WO2021114916A1

WO2021114916A1 - Risk detection method, apparatus and device

Info

Publication number: WO2021114916A1
Application number: PCT/CN2020/124141
Authority: WO
Inventors: 曹佳炯
Original assignee: 支付宝(杭州)信息技术有限公司
Priority date: 2019-12-13
Filing date: 2020-10-27
Publication date: 2021-06-17
Also published as: CN111126216A

Abstract

A risk detection method, apparatus and device. The method comprises: obtaining a biometric feature image combination of a user to be subjected to detection, wherein the biometric feature image combination comprises: a plurality of images obtained by using a multi-ocular camera to photograph, a single time, a specified body part of the user to be subjected to detection (S102); performing consistency detection on the plurality of acquired images by means of a pre-trained first detection model, so as to obtain a first detection result, and performing living body detection on the plurality of acquired images by means of a pre-trained second detection model, so as to obtain a second detection result (S104); and according to the first detection result and the second detection result, determining whether the living body detection of the user to be subjected to detection has a risk of attack (S106).

Description

Risk detection method, device and equipment

Technical field

This document relates to the field of data processing technology, in particular to a risk detection method, device and equipment.

Background technique

As people’s requirements for safety become higher and higher, binocular cameras are widely used in various security scenarios, such as access control, security, etc., to detect two images obtained by a single shot of the binocular camera. Avoid live attacks. Generally, the effective imaging area of a binocular camera is the intersection of the imaging areas of the two cameras, and the area outside the effective imaging area is called a blind zone. And only one camera can collect the image of the blind area, and the other camera cannot collect the image of the blind area. Because the blind zone of the binocular camera has this feature, there is still a risk of live attack.

Summary of the invention

The purpose of one or more embodiments of this specification is to provide a risk detection method, device, and equipment, which combine consistency detection and liveness detection to determine whether the liveness detection of the user to be detected is at risk of being attacked, which not only solves the problem The problem of live attacks in the blind area of multi-camera cameras is solved, and the security is greatly improved.

In order to solve the above technical problems, one or more embodiments of this specification are implemented in this way.

One or more embodiments of this specification provide a risk detection, including: acquiring a combination of biometric images of a user to be detected, wherein the combination of biometric images includes: a multi-lens camera on a designated body of the user to be detected Multiple images obtained from a single shot of the part; the first detection model is used to perform consistency detection on the multiple images to obtain the first detection result; and the second detection model is pre-trained to obtain the first detection result; Perform live detection on the multiple images to obtain a second detection result; according to the first detection result and the second detection result, determine whether the live detection of the user to be detected is at risk of being attacked.

One or more embodiments of this specification provide a risk detection device, including: an acquisition module that acquires a combination of biometric images of a user to be detected, wherein the combination of biometric images includes: A plurality of images obtained by a single shot of a designated body part of the user to be detected; a first training module, which uses a pre-trained first detection model to perform consistency detection on the plurality of images to obtain a first detection result; and , Performing live detection on the multiple images through a pre-trained second detection model to obtain a second detection result; a determining module, which determines whether there is a risk of attack based on the first detection result and the second detection result .

One or more embodiments of this specification provide a risk detection device, including: a processor; and, a memory arranged to store computer-executable instructions, which when executed, cause the processor to: Acquire a biometric image combination of the user to be detected, where the biometric image combination includes: multiple images obtained by a single shot of a designated body part of the user to be detected by a multi-lens camera; A detection model, which performs consistency detection on the multiple images to obtain a first detection result; and, through a pre-trained second detection model, performs a living detection on the multiple images to obtain a second detection result; The first detection result and the second detection result determine whether the live detection of the user to be detected is at risk of being attacked.

One or more embodiments of this specification provide a storage medium for storing computer-executable instructions that, when executed, realize the following process: acquiring a combination of biometric images of a user to be detected, where all The biometric image combination includes: multiple images obtained by a single shot of the designated body part of the user to be detected by a multi-lens camera; and the consistency of the multiple images is detected through a pre-trained first detection model , Obtain a first detection result; and, perform a living detection on the plurality of images through a pre-trained second detection model to obtain a second detection result; determine according to the first detection result and the second detection result Whether the live detection of the user to be detected is at risk of being attacked.

An embodiment of this specification is based on the pre-trained first detection model and the second detection model to detect the biometric image combination of the user to be detected; thus, the consistency detection and the living body detection are combined to determine the living body of the user to be detected Detect whether there is a risk of being attacked; it not only ensures the detection of live attacks in the effective imaging area of the multi-eye camera, but also ensures the detection of live attacks in the blind area of the multi-eye camera, which not only solves the live attack in the blind area of the multi-eye camera Problems, and greatly improved security.

Description of the drawings

In order to more clearly describe the technical solutions in one or more embodiments of this specification, the following will briefly introduce the drawings needed in the description of the embodiments. Obviously, the drawings in the following description are only in this specification. For some of the described embodiments, those of ordinary skill in the art can obtain other drawings based on these drawings without creative work.

FIG. 1 is a schematic diagram of a scenario of a risk detection method provided by one or more embodiments of this specification.

FIG. 2 is a schematic diagram of the first flow of a risk detection method provided by one or more embodiments of this specification.

FIG. 3 is a schematic diagram of the second flow of a risk detection method provided by one or more embodiments of this specification.

FIG. 4 is a detailed diagram of step S100-4 provided by one or more embodiments of this specification.

FIG. 5 is a detailed diagram of step S100-6 provided by one or more embodiments of this specification.

Fig. 6 is a detailed diagram of step A2 provided in one or more embodiments of this specification.

FIG. 7 is a schematic diagram of the module composition of a risk detection device provided by one or more embodiments of this specification.

Fig. 8 is a schematic structural diagram of a risk detection device provided by one or more embodiments of this specification.

