明利普机器视觉工程有限公司 一种基于图像识别技术的计量表手持抄表终端和抄表方法
CN103209211B (zh) * 2013-03-06 2015-12-02 江苏运赢物联网产业发展有限公司 一种计量表读数识别的方法、计量表自助抄表系统及其抄表方法
CN105528601A (zh) * 2016-02-25 2016-04-27 华中科技大学 基于接触式传感器的身份证图像采集、识别系统及采集识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101000652A (zh) * 2006-12-31 2007-07-18 沈阳工业大学 流量计费表数字远传图像自动识别方法及数字远传抄表系统
CN101673338A (zh) * 2009-10-09 2010-03-17 南京树声科技有限公司 基于多角度投影的模糊车牌识别方法
CN101877050A (zh) * 2009-11-10 2010-11-03 青岛海信网络科技股份有限公司 一种车牌字符的自动提取方法
CN102236788A (zh) * 2010-04-20 2011-11-09 荣科科技股份有限公司 电力电能表图像自动识别方法
US20130050498A1 (en) * 2011-08-31 2013-02-28 Next Future, LLC Analog utility meter reading

Cited By (37)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110689017A (zh) * 2018-07-06 2020-01-14 康姆德润达(无锡)测量技术有限公司 一种滤膜id图像识别系统
CN110060260A (zh) * 2019-04-12 2019-07-26 南京信息工程大学 一种图像处理方法及系统
CN110197164B (zh) * 2019-06-04 2023-04-18 深圳市摩西尔电子有限公司 一种led屏拼接单元编码和编码识别的方法及装置
CN110197164A (zh) * 2019-06-04 2019-09-03 深圳市摩西尔电子有限公司 一种led屏拼接单元编码和编码识别的方法及装置
CN110287967A (zh) * 2019-06-28 2019-09-27 哈尔滨工业大学 一种基于图像的数字及机械表数字识别方法
CN111881913A (zh) * 2019-07-05 2020-11-03 深圳数字生命研究院 图像识别方法及装置、存储介质和处理器
CN112785508A (zh) * 2019-11-11 2021-05-11 珠海金山办公软件有限公司 一种电子文档图片去噪的方法和装置
CN110991437A (zh) * 2019-11-28 2020-04-10 北京嘉楠捷思信息技术有限公司 字符识别方法及其装置、字符识别模型的训练方法及其装置
CN110991437B (zh) * 2019-11-28 2023-11-14 嘉楠明芯(北京)科技有限公司 字符识别方法及其装置、字符识别模型的训练方法及其装置
CN111178395B (zh) * 2019-12-12 2023-04-07 平高集团有限公司 一种隔离开关状态识别方法及装置
CN111178395A (zh) * 2019-12-12 2020-05-19 平高集团有限公司 一种隔离开关状态识别方法及装置
CN111242135A (zh) * 2020-03-23 2020-06-05 贵州电网有限责任公司 一种用于变电站二次屏柜的自动接线图生成设备及其操作方法
CN111401381A (zh) * 2020-03-27 2020-07-10 白志青 慢门成像识别仪表数字监控系统及监测方法
CN111539330A (zh) * 2020-04-17 2020-08-14 西安英诺视通信息技术有限公司 一种基于双svm多分类器的变电站数显仪表识别方法
CN111539330B (zh) * 2020-04-17 2023-03-24 西安英诺视通科技有限公司 一种基于双svm多分类器的变电站数显仪表识别方法
CN111914847B (zh) * 2020-07-23 2023-11-17 厦门商集网络科技有限责任公司 一种基于模板匹配的ocr识别方法及其系统
CN111914847A (zh) * 2020-07-23 2020-11-10 厦门商集网络科技有限责任公司 一种基于模板匹配的ocr识别方法及其系统
CN112489071A (zh) * 2020-11-03 2021-03-12 辽宁长江智能科技股份有限公司 一种指针水表识别方法与系统
CN112434693A (zh) * 2020-11-03 2021-03-02 辽宁长江智能科技股份有限公司 一种数字式水表识别方法与系统
CN112712465B (zh) * 2020-12-31 2023-08-04 四川长虹网络科技有限责任公司 一种优化拍照式抄表终端通信数据量的方法和系统
CN112712465A (zh) * 2020-12-31 2021-04-27 四川长虹网络科技有限责任公司 一种优化拍照式抄表终端通信数据量的方法和系统
CN113076945B (zh) * 2021-03-17 2024-05-14 华夏芯(北京)通用处理器技术有限公司 一种基于改进ransac的摄像直读抄表仪异常点剔除方法
CN113076945A (zh) * 2021-03-17 2021-07-06 华夏芯(北京)通用处理器技术有限公司 一种基于改进ransac的摄像直读抄表仪异常点剔除方法
CN113554022B (zh) * 2021-06-07 2024-04-12 华北电力科学研究院有限责任公司 电力仪器检测试验数据的自动获取方法及装置
CN113554022A (zh) * 2021-06-07 2021-10-26 华北电力科学研究院有限责任公司 电力仪器检测试验数据的自动获取方法及装置
CN113269194A (zh) * 2021-06-11 2021-08-17 四川长虹网络科技有限责任公司 读数表不完整字符识别方法以及读数表字符识别方法
CN114067095B (zh) * 2021-11-29 2023-11-10 黄河勘测规划设计研究院有限公司 基于水尺字符检测识别的水位识别方法
CN114067095A (zh) * 2021-11-29 2022-02-18 黄河勘测规划设计研究院有限公司 基于水尺字符检测识别的水位识别方法
CN114519694A (zh) * 2021-12-28 2022-05-20 河南大学 基于深度学习的七段数码管液晶显示屏识别方法及系统
CN114926839A (zh) * 2022-07-22 2022-08-19 富璟科技(深圳)有限公司 基于rpa和ai的图像识别方法及电子设备
CN114926839B (zh) * 2022-07-22 2022-10-14 富璟科技(深圳)有限公司 基于rpa和ai的图像识别方法及电子设备
CN114998887B (zh) * 2022-08-08 2022-10-11 山东精惠计量检测有限公司 一种电能计量表智能识别方法
CN114998887A (zh) * 2022-08-08 2022-09-02 山东精惠计量检测有限公司 一种电能计量表智能识别方法
CN116645682B (zh) * 2023-07-24 2023-10-20 济南瑞泉电子有限公司 一种水表表盘数字识别方法及系统
CN116645682A (zh) * 2023-07-24 2023-08-25 济南瑞泉电子有限公司 一种水表表盘数字识别方法及系统
CN117095423A (zh) * 2023-10-20 2023-11-21 上海银行股份有限公司 一种银行单据字符的识别方法及装置
CN117095423B (zh) * 2023-10-20 2024-01-05 上海银行股份有限公司 一种银行单据字符的识别方法及装置

