Image recognition security processing method and system for security curtain

Abstract

The application discloses an image recognition security processing method and system for a security curtain, and relates to the technical field of security curtains, and solves the problem of rough original object contour feature processing. The application can effectively lock dynamic objects by analyzing the pixel point difference between adjacent frames of images obtained by a high-definition camera. When confirming dynamic features, the system performs point position analysis and gradient calculation to accurately divide difference areas and contour areas, selects the largest area from multiple feature areas as the dynamic feature, ensures the accuracy of recognition, and reduces the misjudgment rate. In each contour partition, the feature difference values of different internal connecting lines are compared, the connecting line with the smallest difference value is selected as the uniform refinement connecting line, and thus the fine processing of the dynamic feature edge contour is realized, so that the extracted feature contour is more accurate and can better reflect the true shape of the object.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of security curtain screens, in particular to an image recognition security processing method and system for a security curtain screen. BACKGROUND

[0002] In today's society, the importance of security is increasingly prominent, and security curtain screens, as key facilities for ensuring the safety of specific areas, are widely used in military bases, important industrial sites, high-end residential areas and other scenes with extremely high safety requirements.

[0003] The traditional security system is rough in processing object features, and cannot finely process the object contour, resulting in inaccurate extracted features, which is difficult to meet the requirements of scenes with extremely high safety requirements. When facing objects with complex shapes, the features extracted by the traditional method are difficult to reflect the true shape of the object, which brings difficulties to subsequent analysis and judgment. The application combines dynamic features with a two-dimensional coordinate system, determines the center point by calculating the mean coordinate, divides the contour partition according to the feature line length, and selects the mean refinement line in each partition to refine the edge contour of the dynamic feature, which can obtain more accurate object features and refine the edge contour of the object features, so that the overall contour of the object features is more obvious, and the recognition accuracy is better. SUMMARY

[0004] In view of the deficiencies of the prior art, the application provides an image recognition security processing method and system for a security curtain screen, which solves the problem of rough processing of original object contour features.

[0005] To achieve the above purpose, the application is implemented by the following technical scheme: an image recognition security processing method for a security curtain screen, comprising the following steps:

[0006] Step 1: Based on the laser detector set in the security curtain screen, it is identified whether there is an article passing through, and based on the identification result, an image recognition signal is confirmed, and the specific method is:

[0007] The detection signal associated with the laser detector is monitored in real time, and when the signal value associated with the detection signal does not belong to the preset interval, it represents that there is an article passing through, and an image recognition signal is directly generated, and the end point values of the preset interval are both preset values;

[0008] Step 2: Based on the generated image recognition signal, the high-definition camera is controlled to acquire the image of the area where the security curtain screen is located, and the dynamic features existing in the multiple images are locked by analyzing the pixel point difference between adjacent images, and the specific method is:

[0009] The images acquired from different frames are sorted according to the time sequence to confirm the image set;

[0010] Randomly select a point from a single image and mark it as the zero point. Confirm the positional characteristics of other points based on the positional differences between other points and the zero point. In this way, confirm the positional characteristics of several points associated with different frames of images in the image set.

[0011] The pixel values ​​associated with different frames within the image set are labeled as X. i-k Where i represents different frame images, and k represents the points associated with different location features, the pixel difference of points with the same location feature in adjacent frame images is confirmed using XC. j-k =|X j-k -X (j+1)-k |Confirm the pixel difference at the point XC j-k , where j∈i, and i = 1, 2, ..., n, then 1≤j≤n-1;

[0012] This will satisfy: XC j-k Points with a value ≥ Y1 are designated as difference points, and the area covered by several difference points within the corresponding frame image is designated as a difference region. Then, the outer region of the difference region is contour-defined: based on the different pixel values ​​associated with different pixels within the frame image, the Sobe I algorithm is used to determine the lateral gradient H associated with each pixel. k and vertical gradient S k Calibration was performed using: Confirm the overall gradient ZH associated with the corresponding pixel. k This will satisfy ZH k > The pixels of Y2 are labeled as gradient pixels. The internal region associated with several consecutive gradient pixels is recorded as the contour region. Conversely, no labeling is performed, and Y1 and Y2 are preset values.