Detailed ways

In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the following will be combined with the drawings in one or more embodiments of this specification to compare the technical solutions in one or more embodiments of this specification. The technical solution is described clearly and completely. Obviously, the described embodiments are only a part of the embodiments in this specification, rather than all the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by a person of ordinary skill in the art without creative work shall fall within the protection scope of this document.

Face recognition is not unfamiliar to people. The risk in the process of face recognition lies mainly in live attacks, that is, an attacker uses photos, videos and other imitation real people to brush their faces in an attempt to pass face recognition. The use of binocular cameras to collect face images, and the color images (RGB images) and near-infrared images (IR images) obtained from a single collection for live detection are the current mainstream anti-attack measures. However, because the binocular camera has a blind spot for shooting, and the blind spot has the characteristics that one camera can collect its image, and the other camera cannot collect its image, it has become a breakthrough for the attacker to carry out a live attack, thus making the binocular camera The live body detection is downgraded to the live body detection of the monocular camera, and the success rate of the attack is greatly improved. In this regard, one or more embodiments of this specification provide a risk detection method, device, and equipment, which can detect whether there is a risk of being attacked in the live detection of the user to be detected. Specifically, the first detection model and the first detection model are trained in advance. The second detection model, and the first detection model is used to perform consistency detection on the biometric image combination of the user to be detected to obtain the first detection result; and the second detection model is used to perform the living body detection on the biometric image combination of the user to be detected to obtain the first detection model. Two detection results; thereby determining whether the live detection of the user to be detected is at risk of being attacked according to the first detection result and the second detection result. Among them, the biometric image combination includes: multiple images obtained by a single shot of a designated body part of the user to be detected by a multi-lens camera; designated body parts such as iris, human face, and so on. As a result, the consistency detection and the live detection are combined to solve the problem of live attacks in the blind area of the multi-eye camera, and greatly improve the security.

Figure 1 is a schematic diagram of an application scenario of the risk detection method provided by one or more embodiments of this specification. As shown in Figure 1, the scenario includes: a multi-camera and a risk detection device, where the multi-camera includes multiple cameras; risk The detection device may be a device independent of the multi-lens camera, or may be a device deployed in the multi-lens camera.

Specifically, the multi-lens camera takes a single shot of the designated body part of the user to be detected to obtain a biometric image combination including multiple images; the risk detection device obtains the biometric image combination of the user to be detected, and trains the first detection model in advance, Perform consistency detection on the multiple images to obtain a first detection result; and, perform live detection on the multiple images through a pre-trained second detection model to obtain a second detection result; according to the obtained first detection result and The second detection result determines whether the live detection of the user to be detected is at risk of being attacked. As a result, the consistency detection and the live detection are combined to solve the problem of live attacks in the blind area of the multi-eye camera, and greatly improve the security.

Based on the foregoing application scenario architecture, one or more embodiments of this specification provide a risk detection method; FIG. 2 is a schematic flowchart of a risk detection method provided by one or more embodiments of this specification, and the method in FIG. 2 can be modified by The risk detection device in FIG. 1 is executed. As shown in FIG. 2, the method includes steps S102-S106.

Step S102: Acquire a biometric image combination of the user to be detected, where the biometric image combination includes: multiple images obtained by a single shot of a designated body part of the user to be detected by a multi-camera; wherein the designated body part may be Iris, human face, etc.

Step S104: Perform consistency detection on the acquired multiple images through the pre-trained first detection model to obtain a first detection result; and, through the pre-trained second detection model, perform live detection on the multiple acquired images, Obtain the second test result.

Among them, the first detection model is used to detect the consistency of multiple images included in the combination of biometric images; due to the blind spot attack, the difference between multiple images taken by multiple cameras in the multi-camera is relatively large, so The first detection model is used to perform consistency detection on the multiple images included in the biometric image combination, which can effectively intercept the blind spot attack; wherein the training process of the first detection model is described in detail later. The second detection model is used to perform live detection on multiple images included in the combination of biometric images to intercept live attacks in the effective imaging area; the second detection model can be the same or different from the existing live detection model, due to live detection The training process of the detection model is a well-known technical means to those skilled in the art. Therefore, the training process of the second detection model will not be described in detail in the embodiment of this specification.

Step S106: According to the first detection result and the second detection result, it is determined whether the live detection of the user to be detected is at risk of being attacked.

In one or more embodiments of this specification, based on the pre-trained first detection model and the second detection model, the biometric image combination of the user to be detected is detected; thus, the consistency detection and the living body detection are combined to determine Whether the live detection of the user to be detected is at risk of being attacked; it not only ensures the detection of live attacks in the effective imaging area of the multi-camera camera, but also ensures the detection of live attacks in the blind area of the multi-camera camera. The problem of live attacks in the blind area of the camera, and greatly improves the security.

In order to realize the consistency detection of the multiple images included in the biometric image combination, in one or more embodiments of this specification, as shown in FIG. 3, before step S102, S100-2 to S100-6 are further included.

Step S100-2: Obtain the biometric image combination to be trained collected by the multi-lens camera.

In step S100-4, a second preprocessing is performed on the biometric image combination to be trained to obtain a sample set to be trained.

In step S100-6, the first detection model is trained based on the sample set.

Since the shooting angles of the multiple cameras of the multi-camera camera are different, there are also differences between the multiple images acquired by the multi-camera camera at a time. In order to avoid this difference, the detection performance of the trained first detection model is poor. In one or more embodiments of this specification, step S100-3 is further included before step S100-4.

