Multi-modal biological recognition method based on artificial intelligence

Abstract

The invention discloses a multi-modal biological recognition method based on artificial intelligence, relates to the field of biological recognition, solves the problem that the existing biological recognition method is poor in recognition effect, and comprises the following steps: S1, carrying out backlight intensity analysis and backlight angle analysis on each outdoor biological recognition area, s2, dividing the outdoor biological recognition region into a first type of biological recognition region and a second type of biological recognition region to obtain recognition region type division data, and performing biological recognition on the first type of biological recognition region to obtain recognition region type division data; and S3, performing biological palmprint image analysis on the to-be-identified organism in the second type biological identification area, and performing biological identification according to an analysis result. The multi-modal biological identification method can improve the accuracy.

Description

Technical Field

[0001] This invention belongs to the field of biometrics and involves artificial intelligence technology, specifically a multimodal biometric method based on artificial intelligence. Background Technology

[0002] Existing multimodal biometric methods have the following specific drawbacks when performing biometric identification:

[0003] 1. Existing multimodal biometric methods cannot perform backlight intensity and backlight angle analysis in outdoor biometric areas to obtain the backlight recognition influence coefficient for selecting the recognition method, resulting in a lag in the recognition process.

[0004] 2. Existing multimodal biometric methods can only compare pre-recorded palmprint images with the palmprint images to be identified in real time when performing palmprint recognition. They cannot combine historical data to train the image model, resulting in low biometric recognition efficiency.

[0005] To address this, we propose a multimodal biometric method based on artificial intelligence. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a multimodal biometric identification method based on artificial intelligence, aiming to improve the identification efficiency and accuracy of the multimodal biometric identification method.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a multimodal biometric identification method based on artificial intelligence, comprising the following specific steps:

[0008] Step S1: Obtain multiple outdoor biometric recognition areas, and perform backlight intensity analysis and backlight angle analysis on each outdoor biometric recognition area. Based on the analysis results, obtain the backlight recognition influence coefficient corresponding to each outdoor biometric recognition area to obtain regional backlight analysis data.

[0009] Step S2: Analyze the recognition method of the outdoor biometric area based on the backlight analysis data of the area, divide the outdoor biometric area into a first type of biometric area and a second type of biometric area based on the analysis results, obtain the recognition area type classification data, and use the first biometric device to perform biometric recognition on the first type of biometric area.

[0010] Step S3: Obtain the second type of biometric region by dividing the data according to the recognition region type, and use the second biometric device to perform biometric palmprint image analysis on the organism to be identified in the second type of biometric region, and perform biometric identification based on the analysis results.

[0011] Furthermore, step S1 also includes the following specific steps:

[0012] Step S11: Acquire the outdoor area that needs to be biometrically identified, obtain multiple outdoor biometric identification areas, select a sample biometric identification area from the acquired outdoor biometric identification areas, and set two biometric identification devices in the sample biometric identification area, and name them as the first biometric identification device and the second biometric identification device respectively.

[0013] Step S12: Perform bio-facial backlight analysis on the bio-identification area of ​​the sample, and obtain the bio-backlight recognition influence coefficient based on the analysis results;

[0014] Step S13: Obtain the backlighting impact coefficient for each outdoor biometric area to obtain regional backlighting analysis data.

[0015] Furthermore, step S12 also includes the following specific steps:

[0016] Step S121: If there is an organism to be identified in the sample biometric area, then create a spatial coordinate system for the identification area based on the relative position of the organism to be identified and the first biometric device;

[0017] Step S122: Analyze the backlight angle of the face of the organism to be identified based on the spatial coordinate system of the identification area, and obtain the backlight angle of the face of the organism to be identified based on the analysis results.

[0018] Step S123: Obtain the numerical value of the light intensity corresponding to the biometric area of ​​the sample at the current moment to obtain the real-time light intensity value of the area;

[0019] Step S124: Calculate the bio-backlight recognition influence coefficient corresponding to the bio-recognition area of ​​the sample by taking the facial backlight angle corresponding to the bio-recognition area and the real-time illumination intensity value of the area.

[0020] The backlighting impact coefficient corresponding to the biometric recognition area of ​​the sample is calculated using the following formula:

[0021]

[0022] Where Nxk is the biometric backlighting influence coefficient corresponding to the biometric recognition area of ​​the sample, Mjj is the facial backlighting angle corresponding to the biometric being to be recognized, Lss is the real-time illumination intensity value of the area, and Lck is the reference illumination intensity value.

[0023] Furthermore, step S121 also includes the following specific steps:

[0024] Within the sample biometric area, the geometric center point of the face corresponding to the organism to be identified is acquired by the image acquisition device to obtain the first position feature point. A horizontal straight line pointing due east is drawn through the first position feature point to obtain the first position feature line. A vertical straight line pointing to the sky is drawn through the first position feature point to obtain the second position feature line. A horizontal straight line pointing due north is drawn through the first position feature point to obtain the third position feature line.

[0025] The first position feature point is marked as the origin of the coordinate system, the first position feature line is marked as the x-axis, the second position feature line is marked as the y-axis, and the third position feature line is marked as the z-axis, thus obtaining the spatial coordinate system of the recognition area.

[0026] Furthermore, step S122 also includes the following specific steps:

[0027] Step S1221: Perform position analysis on the first biometric device according to the spatial coordinate system of the identification area, and obtain the device's line-of-sight vector based on the analysis results;

[0028] Step S1222: Analyze the position of sunlight based on the spatial coordinate system of the identified area, and obtain the sunlight vector based on the analysis results.