[0013] Based on the contour regions and difference regions identified within the corresponding frame image, the contour region to which the difference region belongs is identified, and this contour region is recorded as a feature region of this frame image.

[0014] Based on the different feature regions determined within different frames of images, the feature region with the largest area is selected and recorded as the dynamic feature of this image set.

[0015] Step 3: Based on the dynamic features associated with the corresponding image set, firstly identify contour partitions within the edge contours associated with the dynamic features, then identify mean-weighted thinning lines within different contour partitions, and refine the edge contours of this dynamic feature according to the different mean-weighted thinning lines associated with different contour partitions. The specific method is as follows:

[0016] The calibrated dynamic feature is combined with a two-dimensional coordinate system to confirm different two-dimensional coordinates associated with different pixel points in the dynamic feature, and a plurality of sets of two-dimensional coordinates are processed by averaging to confirm an average coordinate, and the confirmed average coordinate is calibrated in the dynamic feature as a center point of the dynamic feature;

[0017] An edge point associated with the outermost edge of the dynamic feature is confirmed, the center point and different edge points are connected to lock a point connection line, a plurality of sets of gradient pixel points on the periphery of the dynamic feature are confirmed to cover a gradient region, and a part of the point connection line located in the gradient region is confirmed as a feature connection line corresponding to the gradient region, and the length of each feature connection line is determined;

[0018] The feature confirmation is performed on adjacent feature connection lines to lock a clustered feature connection line set: a set of feature connection lines is randomly selected, and the clockwise direction is taken as the processing direction, and the feature difference = | the length of the previous set of feature connection lines - the length of the next set of feature connection lines | is adopted, if the feature difference ≤ Y3, the two adjacent sets of feature connection lines are calibrated as the clustered feature connection line set, Y3 is a preset value, otherwise, the corresponding feature connection line is calibrated as a separation line;

[0019] Based on the positions of different separation lines in the gradient region, the corresponding gradient region is divided into a plurality of different contour partitions;

[0020] The specific way of calibrating different mean value refinement connection lines associated with different contour partitions is:

[0021] A set of pixel points is randomly selected from the pixel points associated with each different vertical column in the contour partition, and is recorded as a selected point, a plurality of selected points are connected to confirm an internal connection line, the pixel values corresponding to a plurality of pixel points associated with the upper part of the internal connection line are summed to confirm an upper feature sum Z1, the pixel values corresponding to a plurality of pixel points associated with the lower part of the internal connection line are summed to confirm a lower feature sum Z2, and the feature difference CZ is confirmed by CZ = | Z1 - Z2 |;

[0022] A plurality of different point selection processes are performed, the internal connection lines associated with each selected process are different, and different feature differences CZ associated with different internal connection lines are confirmed, the minimum value is selected from a plurality of different feature differences CZ, the internal connection line associated with the minimum value is recorded as the mean value refinement connection line of the contour partition, the mean value refinement connection lines associated with each contour partition are connected as the edge contour line of the dynamic feature after refinement processing, and the edge contour line is taken as the feature contour of the dynamic feature;

[0023] Step four, the dynamic feature after completing the contour thinning processing is pre-alarm checked, the dynamic feature is compared with the feature contour pre-stored in the model library, whether there is a comparison result is confirmed, if there is, pre-alarm processing is carried out, if not, it is not necessary to process, the specific mode is:

[0024] The dynamic feature is compared with the feature contour pre-stored in the model library: the center point of the dynamic feature is coincided with the center point of the feature contour, and the dynamic feature is controlled to rotate, whether the coincidence rate of the dynamic feature and the feature contour reaches 90% is identified, if it reaches, it represents that there is a comparison result, then the alarm is directly controlled to carry out pre-alarm processing and warn the external operator; if it does not reach, it does not carry out any processing.