Step S100-3: Perform calibration processing on multiple cameras in the multi-camera to obtain a conversion matrix;

Specifically, a camera is randomly selected from the multiple cameras included in the multi-camera as the reference camera, and the cameras other than the reference camera are used as the camera to be calibrated; the black and white grid calibration board is used to solve the reference camera and each camera to be calibrated. The internal parameter matrix and the external parameter matrix between the cameras; according to the external parameter matrix, the two images collected by the corresponding reference camera and the camera to be calibrated are converted from the world coordinate system to the camera coordinate system; according to the corresponding internal parameter matrix, Convert the two images from the camera coordinate system to the pixel coordinate system; determine the first coordinate of the same pixel in the two images in the world coordinate system and the second coordinate in the pixel coordinate system; The coordinates are converted into a conversion matrix of the second coordinate, and the determined conversion matrix is used as the conversion matrix between the reference camera and the corresponding camera to be calibrated, and each obtained conversion matrix is used as the conversion matrix of the multi-lens camera. Among them, the process of solving the internal parameter matrix and the external parameter matrix between the reference camera and the camera to be calibrated, and then determining the conversion matrix between the two cameras according to the internal parameter matrix and the external parameter matrix, can refer to the calibration of the existing binocular camera The process will not be described in further detail in the embodiments of this specification.

Further, after selecting the reference camera, it further includes: recording the camera information of the reference camera; and, after obtaining the conversion matrix of the multi-view camera, it further includes: saving the conversion matrix; in order to achieve consistency in the biometric image combination of the user to be detected During detection, determine the reference image in the biometric image combination of the user to be detected according to the recorded camera information of the reference camera, and perform spatial alignment processing on the image to be aligned in the biometric image combination of the user to be detected according to the saved conversion matrix.

After the conversion matrix of the multi-lens camera is obtained, the second preprocessing can be performed on the images in the biometric image combination to be trained according to the conversion matrix, so as to avoid the difference between the images causing the training of the first detection model. Poor detection performance. Specifically, as shown in FIG. 4, step S100-4 includes step S100-4-2 to step S100-4-6.

Step S100-4-2, according to the conversion matrix, perform spatial alignment processing on the images in the biometric image combination to be trained.

Specifically, for each combination of biometric images to be trained, determine the one-to-one correspondence between the multiple images it includes and multiple cameras; take the image corresponding to the reference camera among the multiple images as the reference image, and combine the multiple The images other than the reference image in the image are regarded as the image to be aligned; according to the conversion matrix between the camera corresponding to each image to be aligned and the reference camera, the corresponding image to be aligned is spatially aligned to align it with the reference image. .

Optionally, the image name of each image includes the camera identification of the corresponding camera; correspondingly, determining the one-to-one correspondence between multiple images and multiple cameras includes: determining according to the camera identification included in the image name of each image The one-to-one correspondence between each image and the camera; or, the multiple cameras included in the multi-lens camera respectively establish a transmission channel with the risk detection device in advance, and each camera sends the captured image to the risk detection device through the corresponding transmission channel; corresponding , Determining the one-to-one correspondence between multiple images and multiple cameras includes: obtaining the corresponding camera identification from the corresponding relationship between the channel identification and the camera identification according to the channel identification of the transmission channel through which each image is received; the acquired camera identification The corresponding camera is determined as the camera corresponding to the corresponding image.

In step S100-4-4, the biometric image combination after the spatial alignment processing is used as a sample, and the sample is divided into a positive sample and a negative sample; wherein, the images in the positive sample are consistent, and the images in the negative sample are inconsistent.

Generally, the images in the biometric image combination after spatial alignment are the same. However, when the biometric image combination is collected by the multi-lens camera, there are phenomena such as jitter and partial occlusion due to external force, which makes the biological features after the spatial alignment process The images in the image combination may have slight differences; based on this, the samples are divided into positive samples and negative samples in step S100-4-4, including: determining whether the multiple images included in each sample are consistent, and if they are consistent, the corresponding The sample of is determined as a positive sample; if it is inconsistent, the corresponding sample is determined as a negative sample. Further, considering that when a combination of biometric images is collected by a multi-lens camera, the phenomena such as shaking and partial occlusion due to external forces are accidental phenomena. Therefore, the number of negative samples may not be sufficient for model training; for this, step S100-4- The sample is divided into positive samples and negative samples in 4, which may also include: cross-combining different samples to obtain negative samples; wherein, the negative samples obtained by the cross-combination include one-to-one correspondence with multiple cameras and at different times. Multiple images taken.

For example, the multi-lens camera is a quadruple camera. For ease of description, the four cameras included in the quadruple camera are recorded as camera 1, camera 2, camera 3, and camera 4; and the images captured by each camera are correspondingly recorded as Image 1, image 2, image 3, image 4; image 2 in sample 1 can be exchanged with image 2 in sample 2 to obtain new sample 1 and new sample 2, and combine the new sample 1 and new sample 1 The sample 2 of sample 2 is determined to be a negative sample; you can also replace image 2 in sample 1 with image 2 in sample 2, and replace image 3 in sample 1 with image 3 in sample 4 to obtain a new sample 1, and The new sample 1 is determined to be a negative sample; it should be pointed out that the cross-combination method can be set according to actual needs.

In step S100-4-6, the positive sample and the negative sample are determined as the sample set to be trained.

The second preprocessing is performed on the biometric image combination to be trained and divided into a positive sample and a negative sample to train the first detection model based on the sample set including the positive sample and the negative sample. Specifically, as shown in FIG. 5, step S100-6 includes step S100-6-2 to step S100-6-6.

S100-6-2: Divide the sample set into a training set and a test set; wherein the training set and the test set include the same proportion of positive samples and negative samples.

Specifically, according to a preset ratio of positive samples to negative samples, randomly select a first number of positive samples and a second number of negative samples from the positive samples included in the sample set, and determine the selected samples as the training set; and , Randomly select the third number of positive samples and the fourth number of negative samples from the positive samples included in the sample set, and determine the selected sample as the test set; where, when the first number is the same as the third number, the second The quantity is the same as the fourth quantity; when the first quantity is different from the third quantity, the second quantity is different from the fourth quantity.

S100-6-4: Perform the third preprocessing on the training set and the test set to obtain the target training set and the target test set.

Specifically, the one-to-one correspondence between each image in the training set and the camera is determined; according to the determined correspondence, the image corresponding to each camera is obtained from the training set to obtain the corresponding training subset; according to the images included in each training subset , Determine the conversion parameters of the corresponding camera; According to the conversion parameters, perform preset conversion processing on the images corresponding to the corresponding cameras in the training set and the test set to obtain the target training set and the target test set.