[0029] Step S1223: Vector the sunlight and the device's line-of-sight vector The backlight angle of the face corresponding to the organism to be identified is obtained by calculation;

[0030] The backlight angle corresponding to the face of the organism to be identified is calculated using the following formula:

[0031]

[0032] Where Mjj is the backlight angle of the face of the organism to be identified. The vector of sunlight. This is the line-of-sight vector of the device.

[0033] Furthermore, step S1221 further includes the following specific steps:

[0034] The horizontal distance between the center point of the first biometric device and the geometric center point of the organism's face is obtained by identifying the spatial coordinate system of the region, thus obtaining the first horizontal distance value.

[0035] The vertical distance between the center point of the first biometric device and the geometric center point of the organism's face is obtained by identifying the spatial coordinate system of the region, thus obtaining the first vertical distance value.

[0036] The angle between the horizontal ray passing through the center point of the first biometric device and the vertical plane directly in front of the face of the organism is obtained by identifying the spatial coordinate system of the region; thus, the horizontal offset angle of the first device is obtained.

[0037] The pitch angle of the first biometric device is obtained by identifying the angle between the horizontal ray passing through the center point of the first biometric device and the horizontal plane where the geometric center point of the organism's face is located by the spatial coordinate system of the identification area.

[0038] The vector pointing from the center point of the first biometric device to the geometric center point of the organism's face is labeled as the device's line-of-sight vector.

[0039] Acquisition device line-of-sight vector The projection length in the XZ plane is: dxz = d·cosε;

[0040] Acquisition device line-of-sight vector The corresponding x-axis component: Cx = dxz·sinδ;

[0041] Acquisition device line-of-sight vector The corresponding z-axis component: Cz = dxz·cosδ;

[0042] Acquisition device line-of-sight vector The corresponding y-axis component: Cy = d·cosε;

[0043] Where d is the first horizontal distance value, h is the first vertical distance value, ε is the first device pitch angle, and δ is the first device horizontal offset angle;

[0044] The device's line-of-sight vector is obtained by calculating the first horizontal distance value d and the first vertical distance value h. Length of the module

[0045] Line of sight through the device The corresponding x-axis component Cx, z-axis component Cz, y-axis component Cy, and device line-of-sight vector Length of the module Line of sight vector of the device The calculation is performed using the following formula:

[0046]

[0047] Furthermore, step S1222 further includes the following specific steps:

[0048] Obtain the solar altitude angle α corresponding to the current moment, obtain the angle of the sun relative to due east at the current moment, and obtain the sun's eastward azimuth angle β;

[0049] Obtain the sun ray vector Projected length in the XZ plane: cosα;

[0050] Obtain the sun ray vector The corresponding x-axis component: Sx = cosα·sinβ;

[0051] Obtain the sun ray vector The corresponding z-axis component: Sz = cosα·cosβ;

[0052] Obtain the sun ray vector The corresponding y-axis component: Sy = sinα;

[0053] Then through the vector of sunlight The sun ray vector can be calculated from the projection length, x-axis component, z-axis component, and y-axis component in the XZ plane. The specific formula is as follows:

[0054]

[0055] Furthermore, step S2 also includes the following specific steps:

[0056] Step S21: Obtain regional backlight analysis data, and obtain the backlight recognition impact coefficient corresponding to each outdoor biometric recognition area based on the regional backlight analysis data;

[0057] Step S22: Obtain the baseline range of the backlight recognition influence coefficient. If the backlight recognition influence coefficient is within the baseline range, the corresponding outdoor biometric area is classified as the first type of biometric area. If the backlight recognition influence coefficient is not within the baseline range, the corresponding outdoor biometric area is classified as the second type of biometric area, thus obtaining the recognition area type classification data.

[0058] Step S23: The first biometric device collects the biometric face of the first type of biometric area, and compares the collected result with the existing biometric face image information database. If the comparison is successful, the biometric recognition is successful; if the comparison is unsuccessful, the biometric recognition fails.

[0059] Furthermore, step S3 also includes the following steps:

[0060] Step S31: Obtain the identification area type classification data, and obtain the second type of biometric identification area based on the identification area type classification data;

[0061] Step S32: Obtain the palmprint image of the organism to be identified in the second type of biometric recognition area to obtain the palmprint image to be identified;

[0062] Step S33: Acquire the access palmprint images stored inside the second biometric device to obtain multiple stored palmprint images, and create a palmprint recognition model based on the multiple stored palmprint images to obtain a feature palmprint recognition model.

[0063] Step S34: Use the feature palmprint recognition model to recognize the palmprint image to be recognized. If the feature palmprint recognition model outputs that the image comparison is successful, the biometric recognition is completed. If the feature palmprint recognition model outputs that the image comparison is unsuccessful, the biometric recognition fails.

[0064] Furthermore, step S33 also includes the following steps:

[0065] A model training image acquisition cycle is created by taking the current time point as the end time point of the cycle. The palmprint images acquired by the second biometric device within the model training image acquisition cycle are then acquired to obtain multiple palmprint training images.

[0066] Randomly select a sample palmprint training image from the multiple acquired palmprint training images, and perform consistency matching between the sample palmprint training image and the stored palmprint images. Based on the matching result, identify and label the sample palmprint training image.

[0067] Specifically as follows:

[0068] Randomly select one sample palmprint image from the multiple acquired stored palmprint images, and use an image recognition algorithm to obtain the common area of ​​the palm in the sample palmprint training image and the sample stored palmprint image. Mark the common area of ​​the palm in the sample palmprint training image as the first common area of ​​the palm, and mark the common area of ​​the palm in the sample stored palmprint image as the second common area of ​​the palm.