[0025] Preferably, an image recognition security processing system for a security curtain comprises:

[0026] The feature confirmation end identifies whether there is an article passing based on the laser detector arranged in the security curtain, and confirms the image recognition signal based on the identification result;

[0027] The dynamic feature locking end controls the high-definition camera to acquire the image of the area where the security curtain is located based on the generated image recognition signal, and locks the dynamic feature existing in multiple images by analyzing the pixel point difference between adjacent images;

[0028] The contour thinning processing end confirms the contour partition from the edge contour associated with the dynamic feature, then confirms the equal quantity thinning line from different contour partitions, and thins the edge contour of the dynamic feature according to the different equal quantity thinning lines associated with different contour partitions;

[0029] The pre-alarm processing end pre-alarm checks the dynamic feature after completing the contour thinning processing, compares the dynamic feature with the feature contour pre-stored in the model library, confirms whether there is a comparison result, if there is, pre-alarm processing is carried out, if not, it is not necessary to process.

[0030] The present application provides an image recognition security processing method and system for a security curtain.

[0031] The present application can effectively lock the dynamic object by analyzing the pixel point difference between adjacent images acquired by the high-definition camera; when confirming the dynamic feature, the difference area and the contour area are accurately divided through the point position analysis and gradient calculation of the system, the largest area is selected from multiple feature areas as the dynamic feature, the accuracy of identification is ensured, and the misjudgment rate is reduced.

[0032] In each contour partition, by comparing the feature difference of different internal connection lines, the connection line with the minimum difference value is selected as the equalization refinement connection line, so as to realize the refinement processing of the dynamic feature edge contour, and make the extracted feature contour more accurate and more reflect the true shape of the object.

[0033] The processed dynamic feature is compared with the pre-stored feature contour in the model library, and the possible security threat is quickly and accurately identified through the center point coincidence and the contour coincidence rate. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 It is a method flowchart of the present application;

[0035] Figure 2 It is a principle framework diagram of the present application. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0037] First embodiment

[0038] Please refer to Figure 1 The present application provides an image recognition security processing method for security curtain, comprising the following steps:

[0039] Step one, based on the laser detector set in the security curtain to identify whether there is an article passing, and based on the identification result, confirm the image recognition signal, specifically, the laser detector includes a laser head assembly, a battery, and corresponding legs, the laser head assembly can emit laser to the monitored area, the scanning angle range of the laser head assembly is 0-90°, the scanning length is 30m and the height is 20m, forming a laser curtain in front, when there is an object passing through the laser curtain, the laser signal detected by the laser detector changes, so it represents that there is an article passing, and directly generates an image recognition signal for image processing, wherein the operation mode is: real-time monitoring of the detection signal associated with the laser detector, when the signal value associated with the detection signal is not in the preset interval (that is, the corresponding laser signal has intensity change or value fluctuation), it represents that there is an article passing, and directly generates an image recognition signal, the end point value of the preset interval is a preset value, which is determined in advance by relevant personnel according to experience;

[0040] Step two, based on the generated image recognition signal, control the high-definition camera to acquire the image of the area where the security curtain net is located, and lock the dynamic characteristics existing in multiple frames of images by analyzing the pixel point difference between adjacent frames of images. Specifically, the camera and the corresponding laser detector are placed in the same place, so the camera can effectively acquire the specific image in front, and each frame of image associated during the acquisition process is different, so the corresponding dynamic object can be confirmed based on the image of the corresponding adjacent frame to determine the corresponding dynamic characteristics. The specific way to lock the dynamic characteristics is:

[0041] Sort the acquired images belonging to different frames according to the time sequence to confirm the image set;

[0042] Randomly select a point from a single image and mark it as zero, and confirm the position characteristics of other points based on the position difference between other points and the zero point. In this way, the position characteristics of several points associated with different frames of images in the image set are confirmed;