Also Published As

Publication number Publication date
CN106228159A (zh) 2016-12-14

Similar Documents

Publication Publication Date Title
WO2018018788A1 (zh) 一种基于图像识别的计量表抄表装置及其方法
CN107545239B (zh) 一种基于车牌识别与车辆特征匹配的套牌检测方法
CN109816644B (zh) 一种基于多角度光源影像的轴承缺陷自动检测系统
CN111650220B (zh) 一种基于视觉的图文缺陷检测方法
CN106960208B (zh) 一种仪表液晶数字自动切分和识别的方法及系统
Yu et al. An approach to Korean license plate recognition based on vertical edge matching
WO2018145470A1 (zh) 一种图像检测方法和装置
CN109657632B (zh) 一种车道线检测识别方法
KR101403876B1 (ko) 차량 번호판 인식 방법과 그 장치
CN107491730A (zh) 一种基于图像处理的化验单识别方法
CN109409355B (zh) 一种新型变压器铭牌识别的方法及装置
WO2019000653A1 (zh) 一种图像目标识别方法及装置
CN111382704A (zh) 基于深度学习的车辆压线违章判断方法、装置及存储介质
CN101122952A (zh) 一种图片文字检测的方法
CN101122953A (zh) 一种图片文字分割的方法
Paunwala et al. A novel multiple license plate extraction technique for complex background in Indian traffic conditions
CN104463134B (zh) 一种车牌检测方法和系统
CN103310211A (zh) 一种基于图像处理的填注标记识别方法
CN110648330B (zh) 摄像头玻璃的缺陷检测方法
CN116071763B (zh) 基于文字识别的教辅图书智能校编系统
CN113033558B (zh) 一种用于自然场景的文本检测方法及装置、存储介质
CN110598566A (zh) 图像处理方法、装置、终端和计算机可读存储介质
CN109886168B (zh) 一种基于层阶的地面交通标志识别方法
Sun et al. A visual attention based approach to text extraction
KR20150108118A (ko) 영상 인식 기반 계량기 원격 검침 시스템

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16910367

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16910367

Country of ref document: EP

Kind code of ref document: A1