WO2023134791A2 - 一种环境安防工程监测数据管理方法及系统 - Google Patents

一种环境安防工程监测数据管理方法及系统 Download PDF

Info

Publication number
WO2023134791A2
WO2023134791A2 PCT/CN2023/086005 CN2023086005W WO2023134791A2 WO 2023134791 A2 WO2023134791 A2 WO 2023134791A2 CN 2023086005 W CN2023086005 W CN 2023086005W WO 2023134791 A2 WO2023134791 A2 WO 2023134791A2
Authority
WO
WIPO (PCT)
Prior art keywords
pixel
superpixel
block
edge point
gray
Prior art date
Application number
PCT/CN2023/086005
Other languages
English (en)
French (fr)
Other versions
WO2023134791A3 (zh
Inventor
周皓
Original Assignee
苏州迈创信息技术有限公司
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 苏州迈创信息技术有限公司 filed Critical 苏州迈创信息技术有限公司
Publication of WO2023134791A2 publication Critical patent/WO2023134791A2/zh
Publication of WO2023134791A3 publication Critical patent/WO2023134791A3/zh

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
    • H04N19/103—Selection of coding mode or of prediction mode
    • H04N19/11—Selection of coding mode or of prediction mode among a plurality of spatial predictive coding modes
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
    • H04N19/136—Incoming video signal characteristics or properties
    • H04N19/14—Coding unit complexity, e.g. amount of activity or edge presence estimation
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/169—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding
    • H04N19/17—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object
    • H04N19/176—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object the region being a block, e.g. a macroblock
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/593—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving spatial prediction techniques
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00—Television systems
    • H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Definitions