[0043] And the pixel values associated with different frames of images in the image set are marked as X i-k , where i represents different frames of images, and k represents the point associated with different position characteristics. Confirm the pixel difference of the same position characteristics point in adjacent frames of images, and use XC j-k = |X j-k -X (j+1)-k to confirm the pixel difference XC j-k , where j∈i, i=1, 2, …, n, then 1≤j≤n-1. The subsequent difference point marking is directly marked in the images of the adjacent frames, so the images existing in front and back are the corresponding intermediate images. According to the point characteristics associated with the intermediate image and the front and back images, the corresponding difference point can be determined.

[0044] Mark the point that satisfies XC j-k ≥Y1 as a difference point, and mark the area covered by several difference points in the corresponding frame of image as a difference area, and then mark the outline of the peripheral area of the difference area: based on the different pixel values associated with different pixel points in the frame image, use Sobe l algorithm to mark the horizontal gradient H k and vertical gradient S k associated with each pixel point (as the way to confirm gradient data through algorithm is common in the prior art, so it is not described in detail here. By confirming the grid associated with the corresponding pixel point and the related weight, the comprehensive gradient of the corresponding pixel point is determined), and use: to confirm the comprehensive gradient ZH k associated with the corresponding pixel point, and ZH kThe pixel point of Y2 is marked as a gradient pixel point, and the internal region associated with a plurality of continuous gradient pixel points is recorded as a contour region (the gradient pixel points are confirmed, one object feature is associated with the pixel points on the periphery, and the intermediate region included in the gradient pixel points is the corresponding contour region), otherwise, no marking is performed, Y1 and Y2 are preset values, and specific values are determined by an operator according to experience;

[0045] Based on the contour region and the difference region confirmed in the corresponding frame image, the contour region to which the difference region belongs is confirmed, and the contour region is recorded as the feature region of the frame image. The difference region is a pixel point difference region generated by the corresponding dynamic object during movement. The difference region is generally generated by the change of the corresponding pixel point during movement of the corresponding object, so that the corresponding difference region can be confirmed, and the specific contour region to which the difference region belongs exists;

[0046] Based on the different feature regions determined in different frame images, the feature region with the largest area is selected and recorded as the dynamic feature of the image set (different feature regions have different area values, the feature region associated with the maximum area value is determined from a plurality of area values, and the feature region with the maximum area value is recorded as the dynamic feature of the corresponding image set);

[0047] Step three, based on the dynamic feature associated with the corresponding image set, the contour partition is confirmed from the edge contour associated with the dynamic feature, then the equal-quantity refined connection lines are confirmed from different contour partitions, and the edge contour of the dynamic feature is refined according to different equal-quantity refined connection lines associated with different contour partitions. The specific refinement processing steps are as follows:

[0048] I. Confirming the contour partition from the edge contour associated with the dynamic feature:

[0049] The marked dynamic feature is combined with a two-dimensional coordinate system to confirm different two-dimensional coordinates associated with different pixel points in the dynamic feature, and a plurality of groups of two-dimensional coordinates are subjected to mean value processing to confirm the mean value coordinates. The confirmed mean value coordinates are marked in the dynamic feature and recorded as the center point of the dynamic feature;

[0050] The edge point associated with the outermost edge of the dynamic feature is confirmed, the center point and different edge points are connected, the point connection line is locked, a plurality of groups of gradient pixel points on the periphery of the dynamic feature are confirmed, the gradient region covered by the plurality of groups of gradient pixel points is confirmed, the part of the point connection line located in the gradient region is recorded as the feature connection line of the corresponding gradient region, and the length of each feature connection line is determined;

[0051] Confirming features of adjacent feature lines, locking the clustering feature line set: randomly selecting a group of feature lines, and taking the clockwise direction as the processing direction, using: feature difference = |the length of the previous group of feature lines - the length of the next group of feature lines|, if the feature difference ≤ Y3, then the adjacent two groups of feature lines are marked as the clustering feature line set (if the subsequent length change still belongs to the corresponding numerical range, then it is still marked in the same clustering feature line set), Y3 is a preset value, and its specific value is determined by the operator according to experience, otherwise, the corresponding feature line is marked as a separation line;