[0069] The common areas of the first and second palms are processed into grayscale to obtain the first grayscale image and the second grayscale image respectively.

[0070] The palmprint coverage area in the second grayscale image is marked to obtain the palmprint coverage area of ​​the second image. The palmprint coverage area of ​​the second image is then decomposed into several pixels. The pixel depth corresponding to each pixel is obtained numerically to obtain multiple pixel depth values. The maximum pixel depth value is marked as the first palmprint pixel depth reference value, and the minimum pixel depth value is marked as the second palmprint pixel depth reference value. The first palmprint pixel depth reference value and the second palmprint pixel depth reference value are marked as the palmprint pixel reference interval.

[0071] The first grayscale image is divided into several pixels. The pixel depth corresponding to each pixel is obtained to obtain multiple pixel depth values. Pixels whose pixel depth values ​​are in the palm print pixel reference range are marked as palm print region pixels.

[0072] In the first grayscale image, the image area occupied by the palm print region pixels is marked as the palm print coverage area of ​​the first image;

[0073] The first grayscale image is used to cover the second grayscale image. The overlapping area of ​​the palm print coverage area of ​​the first image and the palm print coverage area of ​​the second image is named the palm print overlapping area. The area of ​​the palm print overlapping area is obtained to get the area value of the palm print overlapping area.

[0074] Obtain the area value of the palmprint coverage area of ​​the second image, obtain the area value of the palmprint region corresponding to the sample stored palmprint image, obtain the area value of the stored palmprint region, calculate the ratio of the area value of the overlapping area of ​​the palmprint in the image to the area value of the stored palmprint region, and obtain the palmprint region similarity between the sample palmprint training image and the sample stored palmprint image.

[0075] The palmprint region similarity between the sample palmprint training image and each stored palmprint image is obtained respectively, and the palmprint region similarity with the largest value is marked as the comprehensive contrast similarity corresponding to the sample palmprint training image.

[0076] Obtain a preset similarity range. If the overall similarity is within the preset similarity range, mark the sample palmprint training image as having passed the comparison. If the overall similarity is not within the preset similarity range, mark the sample palmprint training image as having failed the comparison.

[0077] Each palmprint training image is identified and labeled to obtain palmprint training image labeling data;

[0078] An image recognition model is created using an existing artificial intelligence platform. The model is then trained using palmprint training image labeled data to obtain a feature palmprint recognition model.

[0079] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0080] 1. This invention obtains the backlight recognition influence coefficient by analyzing the backlight intensity and backlight angle of the outdoor biometric recognition area, and selects the recognition method based on the backlight recognition influence coefficient, which can ensure the accuracy of the recognition process.

[0081] 2. This invention creates a feature palmprint recognition model by combining historical recognition images, and performs biometric palmprint recognition through the feature palmprint recognition model. Compared with the traditional recognition method of comparing pre-recorded palmprint images with the palmprint images to be recognized in real time, this invention can effectively improve the efficiency of biometric recognition. Attached Figure Description

[0082] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0083] Figure 1 This is an overall system block diagram of the present invention;

[0084] Figure 2 This is a schematic diagram of the spatial coordinate system of the identification area in this invention;

[0085] Figure 3 This is a schematic diagram of the facial backlight angle in this invention;

[0086] Figure 4 This is a schematic diagram of the palm print coverage area in this invention. Detailed Implementation

[0087] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0088] Example 1

[0089] Please see Figure 1 This invention provides a technical solution: a multimodal biometric identification method based on artificial intelligence, comprising the following specific steps:

[0090] Step S1: Obtain multiple outdoor biometric recognition areas, and perform backlight intensity analysis and backlight angle analysis on each outdoor biometric recognition area. Based on the analysis results, obtain the backlight recognition influence coefficient corresponding to each outdoor biometric recognition area to obtain regional backlight analysis data.

[0091] Step S1 further includes the following specific steps:

[0092] The outdoor areas that need to be biometrically identified are acquired, resulting in multiple outdoor biometric identification areas. A sample biometric identification area is selected from the acquired outdoor biometric identification areas, and two biometric identification devices are set in the sample biometric identification area, which are named the first biometric identification device and the second biometric identification device, respectively.

[0093] It should be noted here that:

[0094] In this application, the first biometric device involved is a facial recognition device, and the second biometric device is a palmprint recognition device;

[0095] Biometric backlighting analysis was performed on the biometric recognition area of ​​the sample, and the biometric backlighting influence coefficient was obtained based on the analysis results.

[0096] Specifically as follows:

[0097] If a biological entity to be identified exists in the sample biometric area, a spatial coordinate system for the identification area is created based on the relative position of the biological entity to be identified and the first biometric device.

[0098] Specifically as follows:

[0099] Please see Figure 2 Within the sample biometric recognition area, the geometric center point of the face corresponding to the organism to be identified is acquired by the image acquisition device to obtain the first position feature point. A horizontal straight line pointing due east is drawn through the first position feature point to obtain the first position feature line. A vertical straight line pointing to the sky is drawn through the first position feature point to obtain the second position feature line. A horizontal straight line pointing due north is drawn through the first position feature point to obtain the third position feature line.

[0100] It should be noted here that:

[0101] In this application, the sample biometric area referred to herein is an outdoor area.

[0102] Mark the first position feature point as the origin, the first position feature line as the x-axis, the second position feature line as the y-axis, and the third position feature line as the z-axis to obtain the spatial coordinate system of the recognition area;

[0103] The backlight angle of the face of the organism to be identified is analyzed based on the spatial coordinate system of the identification area, and the backlight angle of the face of the organism to be identified is obtained based on the analysis results.