  • the invention relates to the technical field of data processing, in particular to a method and system for managing monitoring data of environmental security engineering.
  • the existing technology generally uses lossy compression to compress and store video data, which can greatly reduce the storage space.
  • the commonly used lossy compression method is predictive coding compression.
  • Predictive coding compression divides each frame of image into multiple random sizes Then use the feature values of the pixels in each macroblock to perform linear prediction. If the gray level distribution in the same rectangular macroblock is flat and the gray level correlation between pixels is strong, the linear prediction result is ideal. , if the gray level distribution in the same rectangular macroblock is more random and there are many mutation points, the accuracy of the linear prediction result is not high.
  • the present invention provides a monitoring data management method and system for environmental security engineering to solve the problem of inaccurate linear prediction results of some macroblocks with relatively large gray distribution randomness in the existing randomly divided macroblocks, which leads to low compression efficiency .
  • the fitting quality entropy value of each superpixel block in different directions is obtained; the minimum fitting quality entropy value is selected as The fitting entropy value of the superpixel block will obtain the direction of the minimum fitting quality entropy value as the target direction of the corresponding superpixel block;
  • the prediction model coefficient of each type of superpixel block is obtained;
  • each type of superpixel block in each single-channel image Using the prediction model coefficients of each type of superpixel block in each single-channel image and the gray value of pixels in each type of superpixel block to obtain the prediction bias of each type of superpixel block; according to each type of superpixel block in each single-channel image
  • the prediction offset of the pixel block, the prediction model coefficient and the gray value of the edge point obtain the predicted gray value of each pixel in each type of super pixel block in each single-channel image;
  • the prediction of each pixel Errors are encoded and stored.
  • the method for obtaining the prediction model coefficients of each type of superpixel block includes:
  • the prediction model coefficient of each superpixel block is obtained by using the mean value of the gray change rate between each edge point in each superpixel block and the corresponding edge point in the target direction;
  • the average value of the prediction model coefficients of all super pixel blocks in each type of super pixel block is used as the prediction model coefficient of each type of super pixel block.
  • the method for obtaining the prediction offset of each type of superpixel block includes:
  • each edge point in each superpixel block of each single-channel image and the corresponding edge point in the target direction as a pair of edge point pairs, and each pair of edge point pairs corresponds to an edge point connection line;
  • the predicted value of each pixel point on each edge point line except the edge point is obtained.
  • the mode of the prediction offsets of all pixels in each type of super pixel block is used as the prediction offset of each type of super pixel block.
  • ⁇ k represents the prediction offset of pixel k
  • G b represents the gray value of the target point on the edge point line where pixel k is located
  • ⁇ M represents the Mth class to which the super pixel block where pixel k belongs The prediction model coefficient of the super-pixel block
  • K represents the value corresponding to the Kth pixel after pixel k is the target point on the line connected to the edge point
  • G k represents the gray value of pixel k.
  • methods for obtaining the fitting quality of the fitted straight line include:
  • the method to obtain the prediction error of each pixel in the grayscale image is:
  • the absolute value of the difference between the target predicted gray value and the gray value of each pixel in the gray image is taken as the prediction error of each pixel in the gray image.
  • methods for obtaining superpixel blocks of multiple categories include:
  • the classification principle is: the correlation between any two superpixel blocks in the same category is within the preset correlation threshold interval;
  • An environmental security engineering monitoring data management system including a data acquisition unit, a data analysis unit, and a data compression unit:
  • the data acquisition unit is used to obtain the grayscale image and multiple single-channel images of each frame of the security monitoring video; obtain multiple super-pixel blocks of each single-channel image;
  • the data analysis unit is used to connect each edge point in each superpixel block of each single-channel image with edge points in different directions to obtain edge point connections in different directions, and obtain each edge point connection line in different directions
  • the fitting quality entropy value of each superpixel block in different directions is obtained; the minimum fitting quality entropy value is selected as The fitting entropy value of the superpixel block will obtain the direction of the minimum fitting quality entropy value as the target direction of the corresponding superpixel block;
  • the data encoding unit uses the gray value and the distance between each edge point in each type of super pixel block of each single-channel image and the corresponding edge point in the target direction to obtain the prediction model coefficient of each type of super pixel block;
  • each type of superpixel block in each single-channel image Using the prediction model coefficients of each type of superpixel block in each single-channel image and the gray value of pixels in each type of superpixel block to obtain the prediction bias of each type of superpixel block; according to each type of superpixel block in each single-channel image
  • the prediction offset of the pixel block, the prediction model coefficient and the gray value of the edge point obtain the predicted gray value of each pixel in each type of super pixel block in each single-channel image;
  • the prediction of each pixel Errors are encoded and stored.
  • the beneficial effect of the present invention is: a kind of environmental security engineering monitoring data management method and system of the present invention, because the gray level correlation between the pixel points in each single channel in the image is closer, so for each single channel of the image Channels are analyzed independently, which can make the linear prediction result more accurate; obtain the superpixel blocks in the grayscale image, and make the grayscale image have similar textures and similar grayscales And the pixels with similar distances are divided into the same superpixel block, and then the linear prediction model is analyzed according to the pixel characteristics in the same superpixel block.
  • the prediction model coefficients obtained in this way are more accurate; through Obtain the fitting straight line of the gray level difference on the line between the edge points in each direction in the superpixel block, and then obtain the fitting quality entropy value of the superpixel block, and select the minimum fitting quality entropy value from multiple directions as the superpixel
  • the fitting entropy value of the block that is, the direction of the most regular grayscale change is selected as the direction for linear prediction, which is more in line with the grayscale change law of the pixel point, and can make the subsequent predicted grayscale value closer to the single-channel image
  • the gray value of the pixel itself and then convert the predicted gray value in the single-channel image to the target predicted gray value in the gray image, so that the obtained prediction error is smaller, the encoded space is smaller, and the compression The effect is better; and considering the influence of the prediction bias on the prediction error, compared with directly using the prediction model coefficients to calculate the predicted gray value, the predicted gray value obtained is more accurate, the subsequent prediction error is smaller, and
  • Fig. 1 is a flow chart of the general steps of an embodiment of a method for managing monitoring data of an environmental security project according to the present invention.
  • An embodiment of an environmental security engineering monitoring data management method of the present invention is aimed at images collected by environmental security monitoring, most of which are fire-fighting equipment and fixed buildings, and there are a large number of spatially redundant pixel data in the image, so the establishment of prediction
  • the model compresses the data, as shown in Figure 1, the method includes:
  • each frame of image in the security monitoring video is obtained, wherein each frame of image obtained is an RGB image, and since the pixels in the same channel are more closely related, the pixel points in the same channel in the follow-up
  • the value of the predicted pre- The measurement model will be more accurate, so the R, G, and B channels in each frame of image are processed separately to obtain three single-channel images in each frame of image; and the grayscale image of each frame of image is obtained for subsequent calculation of pixels point forecast error.
  • a single-channel image in one frame image in the security monitoring video is analyzed and processed, and each single-channel image of other frame images is processed in the same manner.
  • the image data compression method selected in this program is predictive coding compression, that is, by establishing a linear prediction model, the predicted gray value of each pixel is obtained, and the prediction error is obtained according to the predicted gray value and the gray value of the pixel itself.
  • the errors are coded and stored.
  • the method of preliminary classification of pixels in a single-channel image is to perform superpixel segmentation on a single-channel image to obtain multiple superpixel blocks, quantify the features of the superpixel blocks, and pass the correlation between the quantized superpixel blocks.
  • Classification of superpixel blocks by degrees there may or may not be a similarity connection between adjacent superpixel blocks, and there may also be a connection between non-adjacent superpixel blocks, and the same prediction model can be used to carry out predict. Therefore, the same predictive model is established for the same type of superpixel blocks with high correlation, and predictive encoding and storage are performed on each frame of image.
  • Superpixel segmentation refers to the division of a series of adjacent pixels with similar color, brightness, and texture features into the same superpixel block area. Superpixel segmentation takes into account the edge information in the image and will be along the Pixels with larger grayscale gradients are segmented, and within the superpixel block, the difference in grayscale values of pixels is very small.
  • Quantitative analysis is performed on the features of each superpixel block in a single-channel image, including the analysis of the brightness value and grayscale variation law of the superpixel block.
  • the gray level change law in the super pixel block it can be analyzed from the gray level change law of the edge point in a certain direction, that is, the gray level difference between adjacent pixel points is linear, and the difference can be stored when storing. The value is further compressed and stored to save storage space.
  • the edge points of each superpixel block are obtained, and the gray scale variation rules of the edge points on the line connecting the 0°, 45°, 90°, and 135° directions are analyzed.
  • the 0° direction is taken as an example
  • 45°, 90°, and 135° directions are treated in the same way.
  • the i'th edge point corresponding to the i-th edge point in the superpixel block in the 0° direction (the corresponding edge point refers to the 0° point passing through the i-th edge point
  • the intersection of the straight line in the direction and the edge line of the superpixel block, that is, the i-th edge point and the corresponding i'th edge point are on the same straight line in the 0° direction
  • the i-th edge point and the corresponding i-th edge point The connection line of the 'th edge point is recorded as the i-th edge point connection line
  • the i'th edge point and the i-th edge point are a pair of edge points.
  • JL a represents the distance from the ath grayscale difference on the edge point connecting line to the fitting straight line corresponding to the edge point connecting line;
  • the parameters of the straight line equation of the fitted line; (x a , y a ) is the abscissa and vertical coordinates of the a-th gray level difference on the i-th edge point connection line, that is, x a represents the i-th edge point connection line y a represents the specific value of the a-th gray-scale difference;
  • the formula for calculating the distance from a point to a straight line is an existing formula, and will not be repeated here.
  • the distance between the gray difference value of adjacent pixel points on the line of each edge point and the corresponding fitting line is obtained.
  • the variance of the distance between all the grayscale differences on the line between each edge point and the fitted straight line is taken as the fitting quality of the fitted straight line.
  • the specific formula for calculating the fitting quality of the fitted straight line corresponding to the i-th edge point connection line is:
  • ZL i represents the fitting quality of the fitted straight line corresponding to the ith edge point connection line
  • u represents the number of pixels on the i-th edge point connection line
  • JL a represents the ath edge point connection line
  • the distance from the grayscale difference to the fitting line corresponding to the line connecting the edge points. Indicates the mean value of the distances from all the grayscale differences on the edge point line to the corresponding fitted straight line, Indicates the variance of the distance between all the grayscale differences on the edge point connection line and the fitting line.
  • the larger the variance the more dispersed the actual grayscale difference is around the fitting line, and the worse the quality of the fitting line. ; the smaller the variance, It is considered that the closer the actual gray level difference is to the fitted straight line, the better the quality of the fitted straight line, and the closer the gray level change relationship on the edge point connection line is to a linear relationship.
  • linear fitting of the gray level difference between adjacent pixels on the edge point line can describe the linear change of the gray level difference.
  • the fitting quality of the fitted straight line is calculated by using the variance of the distance between the gray level difference and the fitted straight line, which can evaluate the linear fitting quality of the gray level difference, and the distance between the gray level difference and the fitted straight line after linear fitting It can reflect the relationship between the grayscale differences, that is, the same linear relationship can be used to integrate the grayscale differences.
  • the fitting quality entropy value of the superpixel block is calculated according to the following formula:
  • S m represents the fitting quality entropy value of the mth superpixel block
  • P ZLi represents the probability that a fitted line with a fitting quality of ZL i appears in all the fitted lines of the mth superpixel block
  • n represents The number of fitted straight lines in the m-th superpixel block, that is, the number of edge point connections
  • i represents the fitted straight line corresponding to the i-th edge point connection.
  • the change law of the gray value inside the super pixel block is more stable.
  • the fitted mass entropy values in the 45°, 90°, and 135° directions were respectively obtained, and the minimum value was selected from the fitted mass entropy values obtained in the four directions as
  • the fitting entropy value of the superpixel block takes the direction of the minimum fitting quality entropy value as the target direction, and it is considered that the gray level change law in the target direction is the most stable and can be used as the optimal prediction direction.
  • the gray mean value of the pixels in each super pixel block in the single-channel image is obtained, and the feature vector of the corresponding super pixel block is formed by using the fitting entropy value and the gray value mean value of each super pixel block;
  • the cosine similarity of the feature vectors of the superpixel blocks is calculated as the correlation between two corresponding superpixel blocks.
  • LX (c,d) represents the correlation between the cth superpixel block and the dth superpixel block;
  • LD c represents the average gray value of pixels in the cth superpixel block;
  • LD d represents the dth superpixel The mean gray value of the pixels in the pixel block;
  • S c represents the fitting entropy value of the cth superpixel block, that is, the degree of confusion of the gray level change law on the line between the edge points in the cth superpixel block;
  • S d represents the The fitting entropy value of the d superpixel block, that is, the degree of confusion of the gray level change law on the line between the edge points in the dth superpixel block;
  • the cth superpixel block is more similar to the dth superpixel block, it means The change law of brightness and gray scale is numerically closer to equal, that is, the feature vector of the cth superpixel block and the feature vector of the dth superpixel block The higher the cosine similarity of (the calculation formula
  • the correlation degree of every two super pixel blocks is obtained, and the super pixel blocks are classified according to the correlation degree of every two super pixel blocks.
  • the reason for choosing classification instead of merging for all superpixel blocks in a single-channel image is that the grayscale change law and brightness value characteristics of pixels in each superpixel block are similar, and the distribution characteristics of pixel points are similar, but If the superpixel blocks are merged, the distribution law of the original pixels in the merged superpixel block will change, which will make the error of the originally constructed prediction model that satisfies the distribution law of the pixel points in the superpixel block before merging be amplified. As a result, the encoding is too long and the compression effect is poor.
  • the correlation threshold interval can be set according to the actual situation.
  • the number of categories of superpixel blocks obtained according to the correlation threshold interval should be reasonable. Not too much or too little, which can not only reduce the amount of calculation and model storage space, but also predict the pixel points of each super pixel block according to the feature change law within the super pixel block, and the prediction result is more accurate.
  • the principle of classification is: the correlation between any two superpixel blocks in the same category is within the preset correlation threshold interval.
  • the degree of correlation between different superpixel blocks of a video image Use the same prediction model for the same type of superpixel blocks to predict, and each superpixel block has a prediction direction that belongs to the current superpixel block, that is, the target direction. There are one or more superpixel blocks in each type of superpixel block.
  • Classify super-pixel blocks with high correlation into the same class that is, according to the gray value and gray-scale change law of the super-pixel block, divide the super-pixel blocks with similar gray-scale mean value and similar gray-scale change law into the same class.
  • the classification method reduces the amount of calculation while ensuring the accuracy of the prediction model coefficients.
  • V m,i represents the gray scale change rate between the i-th edge point in the m-th superpixel block and the corresponding i'-th edge point in the target direction
  • I m,i represents the m-th super-pixel block
  • I m, i' represents the gray value of the i'-th edge point corresponding to the i-th edge point in the target direction in the m-th superpixel block;
  • L m, i Indicates the distance between the i-th edge point and the corresponding i'-th edge point in the target direction.
  • the gray scale change rate on the edge point connection is obtained by the ratio of the absolute value of the overall gray level change difference on the edge point connection line to the edge point connection line length.
  • the edge points and corresponding edge points in the superpixel block indicated in steps S5-S6, and the edge point connection all refer to the edge point connection on the target direction of the superpixel block, and the edge point and the target corresponding edge points in the direction.
  • the same linear prediction model coefficients are used for all pixels in the superpixel block in the same category.
  • the prediction model coefficients of each superpixel block are obtained by using the mean value of the gray change rate between each edge point in each superpixel block and the corresponding edge point in the target direction.
  • the prediction model coefficients of each superpixel block are calculated according to the following formula:
  • ⁇ m represents the prediction model coefficient of the mth superpixel block, that is, the prediction model coefficient of the linear prediction model between the edge point in the mth superpixel block and the corresponding edge point
  • V m,i represents the mth superpixel block
  • n represents the number of edge points in the m-th superpixel block.
  • the prediction model coefficients of each type of super pixel block are obtained by using the mean value of the prediction model coefficients of all super pixel blocks in each type of super pixel block.
  • each edge point in each superpixel block and the corresponding edge point in the target direction as a pair of edge points, and each pair of edge points corresponds to an edge point connection line; compare the gray value of each edge point pair
  • the small edge points are recorded as target points; according to the gray value of the target point on each edge point line in each super pixel block and the prediction model coefficient of the category of the super pixel block, each edge point line except edge
  • the predicted value of each pixel other than the pixel point; the difference between the predicted value and the gray value of each pixel in each super pixel block is used to obtain the predicted offset of each pixel; the super pixel block of each type
  • the mode of the prediction bias of all pixels is used as the prediction bias of each type of super pixel block.
  • the prediction bias of each pixel in the superpixel block is calculated according to the following formula:
  • ⁇ k represents the prediction offset of pixel k
  • G b represents the gray value of the target point on the edge point line where pixel k is located
  • ⁇ M represents the Mth class to which the super pixel block where pixel k belongs
  • K means that the pixel k is the value corresponding to the Kth pixel after the target point on the edge point connection line, for example, the fifth pixel after the target point, K is equal to 5
  • G k means Gray value of pixel k.
  • the gray value of each pixel on the edge point line is obtained according to the gray level change law, and the edge point line is used to The gray value of the target point plus the product of the linear model coefficient and the distance from pixel point k to the target point G b + ⁇ M ⁇ K, to obtain the predicted value of pixel k, because the target point is the gray value of the edge point.
  • the predicted gray value of each pixel is calculated according to the following formula:
  • I k G b + ⁇ M ⁇ K+ ⁇ k,f
  • I k represents the predicted gray value of pixel k
  • G b represents the gray value of the target point on the edge point line where pixel k is located
  • the prediction model coefficient of the class superpixel block K means that the pixel k is the Kth pixel point after the target point on the edge point connection line, for example, the fifth pixel point after the target point, K is equal to 5, that is, the pixel point and The distance between the target points is 5;
  • ⁇ k ,f represents the prediction bias of the fth class superpixel block to which the pixel k belongs.
  • the prediction error of each pixel in the grayscale image of each frame image in the surveillance video is encoded and stored, and the encoding method of the predictive encoding can be selected as run-length encoding.
  • An environmental security engineering monitoring data management system includes a data acquisition unit, a data analysis unit, and a data compression unit. Specifically: a data acquisition unit, configured to acquire grayscale images and multiple single-channel images of each frame of the security monitoring video; and acquire multiple superpixel blocks of each single-channel image.
  • the data analysis unit is used to connect each edge point in each superpixel block of each single-channel image with edge points in different directions to obtain edge point connections in different directions, and obtain each edge point connection line in different directions.
  • the fitting straight line of the gray level difference between adjacent pixels and the fitting quality of the fitting straight line is used to connect each edge point in each superpixel block of each single-channel image with edge points in different directions to obtain edge point connections in different directions, and obtain each edge point connection line in different directions.
  • the fitting quality entropy value of each superpixel block in different directions is obtained; the minimum fitting quality entropy value is selected as The fitting entropy value of the superpixel block will obtain the direction of the minimum fitting quality entropy value as the target direction of the corresponding superpixel block;
  • the data encoding unit uses the gray value and distance between each edge point in each type of super pixel block of each single-channel image and the corresponding edge point in the target direction to obtain the prediction model coefficient of each type of super pixel block.
  • the prediction offset of the pixel block, the prediction model coefficient and the gray value of the edge point are used to obtain the predicted gray value of each pixel in each type of super pixel block in each single-channel image.
  • the prediction of each pixel Errors are encoded and stored.
  • the present invention provides a monitoring data management method and system for environmental security engineering. Since the grayscale connection between pixels in each single channel in the image is closer, each single channel of the image is Independent analysis can make the linear prediction result more accurate; obtain the superpixel blocks in the grayscale image, divide the pixels with similar texture, similar grayscale and similar distance in the grayscale image into the same superpixel block, and follow up according to the same
  • the pixel feature analysis linear prediction model in the super pixel block compared with the random division of the macro block, the prediction model coefficient obtained by this method is more accurate; by obtaining the gray level difference between the edge points in each direction in the super pixel block.
  • the fitting quality entropy value of the superpixel block is obtained, and the minimum fitting quality entropy value is selected from multiple directions as the fitting entropy value of the superpixel block, that is, the direction with the most regular grayscale change is selected as the
  • the direction of linear prediction is determined, which is more in line with the change law of the gray level of the pixel, so that the predicted gray value obtained later