[0052] Based on the location of different separation lines in the gradient region, the corresponding gradient region is divided into multiple different contour partitions (that is, the region associated between adjacent segmentation lines, and the corresponding region is the divided contour partition). There are different feature lines in the overall edge contour of the corresponding dynamic feature. The length of the different feature lines is confirmed, and based on the confirmed specific numerical value, the feature is divided. The corresponding separation line is locked from the divided specific length numerical value, and the region is divided based on the corresponding separation line, so that different contour partitions can be locked, and subsequent specific confirmation of the uniform refinement line can be facilitated.

[0053] II. The specific way of marking different uniform refinement lines associated with different contour partitions is:

[0054] The pixel points are sorted in a grid form, so there are multiple related pixel points in each vertical direction. A group of pixel points is randomly selected from the pixel points associated with each different vertical row in the contour partition, and is marked as a selected point. A number of selected points are connected to confirm a group of internal lines. The pixel values corresponding to the multiple groups of pixel points associated with the upper part of the internal lines are summed to confirm the upper feature sum Z1. The pixel values corresponding to the multiple groups of pixel points associated with the lower part of the internal lines are summed to confirm the lower feature sum Z2. The feature difference CZ is confirmed using CZ = |Z1-Z2|.

[0055] A plurality of different point selection processes are performed, and the internal lines associated with each selected process are different. Different feature differences CZ associated with different internal lines are confirmed. The minimum value is selected from a plurality of different feature differences CZ. The internal line associated with the minimum value is marked as the uniform refinement line of this contour partition. The uniform refinement lines associated with each contour partition are connected to serve as the edge contour line after the dynamic feature refinement processing, and the edge contour line is used as the feature contour of the dynamic feature.

[0056] Specifically, the corresponding overall edge profile is divided into a plurality of different profile partitions, and the different feature connection lines associated with the different profile partitions are analyzed to confirm the specific connection lines with consistent parameters, and the specific connection lines are taken as a connection set. The separation line not belonging to the corresponding connection set is the best separation feature. The corresponding edge profile is profile-partitioned and adjusted to achieve better profile confirmation processing effect. Then, the connection lines in the profile are refined to confirm the specific feature profile of the corresponding dynamic feature, thereby achieving better feature refinement processing process.

[0057] Step four, pre-warning verification is performed on the dynamic feature after the profile refinement processing. The dynamic feature is compared with the pre-stored feature profile in the model library to confirm whether there is a comparison result. If there is, pre-warning processing is performed. If there is not, no processing is needed. The specific way of pre-warning verification is as follows:

[0058] The dynamic feature is compared with the pre-stored feature profile in the model library one by one: the center point of the dynamic feature is coincided with the center point of the feature profile, and the dynamic feature is controlled to rotate to identify whether the coincidence rate of the dynamic feature and the feature profile reaches 90%. If it does, it means that there is a comparison result, and the alarm is directly controlled to perform pre-warning processing to alert the external operator. If it does not, no processing is needed.

[0059] Specifically, after the center point coincidence processing is completed, the coincidence rate between the corresponding profiles is identified to determine whether it meets the standard. If it does, it means that the feature belongs to a human feature or an animal feature, and pre-warning is directly performed. If it does not, it means that the intruder may be a floating object or other object, and no manual intervention is needed.

[0060] Second embodiment

[0061] In combination Figure 2 An image recognition security processing system for a security curtain, comprising:

[0062] A feature confirmation end, configured to identify whether an object passes through based on a laser detector arranged in the security curtain, and to confirm an image recognition signal based on the identification result;

[0063] A dynamic feature locking end, configured to control a high-definition camera to acquire images of an area where the security curtain is located based on the generated image recognition signal, and to lock dynamic features existing in multiple images by analyzing pixel point differences between adjacent images.