Since the pixel range of the images included in the training set and the test set is large, uncontrollable and unfavorable for training; therefore, in the embodiment of this specification, the conversion parameters of the corresponding camera are determined according to the images included in each training subset, and according to The conversion parameters perform preset conversion processing on the images corresponding to the corresponding cameras in the training set and the test set, so as to convert the pixels of each image to a certain interval to facilitate training. Among them, the conversion parameters and the preset conversion processing can be set according to the needs in actual applications; for example, the conversion parameters include pixel average and pixel variance; the preset conversion processing is the pixel value and conversion parameter of each pixel of each image The pixel mean value in the subtraction is obtained, and the subtraction result is divided by the pixel variance to obtain the division result; the image formed by the division result is determined as the target image; thus, the target training set and the target test set are obtained.

Further, after obtaining the conversion parameters of each camera, it further includes: saving the obtained conversion parameters, so that when the consistency detection is performed on the biometric image combination of the user to be detected, the images in the biometric image combination are saved according to the saved conversion parameters. Perform the first pretreatment.

In S100-6-6, a training operation is performed based on the target training set and the target test set to obtain the first detection model.

Specifically, each image included in the target training set is input to the twin network for binary classification training to obtain the initial detection model; the target test set is input to the obtained initial detection model to obtain the detection result, and if the detection result meets the preset condition, then Determine the initial detection model as the first detection model; if the detection result does not meet the preset conditions, retrain the model based on the target training set to obtain a new initial detection model, and perform detection on the new initial detection model based on the test set. Until the first detection model is obtained. The detection result represents the probability that each image in the biometric image combination is consistent, and the preset condition is, for example, that the probability is greater than the preset probability.

By training the first detection model, when the biometric image combination of the user to be detected is obtained, the consistency detection of each image included in the biometric image combination is performed, thereby combining the live detection result to determine whether there is an attack in the live detection risk.

As mentioned above, the shooting angles of the multiple cameras of the multi-lens camera are different. In order to avoid incorrect detection results due to different shooting angles, in one or more embodiments of this specification, the combination of biometric images of the user to be detected When the included multiple images are tested for consistency, first preprocessing is performed on the multiple images. Specifically, in step S104, the multiple images are tested for consistency through the pre-trained first detection model, including Steps A2 and A4.

Step A2: Perform first preprocessing on multiple images to obtain processed images; specifically, as shown in FIG. 6, in one or more embodiments of this specification, step A2 includes step A2-2 to step A2-10.

Step A2-2: Determine the one-to-one correspondence between the multiple images and the multiple cameras in the multi-camera.

Among them, the implementation process of step A2-2 can be referred to the aforementioned related description, and the repetitive parts will not be repeated here.

Step A2-4: Obtain the conversion matrix of the multi-camera and the conversion parameters corresponding to each of the multiple cameras; where the conversion matrix is obtained by calibrating the multiple cameras before training the first detection model; the conversion parameters are When training the first detection model, it is obtained by performing the third preprocessing on the sample set to be trained.

Specifically, the multi-eye camera is recorded as an N-eye camera, where N is greater than or equal to 2; the saved N-1 conversion matrices of the N-eye camera and the conversion parameters corresponding to each of the N cameras are acquired.

Step A2-6: Perform spatial alignment processing on multiple images according to the conversion matrix.

Specifically, the reference camera is determined according to the saved camera information of the reference camera; the image corresponding to the reference camera among the multiple images is used as the reference image, and the images other than the reference image among the multiple images are used as the image to be aligned; The conversion matrix between the camera corresponding to each image to be aligned and the reference camera performs a spatial alignment operation on the corresponding image to be aligned to make it spatially aligned with the corresponding reference image.

Step A2-8: Perform preset conversion processing on the corresponding spatially aligned image according to the conversion parameters.

Among them, the implementation process of step A2-8 can refer to the aforementioned related description, and the repetitive parts will not be repeated here.

Step A2-10: Determine the image processed by the preset conversion as a processed image.

Step A4: Input the processed image to the first detection model to perform consistency detection on the processed image based on the first detection model.

Since the consistency detection based on the first detection model has better detection capabilities for live attacks in the blind area of the multi-eye camera, the live detection based on the second detection model has better detection capabilities for live attacks in the effective imaging area of the multi-eye camera. Detection capability; therefore, in the embodiments of this specification, according to the first detection result obtained based on the first detection model and the second detection result obtained based on the second detection model, it is determined whether the live detection of the user to be detected is at risk of being attacked. Optionally, in one or more embodiments of this specification, step S106 includes: performing a weighted calculation on the first detection result and the second detection result according to a preset weighting coefficient to obtain the calculation result; if the calculation result is greater than the preset weighting coefficient The first threshold determines that the live detection of the user to be detected is not at risk of being attacked; if the calculation result is not greater than the preset first threshold, it is determined that the live detection of the user to be detected is at risk of being attacked.

Among them, the weighting coefficient and the first threshold can be set according to needs in actual applications; thus, the first detection result and the second detection result are weighted and fused to determine whether the live detection of the user to be detected is attacked The risk of improving safety.

Alternatively, in one or more embodiments of the present application, step S106 includes: if the first detection result is greater than a preset second threshold, and the second detection result is greater than a preset third threshold, determining the user’s There is no risk of attack for live detection.