[0104] Specifically as follows:

[0105] The position of the first biometric device is analyzed based on the spatial coordinate system of the identification area, and the line-of-sight vector of the device is obtained based on the analysis results.

[0106] Specifically as follows:

[0107] Please see Figure 3 The horizontal distance between the center point of the first biometric device and the geometric center point of the organism's face is obtained by identifying the spatial coordinate system of the region, thus obtaining the first horizontal distance value.

[0108] The vertical distance between the center point of the first biometric device and the geometric center point of the organism's face is obtained by identifying the spatial coordinate system of the region, thus obtaining the first vertical distance value.

[0109] The angle between the horizontal ray passing through the center point of the first biometric device and the vertical plane directly in front of the face of the organism is obtained by identifying the spatial coordinate system of the region; thus, the horizontal offset angle of the first device is obtained.

[0110] It should be noted here that: if the center point of the first biometric device is on the right side of the organism's face, the horizontal offset angle of the first device is set to a positive value; if the center point of the first biometric device is on the left side of the organism's face, the horizontal offset angle of the first device is set to a negative value.

[0111] The pitch angle of the first biometric device is obtained by identifying the angle between the horizontal ray passing through the center point of the first biometric device and the horizontal plane where the geometric center point of the organism's face is located, using the spatial coordinate system of the identification area.

[0112] The vector pointing from the center point of the first biometric device to the geometric center point of the organism's face is labeled as the device's line-of-sight vector.

[0113] Acquisition device line-of-sight vector The projection length in the XZ plane is: dxz = d·cosε;

[0114] Acquisition device line-of-sight vector The corresponding x-axis component: Cx = dxz·sinδ;

[0115] Acquisition device line-of-sight vector The corresponding z-axis component: Cz = dxz·cosδ;

[0116] Acquisition device line-of-sight vector The corresponding y-axis component: Cy = d·cosε;

[0117] Where d is the first horizontal distance value, h is the first vertical distance value, ε is the first device pitch angle, and δ is the first device horizontal offset angle;

[0118] The device's line-of-sight vector is obtained by calculating the first horizontal distance value d and the first vertical distance value h. Length of the module

[0119] Line of sight through the device The corresponding x-axis component Cx, z-axis component Cz, y-axis component Cy, and device line-of-sight vector Length of the module Line of sight vector of the device The calculation is performed using the following formula:

[0120]

[0121] It should be noted here that:

[0122] In practical implementation, if d = 1.2m, h = 0.15m, and δ = 10°,

[0123] Then ε = arctan(0.15 / 1.2 ≈ 7.13°)

[0124] Device line-of-sight vector

[0125] The position of sunlight is analyzed based on the spatial coordinate system of the identified area, and the sunlight vector is obtained based on the analysis results.

[0126] Specifically as follows:

[0127] Obtain the solar altitude angle α corresponding to the current moment, obtain the angle of the sun relative to due east at the current moment, and obtain the sun's eastward azimuth angle β;

[0128] Obtain the sun ray vector Projected length in the XZ plane: cosα;

[0129] Obtain the sun ray vector The corresponding x-axis component: Sx = cosα·sinβ;

[0130] Obtain the sun ray vector The corresponding z-axis component: Sz = cosα·cosβ;

[0131] Obtain the sun ray vector The corresponding y-axis component: Sy = sinα;

[0132] Then through the vector of sunlight The sun ray vector can be calculated from the projection length, x-axis component, z-axis component, and y-axis component in the XZ plane. The specific formula is as follows:

[0133]

[0134] It should be noted here that:

[0135] In practical implementation, if α = 73.2° and β = 190.5°, the solar ray vector can be calculated.

[0136] Vector of sunlight and the device's line-of-sight vector The backlight angle of the face corresponding to the organism to be identified is obtained by calculation;

[0137] The backlight angle corresponding to the face of the organism to be identified is calculated using the following formula:

[0138]

[0139] Where Mjj is the backlight angle of the face of the organism to be identified. The vector of sunlight. The line-of-sight vector of the device;

[0140] The light intensity corresponding to the biometric region of the sample at the current moment is numerically obtained to obtain the real-time light intensity value of the region;

[0141] The backlight angle of the face of the organism to be identified and the real-time light intensity of the area are used to calculate the bio-backlight recognition influence coefficient of the sample bio-recognition area.

[0142] The backlighting impact coefficient corresponding to the biometric recognition area of ​​the sample is calculated using the following formula:

[0143]

[0144] Where Nxk is the biometric backlighting influence coefficient corresponding to the biometric recognition area of ​​the sample, Mjj is the facial backlighting angle corresponding to the biometric being to be identified, Lss is the real-time illumination intensity value of the area, and Lck is the reference illumination intensity value.

[0145] It should be noted here that:

[0146] In this application, the reference light intensity value is a constant. In specific implementation, the reference light intensity value is set to 50,000 lux.

[0147] In practice, when the backlight angle of the face corresponding to the organism to be identified is measured to be 120° and the real-time illumination intensity of the area corresponding to the biometric identification area of ​​the sample is 120,000 lux, the biometric backlight identification influence coefficient corresponding to the biometric identification area of ​​the sample can be calculated to be 0.7059.

[0148] It should be noted here that:

[0149] Existing biometric identification methods often require the subject to approach the device for facial recognition, and a second biometric identification device is only used after multiple attempts fail. This invention uses a biometric backlighting influence coefficient to pre-judge whether the first biometric identification device can successfully identify the subject. Based on the judgment result, it decides whether to use the first or second biometric identification device to identify the subject, which can further reduce identification time and improve identification efficiency.