[0064] The contour thinning processing end, based on the dynamic feature associated with the corresponding image set, confirms the contour partition from the edge contour associated with the dynamic feature, then confirms the equal quantity thinning connection from different contour partitions, and according to the different equal quantity thinning connections associated with different contour partitions, performs the thinning processing on the edge contour of the dynamic feature.

[0065] The early warning processing end performs early warning verification on the dynamic feature after the contour thinning processing is completed, compares the dynamic feature with the feature contour pre-stored in the model library, confirms whether there is a comparison result, if there is, performs early warning processing, and if not, no processing is needed.

[0066] Part of the data in the above formula is dimensionless for numerical calculation, and the contents not described in detail in the specification all belong to the prior art known to those skilled in the art.

[0067] The above embodiments are only used to illustrate the technical method of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. An image recognition security processing method for a security screen, characterized by, The method comprises the following steps: Step one: based on the laser detector set in the security curtain, identify whether there is an article passing, and based on the identification result, confirm the image recognition signal; Step two: based on the generated image recognition signal, control the high-definition camera to acquire the image of the area where the security curtain is located, and lock the dynamic features existing in multiple images by analyzing the pixel point difference between adjacent image frames; Step three: based on the dynamic features associated with the corresponding image set, preferentially confirm the contour partition from the edge contour associated with the dynamic features, then confirm the equalization refinement connection from different contour partitions, and refine the edge contour of the dynamic features according to different equalization refinement connections associated with different contour partitions, the specific method is: Combine the calibrated dynamic features with the two-dimensional coordinate system to confirm the different two-dimensional coordinates associated with different pixel points in the dynamic features, then perform mean value processing on the several groups of two-dimensional coordinates to confirm the mean value coordinates, and mark the confirmed mean value coordinates in the dynamic features as the center point of the dynamic features; Confirm the edge point associated with the outermost edge of the dynamic features, connect the center point with different edge points to lock the point connection line, then confirm the gradient area covered by the several groups of gradient pixel points based on the several groups of gradient pixel points confirmed on the periphery of the dynamic features, confirm the part of the point connection line located in the gradient area as the feature connection line of the corresponding gradient area, and determine the connection line length associated with each feature connection line; Confirm the feature of adjacent feature connection lines and lock the clustering feature connection line set: randomly select a group of feature connection lines, take the clockwise direction as the processing direction, and use: feature difference = |connection line length of the previous group of feature connection lines-connection line length of the next group of feature connection lines|, if the feature difference ≤Y3, then mark the adjacent two groups of feature connection lines as the clustering feature connection line set, Y3 is a preset value, otherwise, mark the corresponding feature connection line as a separation line; Based on the location of different separation lines in the gradient area, divide the corresponding gradient area into multiple different contour partitions; Step four: perform early warning verification on the dynamic features after contour refinement, compare the dynamic features with the feature contours pre-stored in the model library, confirm whether there is a comparison result, if there is, perform early warning processing, if not, no processing is needed.