Specifically, it is determined whether the first detection result is greater than a preset second threshold, and if it is not greater than the second threshold, it is determined that the multiple images included in the biometric image combination of the user to be detected are inconsistent, and the live detection of the user to be detected is affected. The risk of attack; if it is greater than the second threshold, it is determined that the multiple images included in the combination of biometric images of the user to be detected are consistent, and it is determined whether the second detection result is greater than the preset third threshold, and if it is greater than the third threshold, it is determined The live detection of the user to be detected is not at risk of being attacked. If it is not greater than the third threshold, it is determined that the live detection of the user to be detected is at risk of being attacked. Or, it is determined whether the second detection result is greater than a preset third threshold, if it is not greater than the third threshold, it is determined that the live detection of the user to be detected is at risk of being attacked; if it is greater than the third threshold, it is determined whether the first detection result is It is greater than the preset second threshold. If it is not greater than the second threshold, it is determined that the multiple images included in the biometric image combination of the user to be detected are inconsistent, and the live detection of the user to be detected is at risk of being attacked; if it is greater than the second threshold , It is determined that there is no risk of attack for the live detection of the user to be detected.

As a result, it is determined whether the live detection of the user to be detected is at risk of being attacked according to the first detection result and the second detection result in turn, which improves security.

Based on any of the foregoing embodiments, optionally, the multi-lens camera is a binocular camera.

Corresponding to the risk detection methods described in FIGS. 2 to 6 above, based on the same technical concept, one or more embodiments of this specification also provide a risk detection device. Figure 7 is a schematic diagram of the module composition of a risk detection device provided by one or more embodiments of this specification. The device is used to execute the risk detection method described in Figures 2 to 6. As shown in Figure 7, the device includes: Module 201, which acquires a biometric image combination of a user to be detected, wherein the biometric image combination includes: multiple images obtained by a single shot of a designated body part of the user to be detected by a multi-lens camera; A training module 202, which detects the consistency of the multiple images through a pre-trained first detection model, to obtain a first detection result; and, through a pre-trained second detection model, performs a consistency check on the multiple images Living body detection obtains a second detection result; a determination module 203, which determines whether there is an attack risk based on the first detection result and the second detection result.

The risk detection device provided by one or more embodiments of this specification detects a combination of biometric images of the user to be detected based on the pre-trained first detection model and second detection model; thus, the consistency detection is compared with the live detection Combine it to determine whether the live detection of the user to be detected is at risk of being attacked; it not only ensures the detection of live attacks in the effective imaging area of the multi-camera camera, but also ensures the detection of live attacks in the blind area of the multi-camera camera. Solve the problem of live attacks in the blind area of the multi-eye camera, and greatly improve the security.

Optionally, the first training module 202 performs first preprocessing on the plurality of images to obtain processed images; and, inputs the processed images to the first detection model to be based on the The first detection model performs consistency detection on the processed image.

Optionally, the first training module 202 determines the one-to-one correspondence between the multiple images and the multiple cameras in the multi-camera; and obtains the conversion matrix of the multi-camera and the multi-camera. The conversion parameter corresponding to each camera in the two cameras; wherein, the conversion matrix is obtained by calibrating the multiple cameras before training the first detection model; the conversion parameter is the training of the first detection model In the model, the sample set to be trained is obtained by performing the third preprocessing; according to the conversion matrix, the multiple images are spatially aligned; according to the conversion parameters, the corresponding spatially aligned images are processed Perform a preset conversion process; determine the image after the preset conversion process as a processed image.

Optionally, the determining module 203 performs a weighted calculation on the first detection result and the second detection result according to a preset weighting coefficient to obtain a calculation result; if the calculation result is greater than the preset first detection result Threshold, it is determined that the live detection of the user to be detected is not at risk of being attacked; if the calculation result is not greater than the preset first threshold, it is determined that the live detection of the user to be detected is at risk of being attacked.

Optionally, the determining module 203, if it is determined that the first detection result is greater than a preset second threshold, and the second detection result is greater than a preset third threshold, determine the living body of the user to be detected There is no risk of being attacked by detection.

Optionally, the device further includes: a second training module; the second training module obtains a combination of biometric images to be trained collected by the multi-lens camera; and combines the biometric images to be trained Performing second preprocessing to obtain a sample set to be trained; training the first detection model based on the sample set.

Optionally, the device further includes: a calibration module; the calibration module performs calibration processing on multiple cameras in the multi-camera to obtain a conversion matrix; the second training module, based on the conversion matrix, Spatial alignment processing is performed on the images in the biometric image combination to be trained; and, the biometric image combination after the spatial alignment processing is used as a sample, and the sample is divided into a positive sample and a negative sample; wherein, Each image in the positive sample is consistent, and each image in the negative sample is not consistent; the positive sample and the negative sample are determined as a sample set to be trained.

Optionally, the second training module divides the sample set into a training set and a test set; wherein the training set and the test set include the same proportion of the positive sample and the negative sample; And, performing a third preprocessing on the training set and the test set to obtain a target training set and a target test set; performing a training operation based on the target training set and the target test set to obtain a first detection model.

Optionally, the second training module determines a one-to-one correspondence between each image in the training set and the camera; and, according to the determined correspondence, acquires the location of each camera from the training set. Corresponding images to obtain the corresponding training subset; according to the images included in each training subset, determine the conversion parameters of the corresponding camera; according to the conversion parameters, the training set and the test set corresponding to the camera Perform preset conversion processing on the image to obtain the target training set and target test set.

It should be noted that the embodiment of the risk detection device in this specification and the embodiment of the risk detection method in this specification are based on the same inventive concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the corresponding risk detection method mentioned above. I won't repeat it here.

Further, corresponding to the risk detection method described in FIGS. 2 to 6 above, based on the same technical concept, one or more embodiments of this specification also provide a risk detection device, which is used to execute the above risk detection method. 8 is a schematic structural diagram of a risk detection device provided by one or more embodiments of this specification.

As shown in FIG. 8, the risk detection equipment may have relatively large differences due to different configurations or performances, and may include one or more processors 301 and a memory 302. The memory 302 may store one or more storage application programs or data. Among them, the memory 302 may be short-term storage or persistent storage. The application program stored in the memory 302 may include one or more modules (not shown in the figure), and each module may include a series of computer-executable instructions in the risk detection device. Furthermore, the processor 301 may be configured to communicate with the memory 302, and execute a series of computer-executable instructions in the memory 302 on the risk detection device. The risk detection device may also include one or more power supplies 303, one or more wired or wireless network interfaces 304, one or more input and output interfaces 305, one or more keyboards 306, and so on.