[0150] Repeat the process of obtaining the backlight recognition influence coefficient corresponding to the biometric recognition area of ​​the sample, and obtain the backlight recognition influence coefficient corresponding to each outdoor biometric recognition area to obtain regional backlight analysis data;

[0151] Step S2: Analyze the recognition method of the outdoor biometric area based on the backlight analysis data of the area, divide the outdoor biometric area into a first type of biometric area and a second type of biometric area based on the analysis results, obtain the recognition area type classification data, and use the first biometric device to perform biometric recognition on the first type of biometric area.

[0152] Step S2 further includes the following specific steps:

[0153] Obtain regional backlight analysis data, and based on the regional backlight analysis data, obtain the backlight recognition impact coefficient corresponding to each outdoor biometric recognition area;

[0154] Obtain the baseline range of the backlight recognition influence coefficient. If the backlight recognition influence coefficient is within the baseline range, the corresponding outdoor biometric area is classified as the first type of biometric area. If the backlight recognition influence coefficient is not within the baseline range, the corresponding outdoor biometric area is classified as the second type of biometric area, thus obtaining the identification area type classification data.

[0155] It should be noted here that:

[0156] The baseline range for the backlight recognition impact coefficient was obtained, as follows:

[0157] A number of historical identification records completed through the first biometric identification device are obtained. The backlight identification influence coefficient corresponding to each historical identification record is obtained. The values ​​of the obtained backlight identification influence coefficients are compared. The backlight identification influence coefficient with the largest value is marked as the first backlight identification influence coefficient benchmark value. The backlight identification influence coefficient with the smallest value is marked as the second backlight identification influence coefficient benchmark value. The numerical interval formed by the first backlight identification influence coefficient benchmark value and the second backlight identification influence coefficient is marked as the backlight identification influence coefficient benchmark interval.

[0158] The first biometric device collects the biometric face of the first type of biometric area, and compares the collected result with the existing biometric face image information database. If the comparison is successful, the biometric recognition is successful; if the comparison is unsuccessful, the biometric recognition fails.

[0159] It should be noted here that:

[0160] The first biometric device involved here can be an outdoor facial recognition access control system. The biometric facial recognition performed by the first biometric device can be human facial recognition. If the recognition is successful, the access control system will pass through; if the recognition is unsuccessful, the access control system will not pass through.

[0161] Step S3: Obtain the second type of biometric region by dividing the data according to the recognition region type, and use the second biometric device to perform biometric palmprint image analysis on the organism to be identified in the second type of biometric region, and perform biometric identification based on the analysis results;

[0162] Step S3 further includes the following steps:

[0163] Obtain identification region type classification data, acquire second type biometric regions based on identification region type classification data, obtain multiple second type biometric regions, and select a feature biometric region from the multiple acquired second type biometric regions;

[0164] Perform biometric palmprint analysis on the organism to be identified that is located in the biometric identification area, and obtain the biometric palmprint image based on the analysis results;

[0165] Specifically as follows:

[0166] Obtain the palmprint image corresponding to the organism to be identified;

[0167] The access palmprint images stored inside the second biometric device are acquired to obtain multiple stored palmprint images. A palmprint recognition model is created based on the multiple stored palmprint images to obtain a feature palmprint recognition model.

[0168] Specifically as follows:

[0169] A model training image acquisition cycle is created by taking the current time point as the end time point of the cycle. The palmprint images acquired by the second biometric device within the model training image acquisition cycle are then acquired to obtain multiple palmprint training images.

[0170] It should be noted here that:

[0171] In this application, the duration of the model training image acquisition cycle is one month. If the time interval between the second biometric device and the current time point is less than or equal to one month, the time point when the first biometric device is put into use is marked as the start time point of the cycle. If the time interval between the second biometric device and the current time point is greater than one month, the time point corresponding to one month before the end time point of the cycle is marked as the start time point of the cycle.

[0172] Randomly select a sample palmprint training image from the multiple acquired palmprint training images, and perform consistency matching between the sample palmprint training image and the stored palmprint images. Based on the matching result, identify and label the sample palmprint training image.

[0173] Specifically as follows:

[0174] Randomly select one sample palmprint image from the multiple acquired stored palmprint images, and use an image recognition algorithm to obtain the common area of ​​the palm in the sample palmprint training image and the sample stored palmprint image. Mark the common area of ​​the palm in the sample palmprint training image as the first common area of ​​the palm, and mark the common area of ​​the palm in the sample stored palmprint image as the second common area of ​​the palm.

[0175] The common areas of the first and second palms are processed into grayscale to obtain the first grayscale image and the second grayscale image respectively.

[0176] It should be noted here that:

[0177] The first grayscale image and the second grayscale image referred to here have the same image parameters, including but not limited to image size, image brightness, and image resolution.

[0178] The palmprint coverage area in the second grayscale image is marked to obtain the palmprint coverage area of ​​the second image. The palmprint coverage area of ​​the second image is then decomposed into several pixels. The pixel depth corresponding to each pixel is obtained numerically to obtain multiple pixel depth values. The maximum pixel depth value is marked as the first palmprint pixel depth reference value, and the minimum pixel depth value is marked as the second palmprint pixel depth reference value. The first palmprint pixel depth reference value and the second palmprint pixel depth reference value are marked as the palmprint pixel reference interval.

[0179] It should be noted here that:

[0180] Please see Figure 4 In this application, the palm print coverage area in the second grayscale image is specifically the palm print image stored in the sample and manually marked during the input process.

[0181] The first grayscale image is divided into several pixels. The pixel depth corresponding to each pixel is obtained to obtain multiple pixel depth values. Pixels whose pixel depth values ​​are in the palm print pixel reference range are marked as palm print region pixels.