2. The image recognition security processing method for security curtain according to claim 1, wherein, In step one, the specific method for confirming the image recognition signal is: Real-time monitor the detection signal associated with the laser detector, when the signal value associated with the detection signal does not belong to the preset interval, it represents that there is an article passing, and directly generates an image recognition signal, the end point values of the preset interval are both preset values. 3.The image recognition security processing method for a security curtain according to claim 1, wherein, In step two, the specific method for locking the dynamic features is: Sort the images acquired from different frames according to the time sequence to confirm the image set; Randomly select a point from a single image and mark it as zero, and confirm the position features of other points based on the position difference between other points and the zero point, and so on, to confirm the position features of several points associated with different image frames in the image set; And the pixel value associated with different frame images in the image set is calibrated as X i-k Where i represents different frame images, and k represents the point associated with different position characteristics. The pixel difference value of the point of the same position characteristics in the adjacent frame images is confirmed, and XC j-k = |X j-k -X (j+1)-k The pixel difference value of the point is confirmed XC j-k Where j ∈ i, and i = 1, 2, …, n is proposed, so that 1 ≤ j ≤ n-1. This will satisfy: XC j-k Points with a value ≥ Y1 are designated as difference points, and the area covered by several difference points within the corresponding frame image is designated as a difference region. Then, contour mapping is performed on the outer region of the difference region: based on the different pixel values ​​associated with different pixels within the frame image, the Sobel algorithm is used to map the lateral gradient H associated with each pixel. k and vertical gradient S k Calibration was performed using: Confirm the overall gradient ZH associated with the corresponding pixel. k This will satisfy ZH k >Y2 pixels are labeled as gradient pixels. The internal region associated with several consecutive gradient pixels is called the contour region, and Y1 and Y2 are preset values. Based on the contour region and the difference region identified in the corresponding frame image, the contour region to which the difference region belongs is identified, and the contour region is recorded as a feature region of the frame image; Based on the different feature regions determined in different frame images, the feature region with the largest area is selected as the dynamic feature of the image set.

4. The image recognition security processing method for security curtain according to claim 3, characterized in that, Not satisfying ZH k Pixels not satisfying Y2 are not calibrated at all.

5. The image recognition security processing method for security curtain according to claim 1, wherein, In step three, the specific way of calibrating the different mean value refinement connection lines associated with different contour partitions is: Randomly select a group of pixel points from the pixel points associated with each different vertical row in the contour partition, and mark them as selected points. Then, connect the selected points to identify a group of internal connection lines. Sum the pixel values corresponding to the groups of pixel points associated with the upper part of the internal connection lines to identify the upper feature sum Z1. Then, sum the pixel values corresponding to the groups of pixel points associated with the lower part of the internal connection lines to identify the lower feature sum Z2. Then, identify the feature difference CZ using CZ = |Z1-Z2|. Perform multiple different point selection processes, and the internal connection lines associated with each selection process are different. Identify the different feature differences CZ associated with different internal connection lines. From the multiple different feature differences CZ, select the minimum value. The internal connection line associated with the minimum value is the mean value refinement connection line of the contour partition. Connect the mean value refinement connection lines associated with each contour partition to obtain the edge contour line after the dynamic feature refinement processing. The edge contour line is the feature contour of the dynamic feature.

6. The image recognition security processing method for security curtain according to claim 1, wherein, In step four, the specific way of pre-warning verification of the dynamic feature is: Compare the dynamic feature with the feature contour pre-stored in the model library one by one: make the center point of the dynamic feature coincide with the center point of the feature contour, and control the rotation of the dynamic feature. Identify whether the coincidence rate of the dynamic feature and the feature contour reaches 90%. If it does, it means there is a comparison result, and the alarm is directly controlled for pre-warning processing to alert external operators. If it does not, no processing is performed.

7. An image recognition security processing system for security screen, which is operated according to the image recognition security processing method for security screen according to any one of claims 1-6, characterized in that, It includes: A feature identification end that identifies whether there is an object passing through based on the laser detector set in the security curtain, and identifies the image recognition signal based on the identification result; A dynamic feature locking end that controls the high-definition camera to acquire images of the area where the security curtain is located based on the generated image recognition signal, and locks the dynamic feature existing in multiple frame images by analyzing the pixel point differences between adjacent frame images; A contour refinement processing end that identifies contour partitions from the edge contour associated with the dynamic feature, identifies mean value refinement connection lines from different contour partitions, and refines the edge contour of the dynamic feature according to different mean value refinement connection lines associated with different contour partitions; A pre-warning processing end that performs pre-warning verification on the dynamic feature after contour refinement processing, compares the dynamic feature with the feature contour pre-stored in the model library, and identifies whether there is a comparison result. If there is, pre-warning processing is performed. If there is not, no processing is required.

Images