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)
  • Compression Of Band Width Or Redundancy In Fax (AREA)

Abstract

本发明公开了一种环境安防工程监测数据管理方法及系统，涉及数据处理技术领域，该方法包括：获取监控视频数据中每帧图像中每个单通道图像的多个超像素块；获取超像素块中每个边缘点连线的灰度差值的拟合直线，以及拟合直线的拟合质量；利用超像素块在不同方向的拟合直线的拟合质量得到拟合熵值；利用每两个超像素块的相关度将超像素块分为多个类别；获取每类超像素块的预测模型系数；根据灰度图像中每个像素点的目标预测灰度值和灰度值得到预测误差；对每个像素点的预测误差进行编码存储。本发明构建的预测模型更加准确，得到的预测误差数值更小，编码后的占用空间更小。

Description

一种环境安防工程监测数据管理方法及系统 技术领域
本发明涉及数据处理技术领域，具体涉及一种环境安防工程监测数据管理方法及系统。
背景技术
随着信息时代的发展，数字监控在环境安防工程中的普及越来越广。同时随着监控视频质量要求的提升，摄像头的分辨率不断提升，使得监控采集的视频数据的数据量大幅增加，进而需要大量的存储空间对视频数据进行存储。为了减少监控视频数据的存储占用空间，需要对视频数据进行压缩存储。
现有技术一般采取有损压缩对视频数据进行压缩存储，能够大量减小存储时的占用空间，常用的有损压缩方式为预测编码压缩，预测编码压缩是将每帧图像划分为多个随机大小的矩形宏块，随后利用每个宏块中像素点的特征值进行线性预测，若同一矩形宏块中灰度分布平缓，像素点之间的灰度相关性较强，则线性预测结果较为理想，若同一矩形宏块中灰度分布随机性较大，突变点较多，则线性预测结果准确性不高，常用的随机划分矩形宏块进行预测编码压缩的方式并没有考虑到每个矩形宏块中像素点的灰度分布特点，因此会导致部分灰度分布随机度较大的宏块的预测结果不够准确，从而导致预测误差数值较大，对预测误差进行编码后数据占用空间较大，压缩效率低。
发明内容
本发明提供一种环境安防工程监测数据管理方法及系统，以解决现有的随机划分宏块中存在部分灰度分布随机度较大的宏块线性预测结果不准确，进而导致压缩效率低的问题。
本发明的一种环境安防工程监测数据管理方法及系统，采用如下技术方案：
获取安防监控视频的每帧图像的灰度图像和多个单通道图像；获取每个单通道图像的多个超像素块；
利用每个单通道图像的每个超像素块中每个边缘点与不同方向的边缘点相连得到不同方向的边缘点连线，获取不同方向的每个边缘点连线上相邻像素点的灰度差值的拟合直线以及拟合直线的拟合质量；
利用每个超像素块中不同方向的每个边缘点连线对应的拟合直线的拟合质量，得到每个超像素块在不同方向的拟合质量熵值；选取最小拟合质量熵值作为超像素块的拟合熵值，将得到最小拟合质量熵值的方向作为对应的超像素块的目标方向；
利用每个单通道图像中每两个超像素块的拟合熵值和灰度均值对所有超像素块进行分类得到多个类别的超像素块；
利用每个单通道图像的每类超像素块中每个边缘点和目标方向上对应的边缘点的灰度值和距离，得到每类超像素块的预测模型系数；
利用每个单通道图像中每类超像素块的预测模型系数和每类超像素块中像素点的灰度值得到每类超像素块的预测偏量；根据每个单通道图像中每类超像素块的预测偏量、预测模型系数和边缘点的灰度值得到每个单通道图像中每类超像素块中每个像素点的预测灰度值；
利用灰度图像中每个像素点的灰度值和每个像素点在每个单通道图像中的预测灰度值得到灰度图像中每个像素点的预测误差，对每个像素点的预测误差进行编码并存储。
进一步，得到每类超像素块的预测模型系数的方法包括：
获取每个单通道图像的每类超像素块中每个超像素块的每个边缘点和目标方向上对应的边缘点的灰度差值以及距离；
对每个边缘点和目标方向上对应的边缘点的得到的灰度差值和距离求比值，得到每个边缘点和目标方向上对应的边缘点之间的灰度变化速率；
利用每个超像素块中每个边缘点和目标方向上对应的边缘点之间的灰度变化速率的均值得到每个超像素块的预测模型系数；
将每类超像素块中所有超像素块的预测模型系数的均值作为每类超像素块的预测模型系数。
进一步，得到每类超像素块的预测偏量的方法包括：
将每个单通道图像的每个超像素块中每个边缘点和目标方向上对应的边缘点记为一对边缘点对，每对边缘点对对应一条边缘点连线；
将每个边缘点对中灰度值较小的边缘点记为目标点；
根据每个超像素块中每条边缘点连线上的目标点的灰度值和超像素块所在类别的预测模型系数得到每条边缘点连线上除边缘点以外每个像素点的预测值；
利用每个超像素块中每个像素点的预测值和灰度值之间的差值得到每个像素点的预测偏量；
将每类超像素块中所有像素点的预测偏量的众数作为每类超像素块的预测偏量。
进一步，获取每个超像素块中每个像素点的预测偏量的公式为：
βk＝Gb+αM×K-Gk
其中，βk表示像素点k的预测偏量；Gb表示像素点k所在的边缘点连线上的目标点的灰度值；αM表示像素点k所在的超像素块所属的第M类超像素块的预测模型系数；K表示像素点k为边缘点连线上目标点以后的第K个像素点对应的数值；Gk表示像素点k的灰度值。
进一步，得到每个像素点的预测灰度值的方法为：
获取每个像素点与所在边缘点连线上目标点之间的距离；
利用得到的距离与像素点所在超像素块所属的超像素块类别的预测模型系数相乘，得到每个像素点对应的乘积；
将每个像素点得到的乘积与所在边缘点连线上目标点的灰度值相加，再加上所属超像素块类别的预测偏量，得到每个像素点的预测灰度值。
进一步，获取拟合直线的拟合质量的方法包括：
获取边缘点连线上每对相邻像素点的灰度差值到拟合直线的距离的方差，将每个边缘点连线得到的方差作为边缘点连线对应的拟合直线的拟合质量。
进一步，得到灰度图像中每个像素点的预测误差的方法为：
利用每个像素点在多个单通道图像中的预测灰度值得到每个像素点在灰度图像中的目标预测灰度值；
将灰度图像中每个像素点的目标预测灰度值和灰度值的差值的绝对值，作为灰度图像中每个像素点的预测误差。
进一步，得到多个类别的超像素块的方法包括：
利用每个超像素块的拟合熵值和灰度均值组成对应的超像素块的特征向量；
对每两个超像素块的特征向量计算余弦相似度作为对应两个超像素块的相关度；
分类原则为：同一类别中任意两个超像素块的相关度都在预设的相关度阈值区间中；
利用所述分类原则对每个单通道图像中的超像素块进行分类得到多个类别的超像素块。
一种环境安防工程监测数据管理系统，包括，数据采集单元，数据分析单元，数据压缩单元：
数据采集单元，用于获取安防监控视频的每帧图像的灰度图像和多个单通道图像；获取每个单通道图像的多个超像素块；
数据分析单元，用于利用每个单通道图像的每个超像素块中每个边缘点与不同方向的边缘点相连得到不同方向的边缘点连线，获取不同方向的每个边缘点连线上相邻像素点的灰度差值的拟合直线以及拟合直线的拟合质量；
利用每个超像素块中不同方向的每个边缘点连线对应的拟合直线的拟合质量，得到每个超像素块在不同方向的拟合质量熵值；选取最小拟合质量熵值作为超像素块的拟合熵值，将得到最小拟合质量熵值的方向作为对应的超像素块的目标方向；
利用每个单通道图像中每两个超像素块的拟合熵值和灰度均值对所有超像素块进行分类得到多个类别的超像素块；
数据编码单元，利用每个单通道图像的每类超像素块中每个边缘点和目标方向上对应的边缘点的灰度值和距离，得到每类超像素块的预测模型系数；
利用每个单通道图像中每类超像素块的预测模型系数和每类超像素块中像素点的灰度值得到每类超像素块的预测偏量；根据每个单通道图像中每类超像素块的预测偏量、预测模型系数和边缘点的灰度值得到每个单通道图像中每类超像素块中每个像素点的预测灰度值；
利用灰度图像中每个像素点的灰度值和每个像素点在每个单通道图像中的预测灰度值得到灰度图像中每个像素点的预测误差，对每个像素点的预测误差进行编码并存储。
本发明的有益效果是：本发明的一种环境安防工程监测数据管理方法及系统，由于图像中每个单通道中像素点之间的灰度联系性更为密切，所以对图像的每个单通道进行独立分析，能够使得线性预测结果更准确；获取灰度图像中的超像素块，将灰度图像中纹理相近，灰度相似 且距离相近的像素点划分到了同一个超像素块，后续根据同一个超像素块中的像素点特征分析线性预测模型，相比于随机划分宏块，该方式得到的预测模型系数更加准确；通过获取超像素块中每个方向的边缘点连线上灰度差值的拟合直线，进而得到超像素块的拟合质量熵值，从多个方向中选取最小拟合质量熵值作为超像素块的拟合熵值，即选取了灰度变化最规律的方向作为了进行线性预测的方向，更贴合像素点的灰度变化规律，能够使得后续得到的预测灰度值更加接近单通道图像中像素点本身的灰度值，再将单通道图像中的预测灰度值转换到灰度图像中的目标预测灰度值，使得到的预测误差更小，编码后的占用空间更小，压缩效果更好；并且考虑到了预测偏量对预测误差的影响，相对于直接利用预测模型系数计算预测灰度值，得到的预测灰度值更加准确，后续得到的预测误差就越小，压缩效果越好。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本发明的一种环境安防工程监测数据管理方法的实施例总体步骤的流程图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本发明的一种环境安防工程监测数据管理方法的实施例，针对于环境安防监控采集的图像，其中多为消防器械和固定建筑物，图像中存在大量空间冗余的像素点数据，所以建立预测模型对数据进行压缩，如图1所示，该方法包括：