[0182] In the first grayscale image, the image area occupied by the palm print region pixels is marked as the palm print coverage area of ​​the first image;

[0183] The first grayscale image is used to cover the second grayscale image. The overlapping area of ​​the palm print coverage area of ​​the first image and the palm print coverage area of ​​the second image is named the palm print overlapping area. The area of ​​the palm print overlapping area is obtained to get the area value of the palm print overlapping area.

[0184] Obtain the area value of the palmprint coverage area of ​​the second image, obtain the area value of the palmprint region corresponding to the sample stored palmprint image, obtain the area value of the stored palmprint region, calculate the ratio of the area value of the overlapping area of ​​the palmprint in the image to the area value of the stored palmprint region, and obtain the palmprint region similarity between the sample palmprint training image and the sample stored palmprint image.

[0185] Repeat the process of obtaining the palmprint region similarity between the sample palmprint training image and the sample stored palmprint image, obtain the palmprint region similarity between the sample palmprint training image and each stored palmprint image respectively, and mark the palmprint region similarity with the largest value as the comprehensive comparison similarity corresponding to the sample palmprint training image.

[0186] Obtain a preset similarity range. If the overall similarity is within the preset similarity range, mark the sample palmprint training image as having passed the comparison. If the overall similarity is not within the preset similarity range, mark the sample palmprint training image as having failed the comparison.

[0187] It should be noted here that:

[0188] In this application, the preset range for similarity qualification is obtained, as follows:

[0189] Several historical palmprint images that have passed comparison are obtained. The similarity of the palmprint region corresponding to each historical palmprint image is obtained. The numerical values ​​of the obtained palmprint region similarity are compared. The palmprint region similarity with the largest value is marked as the upper limit of the preset similarity range, and the palmprint region similarity with the smallest value is marked as the lower limit of the preset similarity range, thus obtaining the preset similarity range.

[0190] Repeat the identification and labeling of the sample palmprint training images, and identify and label each palmprint training image separately to obtain palmprint training image labeling data;

[0191] It should be noted here that:

[0192] The process of classifying historical palmprint images into "failed comparison" and "passed comparison" is achieved by the model automatically extracting pixels, overlaying images, and analyzing area values ​​of the first grayscale image. This process enables the comprehensive comparison similarity of the first grayscale image and classifies historical palmprint images by interval comparison. There is no need to manually label the first grayscale image, thus realizing unsupervised machine learning of the feature palmprint recognition model.

[0193] An image recognition model is created using an existing artificial intelligence platform. The model is then trained using palmprint training image labeled data to obtain a feature palmprint recognition model.

[0194] Specifically as follows:

[0195] The palmprint training image labeled data is divided into a palmprint image training set and a palmprint image test set according to the image training-test ratio;

[0196] It should be noted here that:

[0197] In this application, the image training-to-test ratio is specifically set to 7:3, that is, the ratio of the number of medical palmprint images in the palmprint image training set to the number of palmprint image test sets is 7:3.

[0198] An image recognition model is created using an existing artificial intelligence platform. The image recognition model is trained using a palmprint image training set until each palmprint image in the palmprint image training set is used to train the image recognition model once.

[0199] The image recognition model is tested using a palmprint image test set, and the recognition accuracy is obtained. When the recognition accuracy is greater than or equal to the target recognition accuracy, the image recognition model is trained and a feature palmprint recognition model is obtained. When the recognition accuracy is less than the target recognition accuracy, the image recognition model is trained again using a palmprint image training set until the recognition accuracy is greater than or equal to the target recognition accuracy.

[0200] It should be noted here that:

[0201] In this application, the target recognition accuracy is specifically set to 95%.

[0202] The palmprint recognition model is used to identify the palmprint image to be identified. If the palmprint recognition model outputs that the image comparison is successful, the biometric identification is completed. If the palmprint recognition model outputs that the image comparison is unsuccessful, the biometric identification fails.

[0203] It should be noted here that:

[0204] The second biometric device involved here can be an outdoor palm print recognition access control system. The first biometric device can perform biometric palm print recognition, which can be human palm print recognition. If the recognition is successful, the access control system will pass; if the recognition is unsuccessful, the access control system will not pass.

[0205] It should be noted here that:

[0206] The following benefits exist here:

[0207] Feature palmprint recognition models belong to artificial intelligence models. While recognizing biological palmprints, feature palmprint recognition models can continue to train the model with newly recognized palmprint images, which can achieve unsupervised machine learning. Compared with traditional recording methods, palmprint recognition models can improve the accuracy of recognition.

[0208] Existing palmprint recognition models can only compare real-time acquired palmprint images with pre-existing palmprint images from a second biometric device and output recognition results based on the comparison results. Here, multiple historical images that have been identified and saved within a historical period are used to train the artificial intelligence model. The longer the second biometric device is used, the richer the historical images used for training become, and the stronger the recognition ability of the feature palmprint recognition model becomes, which can greatly improve the recognition accuracy.

[0209] In this application, if a corresponding calculation formula appears, the above calculation formula is a dimensionless calculation. The weighting coefficient, proportional coefficient and other coefficients in the formula are set to quantify each parameter to obtain a result value. The size of the weighting coefficient and proportional coefficient is only required to not affect the proportional relationship between the parameter and the result value.