In a specific embodiment, the risk detection device includes a memory and one or more programs, wherein one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each Each module may include a series of computer-executable instructions for the risk detection equipment, and the one or more programs configured to be executed by one or more processors include computer-executable instructions for performing the following: The biometric image combination, wherein the biometric image combination includes: multiple images obtained by a single shot of a designated body part of the user to be detected by a multi-lens camera; Performing consistency detection on the multiple images to obtain a first detection result; and, performing live detection on the multiple images through a pre-trained second detection model to obtain a second detection result; according to the first detection result and The second detection result determines whether the live detection of the user to be detected is at risk of being attacked.

In one or more embodiments of this specification, the user can directly interact with the virtual robot in a group containing virtual robots, and enter the link address of the public opinion to be queried to achieve the purpose of accurate public opinion query; without the user having to extract the query to be queried in advance The keywords of public opinion, and there is no need for the first client to list and display multiple public opinions for users to choose, which not only improves query efficiency and user experience, but also the link address is more certain, ensuring the accuracy of query results.

Optionally, when the computer-executable instructions are executed, the checking the consistency of the plurality of images through the pre-trained first detection model includes: performing first preprocessing on the plurality of images to obtain Processed image; input the processed image to the first detection model to perform consistency detection on the processed image based on the first detection model.

Optionally, when the computer-executable instructions are executed, the performing the first preprocessing on the multiple images to obtain the processed images includes: determining that the multiple images and multiple of the multi-cameras One-to-one correspondence between cameras; acquiring the conversion matrix of the multi-lens camera and the conversion parameter corresponding to each camera in the plurality of cameras; wherein, the conversion matrix is before the training of the first detection model, the A plurality of cameras are obtained by calibration processing; the conversion parameter is obtained by performing a third preprocessing on the sample set to be trained when the first detection model is trained; and the plurality of images are spatially processed according to the conversion matrix Alignment processing; performing preset conversion processing on the corresponding spatially aligned image according to the conversion parameter; determining the image after the preset conversion processing as a processed image.

Optionally, when the computer-executable instructions are executed, the determining whether the live detection of the user to be detected is at risk of being attacked according to the first detection result and the second detection result includes: Set the weighting coefficient to perform weighting calculation on the first detection result and the second detection result to obtain the calculation result; if the calculation result is greater than the preset first threshold, determine the live detection of the user to be detected There is no risk of being attacked; if the calculation result is not greater than the preset first threshold, it is determined that the live detection of the user to be detected is at risk of being attacked.

Optionally, when the computer-executable instructions are executed, the determining whether the live detection of the user to be detected is at risk of being attacked according to the first detection result and the second detection result includes: If the first detection result is greater than the preset second threshold, and the second detection result is greater than the preset third threshold, it is determined that there is no risk of attack in the live detection of the user to be detected.

Optionally, when the computer-executable instructions are executed, before the obtaining the combination of biometric images of the user to be detected, the method further includes: obtaining the combination of biometric images to be trained collected by the multi-camera; The trained biometric image combination is subjected to second preprocessing to obtain a sample set to be trained; and the first detection model is trained based on the sample set.

Optionally, when the computer-executable instructions are executed, before the second preprocessing is performed on the biometric image combination to be trained, the method further includes: calibrating multiple cameras in the multi-camera, Obtain a conversion matrix; the second preprocessing is performed on the combination of biometric images to be trained to obtain a sample set to be trained, including: according to the conversion matrix, the images in the combination of biometric images to be trained Perform spatial alignment processing; take the biometric image combination after the spatial alignment processing as a sample, and divide the sample into a positive sample and a negative sample; wherein each image in the positive sample is consistent, and the negative sample The images of are inconsistent; the positive sample and the negative sample are determined as the sample set to be trained.

Optionally, when the computer-executable instructions are executed, the training of the first detection model based on the sample set includes: dividing the sample set into a training set and a test set; wherein the training set and The test set includes the same ratio of the positive sample and the negative sample; the third preprocessing is performed on the training set and the test set to obtain a target training set and a target test set; based on the target training set Perform a training operation with the target test set to obtain a first detection model.

Optionally, when the computer-executable instructions are executed, the third preprocessing of the training set and the test set to obtain the target training set and the target test set includes: determining each image in the training set One-to-one correspondence with the camera; according to the determined correspondence, the image corresponding to each camera is obtained from the training set to obtain the corresponding training subset; according to the training subset included in each Image, determine the conversion parameters of the corresponding camera; according to the conversion parameters, perform preset conversion processing on the images corresponding to the corresponding cameras in the training set and the test set to obtain a target training set and a target test set.

Further, corresponding to the risk detection methods shown in FIGS. 2 to 6 above, based on the same technical concept, one or more embodiments of this specification also provide a storage medium for storing computer-executable instructions, a specific In an embodiment, the storage medium may be a U disk, an optical disk, a hard disk, etc., and the computer-executable instructions stored in the storage medium can realize the following process when executed by the processor: obtaining a combination of biometric images of the user to be detected, where: The biometric image combination includes: multiple images obtained by a single shot of the designated body part of the user to be detected by a multi-lens camera; and the consistency of the multiple images is performed through a pre-trained first detection model Detection to obtain a first detection result; and, by using a pre-trained second detection model, perform a living detection on the plurality of images to obtain a second detection result; according to the first detection result and the second detection result, It is determined whether the live detection of the user to be detected is at risk of being attacked.

In one or more embodiments of this specification, based on the pre-trained first detection model and second detection model, the biometric image combination of the user to be detected is detected; thus, the consistency detection and the living body detection are combined to determine Whether the live detection of the user to be detected is at risk of being attacked; it not only ensures the detection of live attacks in the effective imaging area of the multi-camera camera, but also ensures the detection of live attacks in the blind area of the multi-camera camera. The problem of live attacks in the blind area of the camera, and greatly improves the security.