S1、获取安防监控视频的每帧图像的灰度图像和多个单通道图像；获取每个单通道图像的多个超像素块。
具体的，获取安防监控视频中的每帧图像，其中，得到的每帧图像为RGB图像，由于相同通道中的像素点之间的联系性更为密切，所以在后续对相同通道内的像素点的值进行预测的预 测模型会更加准确，所以对每帧图像中的R、G、B通道进行单独处理，获取每帧图像中的三个单通道图像；并获取每帧图像的灰度图像，用于后续计算像素点的预测误差。
本实施例中以安防监控视频中的其中一帧图像中的一个单通道图像进行分析处理，其他帧图像的每个单通道的处理方式相同。
本方案选择的图像数据压缩方式为预测编码压缩，即通过建立线性预测模型，得到每个像素点的预测灰度值，根据预测灰度值和像素点本身的灰度值得到预测误差，对预测误差进行编码存储。为了使得线性预测结果更加准确，需要将单通道图像中特征相近的像素点分为一类，将特征差异较大的像素点划分到不同的部分构建线性预测模型。所以对单通道图像中像素点进行初步分类的方式为，对单通道图像进行超像素分割得到多个超像素块，对于超像素块的特征进行量化，通过量化后的超像素块之间的相关度对超像素块划分类别，相邻超像素块之间可能存在相似性的联系，也有可能不存在，不相邻的超像素块之间也可能存在联系性，可以使用同一种预测模型去进行预测。所以对相关度大的同一类超像素块建立同一预测模型，对每帧图像进行预测编码存储。
具体的，对单通道图像进行超像素分割，得到多个超像素块。超像素分割是指将一系列位置相邻且颜色、亮度、纹理特征相似的像素点划分到同一个超像素块区域，超像素分割考虑到了图像中的边缘信息，会在分割时沿着图像中灰度梯度较大的像素点进行分割，在超像素块内部，像素点的灰度值差异是很小的。
S2、利用每个单通道图像的每个超像素块中每个边缘点与不同方向的边缘点相连得到不同方向的边缘点连线，获取不同方向的每个边缘点连线上相邻像素点的灰度差值的拟合直线以及拟合直线的拟合质量。
对单通道图像中每个超像素块的特征进行量化分析，包括分析超像素块的亮度值和灰度变化规律。分析超像素块中的灰度变化规律时，可以从边缘点在某一方向上的灰度变化规律进行分析，即相邻像素点之间的灰度差值呈线性规律，存储时，可将差异值进一步进行压缩存储，更加节约存储空间。
具体的，获取每个超像素块的边缘点，分析边缘点在0°，45°，90°，135°方向连线上像素点的灰度变化规律，本实施例中以0°方向为例，45°，90°，135°方向上的处理方式相同。
以第m个超像素块为分析对象：获取超像素块中第i个边缘点在0°方向上对应的第i'个边缘点(对应的边缘点是指经过第i个边缘点的0°方向的直线与超像素块的边缘线的交点，即第i个边缘点和对应的第i'个边缘点在同一条0°方向的直线上)，将第i个边缘点和对应的第i'个边缘点的连线记为第i条边缘点连线，且第i'个边缘点和第i个边缘点为一对边缘点对。
获取第i条边缘点连线上第i个边缘点和对应的第i'个边缘点之间的所有相邻像素点之间的灰度变化规律。具体的，计算第i条边缘点连线上相邻像素点之间的灰度差值，将第i条边缘点连线上所有的灰度差值进行线性拟合得到第i条边缘点连线的灰度差值的拟合直线(线性拟合为现有技术，在此不作赘述)，拟合直线的直线方程为kix-y+bi＝0。
根据下式计算边缘点连线上的每个相邻像素点的灰度差值到边缘点连线对应的拟合直线的距离：
其中，JLa表示边缘点连线上的第a个灰度差值到边缘点连线对应的拟合直线的距离；ki、bi为第i条边缘点连线的灰度差值的拟合直线的直线方程的参数；(xa，ya)为第i条边缘点连线上的第a个灰度差值的横纵坐标，即xa表示第i条边缘点连线上的第a个灰度差值，ya表示第a个灰度差值的具体数值；点到直线的距离计算公式为现有公式，在此不做赘述。
根据此方法获取每个边缘点连线上相邻像素点的灰度差值到对应的拟合直线的距离。
将每个边缘点连线上所有的灰度差值到拟合直线的距离的方差，作为拟合直线的拟合质量。具体计算第i条边缘点连线对应的拟合直线的拟合质量的公式为：
其中，ZLi表示表示第i条边缘点连线对应的拟合直线的拟合质量；u表示第i条边缘点连线上像素点的数量；JLa表示边缘点连线上的第a个灰度差值到边缘点连线对应的拟合直线的距离。表示边缘点连线上所有灰度差值到对应的拟合直线的距离的均值， 表示边缘点连线上所有的灰度差值到拟合直线的距离的方差，方差越大，认为实际的灰度差值在拟合直线周围的分布越分散，拟合直线的质量越不好；方差越小， 认为实际的灰度差值与拟合直线越贴合，拟合直线的质量越好，边缘点连接线上的灰度变化关系越趋近于线性关系。
需要说明的是，对边缘点连线上相邻像素点的灰度差值进行线性拟合可以对灰度差值的线性变化进行描述，当边缘点连线上的灰度差值越趋近于线性变化时，在存储过程中，可以对相似的灰度差值再进行压缩，能更好的节省压缩空间。利用灰度差值到拟合直线的距离的方差计算拟合直线的拟合质量，可以对灰度差值的线性拟合质量进行评价，线性拟合后灰度差值到拟合直线的距离可以反映灰度差值之间的联系性，即能使用同一个线性关系对灰度差值进行整合。
S3、利用每个超像素块中不同方向的每个边缘点连线对应的拟合直线的拟合质量，得到每个超像素块在不同方向的拟合质量熵值；选取最小拟合质量熵值作为超像素块的拟合熵值，将得到最小拟合质量熵值的方向作为对应的超像素块的目标方向。
具体的，以第m个超像素块为例，根据下式计算超像素块的拟合质量熵值：
其中，Sm表示第m个超像素块的拟合质量熵值；PZLi表示拟合质量为ZLi的拟合直线在第m个超像素块的所有拟合直线中出现的概率；n表示第m个超像素块中拟合直线的数量，即边缘点连线的数量；i表示第i条边缘点连线对应的拟合直线。为信息熵的计算公式，为现有技术，在此处用于表示超像素块中所有拟合直线的拟合质量的混乱程度，拟合质量熵值越大，表示第m个超像素块中拟合直线的拟合质量变化越混乱，即超像素块中灰度值变化越混乱，熵值Sm越小，表示超像素块中所有拟合直线的拟合质量越接近，即第m个超像素块内部灰度值的变化规律越稳定。
至此，得到了第m个超像素块在0°方向上的拟合质量熵值。
根据得到0°方向上的拟合质量熵值的方法分别获取在45°,90°,135°方向上的拟合质量熵值，从四个方向得到的拟合质量熵值中选取最小值作为该超像素块的拟合熵值，将得到最小拟合质量熵值的方向作为目标方向，认为在目标方向上的灰度变化规律是最稳定的，可以作为最优的预测方向。
S4、利用每个单通道图像中每两个超像素块的拟合熵值和灰度均值对所有超像素块进行分类得到多个类别的超像素块。
若单通道图像中任意两个超像素块的亮度和灰度变化规律的混乱程度越接近，认为这两个超像素块的相关度越高，超像素块的亮度可以用像素点的灰度值表示，灰度变化规律的混乱程度则可以用拟合熵值表示。
具体的，获取单通道图像中每个超像素块中像素点的灰度均值，利用每个超像素块的拟合熵值和灰度均值组成对应的超像素块的特征向量；对每两个超像素块的特征向量计算余弦相似度作为对应两个超像素块的相关度。
以第c个超像素块和第d个超像素块为例，计算两个超像素块的相关度的公式为：
其中，LX(c,d)表示第c个超像素块和第d个超像素块的相关度；LDc表示第c个超像素块中像素点的灰度均值；LDd表示第d个超像素块中像素点的灰度均值；Sc表示第c个超像素块的拟合熵值，即第c个超像素块中边缘点连线上灰度变化规律的混乱程度；Sd表示第d个超像素块的拟合熵值，即第d个超像素块中边缘点连线上灰度变化规律的混乱程度；若第c个超像素块和第d个超像素块越相似，表明亮度和灰度变化规律在数值上是越趋近于相等的，即第c个超像素块的特征向量和第d个超像素块的特征向量的余弦相似度越高(余弦相似度计算公式为现有公式，在此不作赘述)，即LX(c,d)的值越趋近于1。
至此，得到每两个超像素块的相关度，根据每两个超像素块的相关度对超像素块进行分类。
对单通道图像中的所有超像素块选择分类而不是合并的原因是，每个超像素块内像素点的灰度变化规律、亮度值特征是存在相似性的，像素点的分布特征相似，但是若进行超像素块合并的话，合并后的超像素块中原本像素点的分布规律发生变化，会使得原本构建的满足未合并之前的超像素块内像素点的分布规律的预测模型的误差放大，从而导致编码过长，压缩效果差。
根据超像素块之间的相关度对单通道图像中所有的超像素块进行分类，相关度阈值区间可根据实际情况自行设置，应使得根据相关度阈值区间得到的超像素块的类别数量合理，不过多或过少，这样既可以减少计算量和模型存储空间，又可以根据超像素块之内的特征变化规律对每个超像素块的像素点进行预测，预测结果更准确。