[0210] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An artificial intelligence-based multi-modal biometric recognition method, characterized by, Comprise: Step S1: obtain a plurality of outdoor biometric identification area, respectively to each outdoor biometric identification area for inverse light intensity analysis and inverse light angle analysis, according to the analysis result obtains each outdoor biometric identification area corresponding to the inverse light identification influence coefficient, get area inverse light analysis data; Step S2: according to the area inverse light analysis data on outdoor biometric identification area identification mode analysis, according to the analysis result will outdoor biometric identification area is divided into first type biometric identification area and second type biometric identification area, get identification area type division data, and using the first biometric identification device to the first type biometric identification area for biometric identification; Step S3: according to the identification area type division data to the second type biometric identification area for acquisition, and using the second biometric identification device to the second type biometric identification area in the biological identification biological palm print image analysis, and according to the analysis result for biometric identification.

2. The multi-modal biometric recognition method based on artificial intelligence according to claim 1, characterized in that, The step S1 also includes the following specific steps: Step S11: to the outdoor area for biometric identification needs to be obtained, get a plurality of outdoor biometric identification area, and in the sample biometric identification area selected in the obtained outdoor biometric identification area, and set the first biometric identification device and the second biometric identification device in the sample biometric identification area; Step S12: inverse light intensity analysis and inverse light angle analysis are carried out on the sample biometric identification area, and the biological inverse light identification influence coefficient is obtained according to the analysis result; Step S13: respectively obtain the inverse light identification influence coefficient corresponding to each outdoor biometric identification area, get area inverse light analysis data.

3. The multi-modal biometric recognition method based on artificial intelligence according to claim 2, characterized in that, The step S12 also includes the following specific steps: Step S121: according to the relative position of the biological body to be identified and the first biometric identification device, the coordinate system is created, and the identification area space coordinate system is obtained; Step S122: according to the identification area space coordinate system, the face inverse light angle analysis is carried out on the biological body to be identified, and the face inverse light angle corresponding to the biological body to be identified is obtained according to the analysis result; Step S123: the light intensity corresponding to the sample biometric identification area at the current time is obtained, and the area real-time light intensity value is obtained; Step S124: the face inverse light angle and the area real-time light intensity value are calculated to obtain the biological inverse light identification influence coefficient corresponding to the sample biometric identification area; The inverse light identification influence coefficient corresponding to the sample biometric identification area is calculated, and the specific formula is as follows: Wherein, Nxk is the biological inverse light identification influence coefficient corresponding to the sample biometric identification area, Mjj is the face inverse light angle, Lss is the area real-time light intensity value, and Lck is the reference light intensity value.

4. The multi-modal biometric recognition method based on artificial intelligence according to claim 3, characterized in that, The step S121 also includes the following specific steps: In the sample biological recognition region, a face geometric center point corresponding to the to-be-recognized organism is acquired by an image acquisition device to obtain a first position feature point, a straight line is drawn through the first position feature point and horizontally to one side to obtain a first position feature straight line, a straight line is drawn through the first position feature point and vertically to obtain a second position feature straight line, and a straight line is drawn through the first position feature point and horizontally to the other side to obtain a third position feature straight line; The first position feature point is marked as a coordinate origin, the first position feature straight line is marked as a coordinate x-axis, the second position feature straight line is marked as a coordinate y-axis, and the third position feature straight line is marked as a coordinate z-axis to obtain a recognition region space coordinate system.

5. The multi-modal biometric recognition method based on artificial intelligence according to claim 3, characterized in that, The step S122 further includes the following specific steps: Step S1221: performing position analysis on the first biological recognition device according to the recognition region space coordinate system, and obtaining a device line-of-sight vector according to an analysis result; Step S1222: According to the identification area space coordinate system, the position of the sunlight is analyzed, and the sunlight vector is obtained according to the analysis result Step S1223: obtaining the sunlight vector and the device line-of-sight vector by calculation, the face inverse-lighting angle corresponding to the to-be-identified organism is obtained; An inverse light included angle of the face of the to-be-recognized organism is calculated, and a specific formula is as follows: Wherein, Mjj is the face inverse light angle corresponding to the to-be-identified biological, is the sunlight vector, is the device line-of-sight vector.

6. The multi-modal biometric recognition method based on artificial intelligence according to claim 5, characterized in that, The step S1221 further includes the following specific steps: A first horizontal distance numerical value between a first biological recognition device center point and a face geometric center point of the organism is obtained through the recognition region space coordinate system to obtain the first horizontal distance numerical value; A first vertical distance numerical value between the first biological recognition device center point and the face geometric center point of the organism is obtained to obtain the first vertical distance numerical value; An angle between a horizontal ray passing through the first biological recognition device center point and a vertical plane in front of the face of the organism is obtained to obtain a first device horizontal offset angle; An angle between the horizontal ray passing through the first biological recognition device center point and a horizontal plane in which the face geometric center point of the organism is located is obtained to obtain a first device pitch angle; A vector from the first biometric device center point to the biometric face geometric center point is labeled as the device line of sight vector Acquisition device line of sight vector In the projection length dxz in the coordinate XZ plane: dxz = d cos ε; Acquisition device line of sight vector The corresponding x-axis component Cx: Cx = dxz sin δ; Acquisition device line of sight vector The corresponding z-axis component Cy: Cz = dxz cos δ; Acquisition device line of sight vector The corresponding y-axis component Cz: Cy= d - cos ε; Wherein, d is the first horizontal distance numerical value, h is the first vertical distance numerical value, ε is the first device pitch angle, and δ is the first device horizontal offset angle. The first horizontal distance value d and the first vertical distance value h are calculated by the device sight line vector length The x-axis component Cx, the z-axis component Cz, and the y-axis component Cy of the device sight line vector corresponding to the device sight line vector length The device sight line vector is calculated by the following formula:

7. The multi-modal biometric recognition method based on artificial intelligence according to claim 5, characterized in that, The step S1222 further includes the following specific steps: A solar elevation angle α corresponding to a current moment is obtained, and an angle of the sun relative to a due east direction at the current moment is obtained to obtain a solar east direction azimuth angle β; Acquiring a sun ray vector The projection length in the coordinate XZ plane: cos α; Acquiring a sun ray vector The corresponding x-axis component Sx: Sx = cos α · sin β; Acquiring a sun ray vector The corresponding z-axis component Sz: Sz = cos a · cos β; Acquiring a sun ray vector The corresponding y-axis component Sy: Sy = sin a; The solar ray vector is then given by The projection length on the coordinate XZ plane, the x-axis component, the z-axis component, and the y-axis component can be calculated from the solar ray vector The specific formula is as follows:

8. The multi-modal biometric recognition method based on artificial intelligence according to claim 1, characterized in that, The step S2 further includes the following specific steps: Step S21: obtaining region backlight analysis data, and obtaining an inverse light recognition influence coefficient corresponding to each outdoor biological recognition region according to the region backlight analysis data; Step S22: obtaining an inverse light recognition influence coefficient reference interval, if the inverse light recognition influence coefficient is in the inverse light recognition influence coefficient reference interval, an outdoor biological recognition region corresponding to the inverse light recognition influence coefficient is divided into a first type biological recognition region, if the inverse light recognition influence coefficient is not in the inverse light recognition influence coefficient reference interval, the outdoor biological recognition region corresponding to the inverse light recognition influence coefficient is divided into a second type biological recognition region, and recognition region type division data is obtained; Step S23: collecting a biological face of the first type biological recognition region, and performing image consistency comparison on a collection result and an existing biological face image information library, if the comparison passes, biological recognition is successful, and if the comparison does not pass, biological recognition fails.

9. The multi-modal biometric recognition method based on artificial intelligence according to claim 1, characterized in that, The step S3 further includes the following steps: Step S31: Obtain identification area type division data, and obtain the second type of biological identification area according to the identification area type division data; Step S32: Obtain a biological palmprint image of the to-be-identified biological in the second type of biological identification area, to obtain a to-be-identified palmprint image; Step S33: Obtain an access palmprint image stored in the second biological identification device, to obtain a plurality of stored palmprint images, create a palmprint identification model according to the plurality of stored palmprint images, and obtain a feature palmprint identification model; Step S34: Identify the to-be-identified palmprint image using the feature palmprint identification model, if the feature palmprint identification model output result is image comparison passing, the biological identification is completed, if the feature palmprint identification model output result is image comparison failing, the biological identification fails.

10. The multi-modal biometric recognition method based on artificial intelligence according to claim 9, characterized in that, In the step S33, the following steps are further included: A model training image acquisition period is created by taking the time point corresponding to the current time as the period end time point, and palmprint images obtained by the second biological identification device in the model training image acquisition period are obtained, to obtain a plurality of palmprint training images; An arbitrary sample palmprint training image is selected from the obtained plurality of palmprint training images, and the sample palmprint training image is matched with the stored palmprint image for consistency, and the sample palmprint training image is identified according to the matching result; Specifically as follows: An arbitrary sample stored palmprint image is selected from the obtained plurality of stored palmprint images, and the palm common area in the sample palmprint training image and the sample stored palmprint image is obtained through an image recognition algorithm, the palm common area in the sample palmprint training image is marked as a first palm common area, and the palm common area in the sample stored palmprint image is marked as a second palm common area; The first palm common area and the second palm common area are respectively subjected to image grayscale processing, to obtain a first grayscale image and a second grayscale image; The palmprint coverage area in the second grayscale image is marked, to obtain a second image palmprint coverage area, and the second image palmprint coverage area is disassembled into a plurality of pixel points, the pixel depth corresponding to each pixel point is respectively obtained, to obtain a plurality of pixel depth values, the maximum pixel depth value is marked as a first palmprint pixel depth reference value, the minimum pixel depth value is marked as a second palmprint pixel depth reference value, and the first palmprint pixel depth reference value and the second palmprint pixel depth reference value are marked as a palmprint pixel point reference interval; The first grayscale image is disassembled into a plurality of pixel points, the pixel depth corresponding to each pixel point is obtained, to obtain a plurality of pixel depth values, and the pixel point whose pixel depth value is in the palmprint pixel point reference interval is marked as a palmprint region pixel point; In the first grayscale image, the image region occupied by the palmprint region pixel point is marked as a first image palmprint coverage area; The first grayscale image is used to cover the second grayscale image, the overlapping area of the first image palm print covering area and the second image palm print covering area is named as an image palm print overlapping area, the area of the image palm print overlapping area is acquired to obtain an image palm print overlapping area value; An area value of the second image palm print covering area is acquired to obtain a palm print area value corresponding to the sample storage palm print image, obtain a storage palm print area value, and calculate the ratio of the image palm print overlapping area value to the storage palm print area value to obtain a palm print area similarity between the sample palm print training image and the sample storage palm print image; The palm print area similarity between the sample palm print training image and each storage palm print image is acquired respectively, and the maximum palm print area similarity is marked as a comprehensive comparison similarity corresponding to the sample palm print training image; A similarity qualified preset interval is acquired, if the comprehensive comparison similarity is in the similarity qualified preset interval, the sample palm print training image is marked as comparison passed, and if the comprehensive comparison similarity is not in the similarity qualified preset interval, the sample palm print training image is marked as comparison failed; Each palm print training image is marked respectively to obtain palm print training image marking data; An image recognition model is created, the image recognition model is trained using the palm print training image marking data, and a feature palm print recognition model is obtained.

Images