Optionally, when the computer-executable instructions stored in the storage medium are executed by the processor, the first detection model that is pre-trained to perform consistency detection on the multiple images includes: Perform a first preprocessing to obtain a processed image; input the processed image to the first detection model to perform consistency detection on the processed image based on the first detection model.

Optionally, when the computer-executable instructions stored in the storage medium are executed by the processor, the performing the first preprocessing on the plurality of images to obtain the processed images includes: determining the relationship between the plurality of images and the One-to-one correspondence of the multiple cameras in the multi-camera; acquiring the conversion matrix of the multi-camera and the conversion parameters corresponding to each of the multiple cameras; wherein the conversion matrix is for training the first Before detecting the model, it is obtained by calibrating the multiple cameras; the conversion parameter is obtained by performing the third preprocessing on the sample set to be trained when training the first detection model; according to the conversion matrix, Performing spatial alignment processing on the plurality of images; performing preset conversion processing on the corresponding spatial alignment processed image according to the conversion parameter; determining the image after the preset conversion processing as a processed image.

Optionally, when the computer executable instructions stored in the storage medium are executed by the processor, the first detection result and the second detection result are used to determine whether the live detection of the user to be detected is attacked The risk includes: performing a weighted calculation on the first detection result and the second detection result according to a preset weighting coefficient to obtain a calculation result; if the calculation result is greater than a preset first threshold, determining all The live detection of the user to be detected is not at risk of being attacked; if the calculation result is not greater than the preset first threshold, it is determined that the live detection of the user to be detected is at risk of being attacked.

Optionally, when the computer executable instructions stored in the storage medium are executed by the processor, the first detection result and the second detection result are used to determine whether the live detection of the user to be detected is attacked The risk includes: if the first detection result is greater than a preset second threshold, and the second detection result is greater than a preset third threshold, determining that the live detection of the user to be detected is not attacked risk.

Optionally, when the computer-executable instructions stored in the storage medium are executed by the processor, before the obtaining the combination of biometric images of the user to be detected, the method further includes: obtaining biometrics to be trained collected by the multi-lens camera Image combination; performing a second preprocessing on the biometric image combination to be trained to obtain a sample set to be trained; training the first detection model based on the sample set.

Optionally, when the computer-executable instructions stored in the storage medium are executed by the processor, before the second preprocessing is performed on the biometric image combination to be trained, the method further includes: A plurality of cameras are calibrated to obtain a conversion matrix; the second preprocessing of the combination of biometric images to be trained to obtain a sample set to be trained includes: according to the conversion matrix, performing the second preprocessing on the to-be-trained The images in the biometric image combination are subjected to spatial alignment processing; the biometric image combination after the spatial alignment processing is used as a sample, and the sample is divided into a positive sample and a negative sample; wherein each image in the positive sample Consistent, the images in the negative sample are not consistent; the positive sample and the negative sample are determined as the sample set to be trained.

Optionally, when the computer-executable instructions stored in the storage medium are executed by the processor, the training of the first detection model based on the sample set includes: dividing the sample set into a training set and a test set; Wherein, the training set and the test set include the same proportion of the positive sample and the negative sample; the third preprocessing is performed on the training set and the test set to obtain a target training set and a target test set ; Perform a training operation based on the target training set and the target test set to obtain a first detection model.

Optionally, when the computer executable instructions stored in the storage medium are executed by the processor, the third preprocessing is performed on the training set and the test set to obtain the target training set and the target test set, including: determining The one-to-one correspondence between each image in the training set and the camera; according to the determined correspondence, the image corresponding to each camera is obtained from the training set to obtain a corresponding training subset; For the images included in the training subset, the conversion parameters of the corresponding cameras are determined; according to the conversion parameters, preset conversion processing is performed on the images corresponding to the corresponding cameras in the training set and the test set to obtain the target training set and the target Test set.

When the computer-executable instructions stored in the storage medium provided by one or more embodiments of the present specification are executed by the processor, they are detected based on the pre-trained first detection model and the second detection model, based on the combination of biometric images of the user to be detected; Therefore, the consistency detection and the live detection are combined to determine whether the live detection of the user to be detected is at risk of being attacked; it not only ensures the detection of live attacks in the effective imaging area of the multi-eye camera, but also ensures the detection of multiple The detection of live attacks in the blind area of the eye camera not only solves the problem of live attacks in the blind area of the multi-eye camera, but also greatly improves the security.

It should be noted that the embodiment of the storage medium in this specification and the embodiment of the risk detection method in this specification are based on the same inventive concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the corresponding risk detection method mentioned above, and repeat it. I won't repeat it here.

The foregoing describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than in the embodiments and still achieve desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown in order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

In the 1930s, the improvement of a technology can be clearly distinguished between hardware improvements (for example, improvements in circuit structures such as diodes, transistors, switches, etc.) and software improvements (improvements in method flow). However, with the development of technology, the improvement of many methods and processes of today can be regarded as a direct improvement of the hardware circuit structure. Designers almost always get the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be realized by the hardware entity module. For example, a programmable logic device (Programmable Logic Device, PLD) (such as a Field Programmable Gate Array (Field Programmable Gate Array, FPGA)) is such an integrated circuit whose logic function is determined by the user's programming of the device. It is programmed by the designer to "integrate" a digital system on a piece of PLD, without requiring chip manufacturers to design and manufacture dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly realized with "logic compiler" software, which is similar to the software compiler used in program development and writing, but before compilation The original code must also be written in a specific programming language, which is called Hardware Description Language (HDL), and there is not only one type of HDL, but many types, such as ABEL (Advanced Boolean Expression Language) , AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description), etc., currently most commonly used It is VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. It should also be clear to those skilled in the art that just a little bit of logic programming of the method flow in the above-mentioned hardware description languages and programming into an integrated circuit can easily obtain the hardware circuit that implements the logic method flow.