分类的原则为：同一类别中任意两个超像素块的相关度都在预设的相关度阈值区间中。根据该分类原则对每个单通道图像中的超像素块进行分类得到每个单通道图像中的多类超像素 块，设置超像素块间相关度阈值区间为[0.7,1]，在此区间内的超像素块认为联系较为紧密，此时无需关注超像素块的大小、形状、位置，只需关注安防监控视频图像不同超像素块之间的相关度。对同一类超像素块使用同一个预测模型进行预测，在每个超像素块内有属于当前超像素块内的预测方向，即目标方向。每类超像素块中有一个或多个超像素块。
将相关度大的超像素块分为同一类，即根据超像素块的灰度值和灰度变化规律，将灰度均值相近且灰度变化规律相近的超像素块分为了同一类，相比于对每个超像素块构建线性预测模型，分类的方式在保证预测模型系数准确的同时减少了计算量。
S5、利用每个单通道图像的每类超像素块中每个边缘点和目标方向上对应的边缘点的灰度值和距离，得到每类超像素块的预测模型系数。
具体的，获取单通道图像中每类超像素块中每个超像素块的每个边缘点和目标方向上对应的边缘点的灰度差值以及距离；对每个边缘点和目标方向上对应的边缘点的得到的灰度差值和距离求比值，得到每个边缘点和目标方向上对应的边缘点之间的灰度变化速率；根据下式计算每个边缘点和目标方向上对应边缘点之间的灰度变化速率：
其中，Vm,i表示第m个超像素块中第i个边缘点和目标方向上对应的第i'个边缘点之间的灰度变化速率；Im,i表示第m个超像素块中第i个边缘点的灰度值；Im,i'表示第m个超像素块中第i个边缘点在目标方向上对应的第i'个边缘点的灰度值；Lm,i表示第i个边缘点和目标方向上对应的第i'个边缘点之间的距离。即通过边缘点连线上总体的灰度变化差值的绝对值与边缘点连线长度的比值得到了边缘点连线上的灰度变化速率。(步骤S5-S6中所指的超像素块中边缘点和对应的边缘点，以及边缘点连线，全部都指的是超像素块的目标方向上的边缘点连线，以及边缘点和目标方向上对应的边缘点。)
由于超像素块内的像素点为特征值较为相似的像素点，且同一类别中超像素块的相关度较高，所以对同一类别的超像素块内的所有像素点使用相同的线性预测模型系数。
利用每个超像素块中每个边缘点和目标方向上对应的边缘点之间的灰度变化速率的均值得到每个超像素块的预测模型系数。
具体的，以第m个超像素块为例，根据下式计算每个超像素块的预测模型系数：
其中，αm表示第m个超像素块的预测模型系数，即第m个超像素块内的边缘点和对应边缘点之间的线性预测模型的预测模型系数；Vm,i表示第m个超像素块中第i个边缘点和目标方向上对应的第i'个边缘点之间的灰度变化速率，n表示第m个超像素块中边缘点连线的数量。
利用每类超像素块中所有超像素块的预测模型系数的均值得到每类超像素块的预测模型系数。
S6、利用每个单通道图像中每类超像素块的预测模型系数和每类超像素块中像素点的灰度值得到每类超像素块的预测偏量；根据每个单通道图像中每类超像素块的预测偏量、预测模型系数和边缘点的灰度值得到每个单通道图像中每类超像素块中每个像素点的预测灰度值。
将每个超像素块中每个边缘点和目标方向上对应的边缘点记为一对边缘点对，每对边缘点对对应一条边缘点连线；将每个边缘点对中灰度值较小的边缘点记为目标点；根据每个超像素块中每条边缘点连线上的目标点的灰度值和超像素块所在类别的预测模型系数得到每条边缘点连线上除边缘点以外每个像素点的预测值；利用每个超像素块中每个像素点的预测值和灰度值之间的差值得到每个像素点的预测偏量；将每类超像素块中所有像素点的预测偏量的众数作为每类超像素块的预测偏量。
具体的，对于超像素块中每个像素点进行预测时会得到不同的预测值，即也会存在不同的预测偏量，根据下式计算超像素块中每个像素点的预测偏量：
βk＝Gb+αM×K-Gk
其中，βk表示像素点k的预测偏量；Gb表示像素点k所在的边缘点连线上的目标点的灰度值；αM表示像素点k所在的超像素块所属的第M类超像素块的预测模型系数；K表示像素点k为边缘点连线上目标点以后的第K个像素点对应的数值，例如目标点以后第5个像素点，K就等于5；Gk表示像素点k的灰度值。由于αM是通过边缘点连线上边缘点对之间的灰度变化规律得到的，所以根据灰度变化规律获取边缘点连线上每个像素点的灰度值，利用边缘点连线上的目标点的灰度值加上线性模型系数与像素点k到目标点的距离的乘积Gb+αM×K，得到像素点k的预测值，由于目标点是边缘点对中灰度值较小的一个，所以用目标点加上灰度变化规律， 如果目标点选择的是灰度值较大的一个边缘点，则用目标点减去灰度变化规律；根据预测值和像素点本身的灰度值的差值得到像素点k的预测偏量。
获取每类超像素块中所有像素点的预测偏量的众数作为每类超像素块的预测偏量，直接存储每一类的预测偏量，相对于存储每一个预测偏量减少了存储占用空间。
获取每个像素点与所在边缘点连线上目标点之间的距离；利用得到的距离与像素点所在超像素块所属的超像素块类别的预测模型系数相乘，得到每个像素点对应的乘积；将每个像素点得到的乘积与所在边缘点连线上目标点的灰度值相加，再加上所属超像素块类别的预测偏量，得到每个像素点的预测灰度值。具体的，根据下式计算每个像素点的预测灰度值：
Ik＝Gb+αM×K+βk,f
其中，Ik表示像素点k的预测灰度值；Gb表示像素点k所在的边缘点连线上的目标点的灰度值；αM表示像素点k所在的超像素块所属的第M类超像素块的预测模型系数；K表示像素点k为边缘点连线上目标点以后的第K个像素点，例如目标点以后第5个像素点，K就等于5，即该像素点与目标点之间的距离为5；βk,f表示像素点k所属的第f类超像素块的预测偏量。利用边缘点连线上的目标点的灰度值加上线性模型系数与像素点k到目标点的距离的乘积Gb+αM×K，得到像素点k的预测值，由于目标点是边缘点对中灰度值较小的一个，所以用目标点加上灰度变化规律，如果目标点选择的是边缘点对中灰度值较大的一个边缘点，则用目标点减去灰度变化规律；考虑到了预测偏量对预测值的影响，加上所属的超像素块类别的预测偏量，相对于直接利用预测模型系数计算预测灰度值，得到的预测灰度值更加准确，后续得到的预测误差就越小，压缩效果越好。
S7、利用灰度图像中每个像素点的灰度值和每个像素点在每个单通道图像中的预测灰度值得到灰度图像中每个像素点的预测误差，对每个像素点的预测误差进行编码并存储。
具体的，利用每个像素点在多个单通道图像中的预测灰度值得到每个像素点在灰度图像中的目标预测灰度值(将像素点在多个单通道图像的灰度值转化为像素点在灰度图像中的灰度值为现有技术，在此不作赘述)；利用灰度图像中每个像素点的目标预测灰度值减去像素点在灰度图像中的灰度值得到差值，对差值求绝对值得到作为灰度图像中每个像素点的预测误差。
对监控视频中每帧图像的灰度图像中每个像素点的预测误差进行编码并存储，预测编码的编码方式可选择游程编码。
一种环境安防工程监测数据管理系统，包括，数据采集单元，数据分析单元，数据压缩单元。具体的：数据采集单元，用于获取安防监控视频的每帧图像的灰度图像和多个单通道图像；获取每个单通道图像的多个超像素块。
数据分析单元，用于利用每个单通道图像的每个超像素块中每个边缘点与不同方向的边缘点相连得到不同方向的边缘点连线，获取不同方向的每个边缘点连线上相邻像素点的灰度差值的拟合直线以及拟合直线的拟合质量。
利用每个超像素块中不同方向的每个边缘点连线对应的拟合直线的拟合质量，得到每个超像素块在不同方向的拟合质量熵值；选取最小拟合质量熵值作为超像素块的拟合熵值，将得到最小拟合质量熵值的方向作为对应的超像素块的目标方向；
利用每个单通道图像中每两个超像素块的拟合熵值和灰度均值对所有超像素块进行分类得到多个类别的超像素块。
数据编码单元，利用每个单通道图像的每类超像素块中每个边缘点和目标方向上对应的边缘点的灰度值和距离，得到每类超像素块的预测模型系数。
利用每个单通道图像中每类超像素块的预测模型系数和每类超像素块中像素点的灰度值得到每类超像素块的预测偏量；根据每个单通道图像中每类超像素块的预测偏量、预测模型系数和边缘点的灰度值得到每个单通道图像中每类超像素块中每个像素点的预测灰度值。
利用灰度图像中每个像素点的灰度值和每个像素点在每个单通道图像中的预测灰度值得到灰度图像中每个像素点的预测误差，对每个像素点的预测误差进行编码并存储。
综上所述，本发明提供一种环境安防工程监测数据管理方法及系统，由于图像中每个单通道中像素点之间的灰度联系性更为密切，所以对图像的每个单通道进行独立分析，能够使得线性预测结果更准确；获取灰度图像中的超像素块，将灰度图像中纹理相近，灰度相似且距离相近的像素点划分到了同一个超像素块，后续根据同一个超像素块中的像素点特征分析线性预测模型，相比于随机划分宏块，该方式得到的预测模型系数更加准确；通过获取超像素块中每个方向的边缘点连线上灰度差值的拟合直线，进而得到超像素块的拟合质量熵值，从多个方向中选取最小拟合质量熵值作为超像素块的拟合熵值，即选取了灰度变化最规律的方向作为了进行线性预测的方向，更贴合像素点的灰度变化规律，能够使得后续得到的预测灰度值更加接近单通道图像中像素点本身的灰度值，再将单通道图像中的预测灰度值转换到灰度图像中的目标预测灰度值，使得到的预测误差更小，编码后的占用空间更小，压缩效果更好；并且考虑到了预 测偏量对预测误差的影响，相对于直接利用预测模型系数计算预测灰度值，得到的预测灰度值更加准确，后续得到的预测误差就越小，压缩效果越好。
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (8)

  1. 一种环境安防工程监测数据管理方法，其特征在于：
    获取安防监控视频的每帧图像的灰度图像和多个单通道图像；获取每个单通道图像的多个超像素块；
    利用每个单通道图像的每个超像素块中每个边缘点与不同方向的边缘点相连得到不同方向的边缘点连线，获取不同方向的每个边缘点连线上相邻像素点的灰度差值的拟合直线以及拟合直线的拟合质量；
    利用每个超像素块中不同方向的每个边缘点连线对应的拟合直线的拟合质量，得到每个超像素块在不同方向的拟合质量熵值；选取最小拟合质量熵值作为超像素块的拟合熵值，将得到最小拟合质量熵值的方向作为对应的超像素块的目标方向；
    利用每个单通道图像中每两个超像素块的拟合熵值和灰度均值对所有超像素块进行分类得到多个类别的超像素块；
    获取每个单通道图像的每类超像素块中每个超像素块的每个边缘点和目标方向上对应的边缘点的灰度差值以及距离；对每个边缘点和目标方向上对应的边缘点的得到的灰度差值和距离求比值，得到每个边缘点和目标方向上对应的边缘点之间的灰度变化速率；
    利用每个超像素块中每个边缘点和目标方向上对应的边缘点之间的灰度变化速率的均值得到每个超像素块的预测模型系数；将每类超像素块中所有超像素块的预测模型系数的均值作为每类超像素块的预测模型系数；
    利用每个单通道图像中每类超像素块的预测模型系数和每类超像素块中像素点的灰度值得到每类超像素块的预测偏量；根据每个单通道图像中每类超像素块的预测偏量、预测模型系数和边缘点的灰度值得到每个单通道图像中每类超像素块中每个像素点的预测灰度值；
    利用灰度图像中每个像素点的灰度值和每个像素点在每个单通道图像中的预测灰度值得到灰度图像中每个像素点的预测误差，对每个像素点的预测误差进行编码并存储。
  2. 根据权利要求1所述的一种环境安防工程监测数据管理方法，其特征在于，得到每类超像素块的预测偏量的方法包括：
    将每个单通道图像的每个超像素块中每个边缘点和目标方向上对应的边缘点记为一对边缘点对，每对边缘点对对应一条边缘点连线；
    将每个边缘点对中灰度值较小的边缘点记为目标点；
    根据每个超像素块中每条边缘点连线上的目标点的灰度值和超像素块所在类别的预测模型系数得到每条边缘点连线上除边缘点以外每个像素点的预测值；
    利用每个超像素块中每个像素点的预测值和灰度值之间的差值得到每个像素点的预测偏量；
    将每类超像素块中所有像素点的预测偏量的众数作为每类超像素块的预测偏量。
  3. 根据权利要求2所述的一种环境安防工程监测数据管理方法，其特征在于，获取每个超像素块中每个像素点的预测偏量的公式为：
    βk＝Gb+αM×K-Gk
    其中，βk表示像素点k的预测偏量；Gb表示像素点k所在的边缘点连线上的目标点的灰度值；αM表示像素点k所在的超像素块所属的第M类超像素块的预测模型系数；K表示像素点k为边缘点连线上目标点以后的第K个像素点对应的数值；Gk表示像素点k的灰度值。
  4. 根据权利要求2所述的一种环境安防工程监测数据管理方法，其特征在于，得到每个像素点的预测灰度值的方法为：
    获取每个像素点与所在边缘点连线上目标点之间的距离；
    利用得到的距离与像素点所在超像素块所属的超像素块类别的预测模型系数相乘，得到每个像素点对应的乘积；
    将每个像素点得到的乘积与所在边缘点连线上目标点的灰度值相加，再加上所属超像素块类别的预测偏量，得到每个像素点的预测灰度值。
  5. 根据权利要求1所述的一种环境安防工程监测数据管理方法，其特征在于，获取拟合直线的拟合质量的方法包括：
    获取边缘点连线上每对相邻像素点的灰度差值到拟合直线的距离的方差，将每个边缘点连线得到的方差作为边缘点连线对应的拟合直线的拟合质量。
  6. 根据权利要求1所述的一种环境安防工程监测数据管理方法，其特征在于，得到灰度图像中每个像素点的预测误差的方法为：
    利用每个像素点在多个单通道图像中的预测灰度值得到每个像素点在灰度图像中的目标预测灰度值；
    将灰度图像中每个像素点的目标预测灰度值和灰度值的差值的绝对值，作为灰度图像中每个像素点的预测误差。
  7. 根据权利要求1所述的一种环境安防工程监测数据管理方法，其特征在于，得到多个类别的超像素块的方法包括：
    利用每个超像素块的拟合熵值和灰度均值组成对应的超像素块的特征向量；
    对每两个超像素块的特征向量计算余弦相似度作为对应两个超像素块的相关度；
    分类原则为：同一类别中任意两个超像素块的相关度都在预设的相关度阈值区间中；
    利用所述分类原则对每个单通道图像中的超像素块进行分类得到多个类别的超像素块。
  8. 一种环境安防工程监测数据管理系统，包括，数据采集单元，数据分析单元，数据压缩单元，其特征在于：
    数据采集单元，用于获取安防监控视频的每帧图像的灰度图像和多个单通道图像；获取每个单通道图像的多个超像素块；
    数据分析单元，用于利用每个单通道图像的每个超像素块中每个边缘点与不同方向的边缘点相连得到不同方向的边缘点连线，获取不同方向的每个边缘点连线上相邻像素点的灰度差值的拟合直线以及拟合直线的拟合质量；
    利用每个超像素块中不同方向的每个边缘点连线对应的拟合直线的拟合质量，得到每个超像素块在不同方向的拟合质量熵值；选取最小拟合质量熵值作为超像素块的拟合熵值，将得到最小拟合质量熵值的方向作为对应的超像素块的目标方向；
    利用每个单通道图像中每两个超像素块的拟合熵值和灰度均值对所有超像素块进行分类得到多个类别的超像素块；
    数据编码单元，利用每个单通道图像的每类超像素块中每个边缘点和目标方向上对应的边缘点的灰度值和距离，得到每类超像素块的预测模型系数；
    利用每个单通道图像中每类超像素块的预测模型系数和每类超像素块中像素点的灰度值得到每类超像素块的预测偏量；根据每个单通道图像中每类超像素块的预测偏量、 预测模型系数和边缘点的灰度值得到每个单通道图像中每类超像素块中每个像素点的预测灰度值；
    利用灰度图像中每个像素点的灰度值和每个像素点在每个单通道图像中的预测灰度值得到灰度图像中每个像素点的预测误差，对每个像素点的预测误差进行编码并存储。
PCT/CN2023/086005 2022-12-16 2023-04-03 一种环境安防工程监测数据管理方法及系统 WO2023134791A2 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211619607.9 2022-12-16
CN202211619607.9A CN115914634A (zh) 2022-12-16 2022-12-16 一种环境安防工程监测数据管理方法及系统

Publications (2)

Publication Number Publication Date
WO2023134791A2 true WO2023134791A2 (zh) 2023-07-20
WO2023134791A3 WO2023134791A3 (zh) 2023-09-21

Family

ID=86471077

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/086005 WO2023134791A2 (zh) 2022-12-16 2023-04-03 一种环境安防工程监测数据管理方法及系统

Country Status (3)

Country Link
CN (1) CN115914634A (zh)
LU (1) LU504265B1 (zh)
WO (1) WO2023134791A2 (zh)

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116665137A (zh) * 2023-08-01 2023-08-29 聊城市彩烁农业科技有限公司 一种基于机器视觉的畜牧养殖废水处理方法
CN116778431A (zh) * 2023-08-25 2023-09-19 青岛娄山河水务有限公司 基于计算机视觉的污泥处理自动监测方法
CN116863253A (zh) * 2023-09-05 2023-10-10 光谷技术有限公司 基于大数据分析的运维风险预警方法
CN117079197A (zh) * 2023-10-18 2023-11-17 山东诚祥建设集团股份有限公司 一种智慧工地管理方法及系统
CN117115196A (zh) * 2023-10-25 2023-11-24 东莞雕宝自动化设备有限公司 一种切割机刀具磨损视觉检测方法及系统
CN117221609A (zh) * 2023-11-07 2023-12-12 深圳微云通科技有限公司 一种高速公路收费业务集中监控值机系统
CN117237339A (zh) * 2023-11-10 2023-12-15 山东多沃基础工程有限公司 一种基于图像处理的地螺丝打孔点位选取方法及系统
CN117351433A (zh) * 2023-12-05 2024-01-05 山东质能新型材料有限公司 基于计算机视觉的胶固砂浆饱满度监控系统
CN117478891A (zh) * 2023-12-28 2024-01-30 辽宁云也智能信息科技有限公司 一种建筑施工智慧管理系统
CN117831135A (zh) * 2024-03-04 2024-04-05 陕西一览科技有限公司 一种基于图像处理的人迹检测方法
CN117853933A (zh) * 2024-03-07 2024-04-09 山东矿通智能装备有限公司 一种用于露天煤矿开采的煤层识别方法
CN117853493A (zh) * 2024-03-08 2024-04-09 山东天意机械股份有限公司 一种基于图像处理的墙板生产质量检测方法
CN117876971A (zh) * 2024-03-12 2024-04-12 武汉同创万智数字科技有限公司 基于机器视觉的建筑施工安全监测及预警方法
CN117853493B (zh) * 2024-03-08 2024-05-24 山东天意机械股份有限公司 一种基于图像处理的墙板生产质量检测方法

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115914634A (zh) * 2022-12-16 2023-04-04 苏州迈创信息技术有限公司 一种环境安防工程监测数据管理方法及系统
CN116233479B (zh) * 2023-04-28 2023-07-21 中译文娱科技(青岛)有限公司 基于数据处理的直播信息内容审核系统及方法
CN116320042B (zh) * 2023-05-16 2023-08-04 陕西思极科技有限公司 边缘计算的物联终端监测控制系统
CN116703787B (zh) * 2023-08-09 2023-10-31 中铁建工集团第二建设有限公司 一种建筑施工安全风险预警方法及系统
CN117615088B (zh) * 2024-01-22 2024-04-05 沈阳市锦拓电子工程有限公司 一种安全监控的视频数据高效存储方法
CN117692011B (zh) * 2024-01-29 2024-04-30 航天亮丽电气有限责任公司 一种用于消防救援环境监测系统的监测数据预警方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5482007B2 (ja) * 2008-10-08 2014-04-23 株式会社ニコン 画像処理方法
WO2018176185A1 (zh) * 2017-03-27 2018-10-04 中国科学院深圳先进技术研究院 一种纹理合成方法及其装置
US20230289979A1 (en) * 2020-11-13 2023-09-14 Zhejiang University A method for video moving object detection based on relative statistical characteristics of image pixels
CN115439474B (zh) * 2022-11-07 2023-01-24 山东天意机械股份有限公司 一种电力设备故障快速定位方法
CN115914634A (zh) * 2022-12-16 2023-04-04 苏州迈创信息技术有限公司 一种环境安防工程监测数据管理方法及系统

Cited By (26)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116665137A (zh) * 2023-08-01 2023-08-29 聊城市彩烁农业科技有限公司 一种基于机器视觉的畜牧养殖废水处理方法
CN116665137B (zh) * 2023-08-01 2023-10-10 聊城市彩烁农业科技有限公司 一种基于机器视觉的畜牧养殖废水处理方法
CN116778431A (zh) * 2023-08-25 2023-09-19 青岛娄山河水务有限公司 基于计算机视觉的污泥处理自动监测方法
CN116778431B (zh) * 2023-08-25 2023-11-10 青岛娄山河水务有限公司 基于计算机视觉的污泥处理自动监测方法
CN116863253A (zh) * 2023-09-05 2023-10-10 光谷技术有限公司 基于大数据分析的运维风险预警方法
CN116863253B (zh) * 2023-09-05 2023-11-17 光谷技术有限公司 基于大数据分析的运维风险预警方法
CN117079197A (zh) * 2023-10-18 2023-11-17 山东诚祥建设集团股份有限公司 一种智慧工地管理方法及系统
CN117079197B (zh) * 2023-10-18 2024-03-05 山东诚祥建设集团股份有限公司 一种智慧工地管理方法及系统
CN117115196B (zh) * 2023-10-25 2024-02-06 东莞雕宝自动化设备有限公司 一种切割机刀具磨损视觉检测方法及系统
CN117115196A (zh) * 2023-10-25 2023-11-24 东莞雕宝自动化设备有限公司 一种切割机刀具磨损视觉检测方法及系统
CN117221609B (zh) * 2023-11-07 2024-03-12 深圳微云通科技有限公司 一种高速公路收费业务集中监控值机系统
CN117221609A (zh) * 2023-11-07 2023-12-12 深圳微云通科技有限公司 一种高速公路收费业务集中监控值机系统
CN117237339A (zh) * 2023-11-10 2023-12-15 山东多沃基础工程有限公司 一种基于图像处理的地螺丝打孔点位选取方法及系统
CN117237339B (zh) * 2023-11-10 2024-02-27 山东多沃基础工程有限公司 一种基于图像处理的地螺丝打孔点位选取方法及系统
CN117351433A (zh) * 2023-12-05 2024-01-05 山东质能新型材料有限公司 基于计算机视觉的胶固砂浆饱满度监控系统
CN117351433B (zh) * 2023-12-05 2024-02-23 山东质能新型材料有限公司 基于计算机视觉的胶固砂浆饱满度监控系统
CN117478891B (zh) * 2023-12-28 2024-03-15 辽宁云也智能信息科技有限公司 一种建筑施工智慧管理系统
CN117478891A (zh) * 2023-12-28 2024-01-30 辽宁云也智能信息科技有限公司 一种建筑施工智慧管理系统
CN117831135A (zh) * 2024-03-04 2024-04-05 陕西一览科技有限公司 一种基于图像处理的人迹检测方法
CN117831135B (zh) * 2024-03-04 2024-05-10 陕西一览科技有限公司 一种基于图像处理的人迹检测方法
CN117853933A (zh) * 2024-03-07 2024-04-09 山东矿通智能装备有限公司 一种用于露天煤矿开采的煤层识别方法
CN117853933B (zh) * 2024-03-07 2024-05-17 山东矿通智能装备有限公司 一种用于露天煤矿开采的煤层识别方法
CN117853493A (zh) * 2024-03-08 2024-04-09 山东天意机械股份有限公司 一种基于图像处理的墙板生产质量检测方法
CN117853493B (zh) * 2024-03-08 2024-05-24 山东天意机械股份有限公司 一种基于图像处理的墙板生产质量检测方法
CN117876971A (zh) * 2024-03-12 2024-04-12 武汉同创万智数字科技有限公司 基于机器视觉的建筑施工安全监测及预警方法
CN117876971B (zh) * 2024-03-12 2024-05-28 武汉同创万智数字科技有限公司 基于机器视觉的建筑施工安全监测及预警方法

Also Published As

Publication number Publication date
WO2023134791A3 (zh) 2023-09-21
CN115914634A (zh) 2023-04-04
LU504265B1 (en) 2023-07-31

Similar Documents

Publication Publication Date Title
WO2023134791A2 (zh) 一种环境安防工程监测数据管理方法及系统
US20220217337A1 (en) Method, codec device for intra frame and inter frame joint prediction
CN111526361B (zh) 图像编解码方法及装置
CN103037214A (zh) 视频压缩方法
CN107846589B (zh) 一种基于局部动态量化的图像压缩方法
CN101115131A (zh) 基于像素空间相关性的图像融合效果实时评价方法及装置
CN112291562B (zh) 针对h.266/vvc的快速cu分区和帧内模式决策方法
CN111046766A (zh) 一种行为识别方法、装置及计算机存储介质
CN111447452A (zh) 数据编码方法及系统
CN112399177B (zh) 一种视频编码方法、装置、计算机设备及存储介质
CN116233479B (zh) 基于数据处理的直播信息内容审核系统及方法
CN105263026A (zh) 基于概率统计与图像梯度信息的全局矢量获取方法
CN115665359B (zh) 一种环境监测数据智能压缩方法
CN109547798B (zh) 一种快速的hevc帧间模式选择方法
CN113573058B (zh) 一种基于时空显著性融合的帧间图像编码方法
CN113365080B (zh) 串编码技术的编解码方法、设备及存储介质
US7170940B2 (en) Moving picture encoding system
CN112565760B (zh) 串编码技术的编码方法、设备及存储介质
CN112218083B (zh) 高效视频编码标准帧内图像码率估计方法
CN115278225A (zh) 色度编码模式的选择方法、装置和计算机设备
CN110519597B (zh) 一种基于hevc的编码方法、装置、计算设备和介质
CN115278255B (zh) 一种用于力量器械安全管理的数据存储系统
CN117115152B (zh) 基于图像处理的钢绞线生产监测方法
CN109587496B (zh) 一种Skip块判别方法、编码器、电子设备及可读存储介质
Li et al. You Can Mask More For Extremely Low-Bitrate Image Compression

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: 23740131

Country of ref document: EP

Kind code of ref document: A2