The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program codes (such as software or firmware) executable by the (micro)processor. , Logic gates, switches, application specific integrated circuits (ASICs), programmable logic controllers and embedded microcontrollers. Examples of controllers include but are not limited to the following microcontrollers: ARC625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicon Labs C8051F320, the memory controller can also be implemented as part of the memory control logic. Those skilled in the art also know that, in addition to implementing the controller in a purely computer-readable program code manner, it is entirely possible to program the method steps to make the controller use logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded The same function can be realized in the form of a microcontroller, etc. Therefore, such a controller can be regarded as a hardware component, and the devices included in it for realizing various functions can also be regarded as a structure within the hardware component. Or even, the device for realizing various functions can be regarded as both a software module for realizing the method and a structure within a hardware component.

The systems, devices, modules, or units illustrated in the above embodiments may be specifically implemented by computer chips or entities, or implemented by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cell phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or Any combination of these devices.

For the convenience of description, when describing the above device, the functions are divided into various units and described separately. Of course, when implementing the embodiments of this specification, the functions of each unit can be implemented in the same one or more software and/or hardware.

Those skilled in the art should understand that one or more embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, one or more embodiments of this specification may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

This specification is described with reference to flowcharts and/or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and/or block in the flowchart and/or block diagram, and the combination of processes and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing equipment to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing equipment are used to generate It is a device that realizes the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing equipment to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction device. The device implements the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be loaded on a computer or other programmable data processing equipment, so that a series of operation steps are executed on the computer or other programmable equipment to produce computer-implemented processing, so as to execute on the computer or other programmable equipment. The instructions provide steps for implementing the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

In a typical configuration, the computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include non-permanent memory in computer readable media, random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of computer readable media.

Computer-readable media include permanent and non-permanent, removable and non-removable media, and information storage can be realized by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, CD-ROM, digital versatile disc (DVD) or other optical storage, Magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media can be used to store information that can be accessed by computing devices. According to the definition in this article, computer-readable media does not include transitory media, such as modulated data signals and carrier waves.

It should also be noted that the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or equipment including a series of elements includes not only those elements, but also Other elements that are not explicitly listed, or also include elements inherent to such processes, methods, commodities, or equipment. If there are no more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, method, commodity, or equipment that includes the element.

One or more embodiments of this specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of this specification can also be practiced in distributed computing environments. In these distributed computing environments, tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, as for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment.

The above descriptions are only examples of this document, and are not intended to limit this document. For those skilled in the art, this document can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this document shall be included in the scope of the claims of this document.

Claims

A risk detection method, including:

Acquiring a biometric image combination of the user to be detected, where the biometric image combination includes: multiple images obtained by a single shot of a designated body part of the user to be detected by a multi-lens camera;

Perform consistency detection on the multiple images through a pre-trained first detection model to obtain a first detection result; and, through a pre-trained second detection model, perform a living detection on the multiple images to obtain a second Test results;

According to the first detection result and the second detection result, it is determined whether the live detection of the to-be-detected user is at risk of being attacked.
The method according to claim 1, wherein said detecting the consistency of the plurality of images through the pre-trained first detection model comprises:

Performing first preprocessing on the plurality of images to obtain processed images;

The processed image is input to the first detection model to perform consistency detection on the processed image based on the first detection model.
The method according to claim 2, wherein the performing the first preprocessing on the plurality of images to obtain the processed images includes:

Determining a one-to-one correspondence between the multiple images and the multiple cameras in the multi-camera;

Acquire the conversion matrix of the multi-lens camera and the conversion parameter corresponding to each camera of the multiple cameras; wherein, the conversion matrix is obtained by calibrating the multiple cameras before training the first detection model得; The conversion parameter is obtained by performing the third preprocessing on the sample set to be trained when training the first detection model;

Performing spatial alignment processing on the plurality of images according to the conversion matrix;

Performing preset conversion processing on the corresponding spatially aligned image according to the conversion parameter;

The image after the preset conversion processing is determined as a processed image.
The method according to claim 1, wherein the determining whether the live detection of the user to be detected is at risk of being attacked according to the first detection result and the second detection result comprises:

Performing a weighted calculation on the first detection result and the second detection result according to a preset weighting coefficient to obtain a calculation result;

If the calculation result is greater than the preset first threshold, it is determined that there is no risk of being attacked in the live detection of the user to be detected;

If the calculation result is not greater than the preset first threshold, it is determined that the live detection of the to-be-detected user is at risk of being attacked.
The method according to claim 1, wherein the determining whether the live detection of the user to be detected is at risk of being attacked according to the first detection result and the second detection result comprises:

If the first detection result is greater than the preset second threshold and the second detection result is greater than the preset third threshold, it is determined that there is no risk of attack in the live detection of the user to be detected.
The method according to claim 1, wherein the multi-eye camera is a binocular camera.
The method according to any one of claims 1-6, before said acquiring a combination of biometric images of the user to be detected, further comprising:

Acquiring a combination of biometric images to be trained collected by the multi-lens camera;

Performing second preprocessing on the biometric image combination to be trained to obtain a sample set to be trained;

Training the first detection model based on the sample set.
The method according to claim 7, before the second preprocessing is performed on the biometric image combination to be trained, further comprising:

Performing calibration processing on multiple cameras in the multi-camera to obtain a conversion matrix;

The second preprocessing performed on the combination of biometric images to be trained to obtain a sample set to be trained includes:

Performing spatial alignment processing on the images in the biometric image combination to be trained according to the conversion matrix;

Taking the biometric image combination after the spatial alignment processing as a sample, and dividing the sample into a positive sample and a negative sample; wherein each image in the positive sample is consistent, and each image in the negative sample is not consistent;

The positive sample and the negative sample are determined as a sample set to be trained.
The method according to claim 8, wherein the training of the first detection model based on the sample set comprises:

Dividing the sample set into a training set and a test set; wherein the training set and the test set include the same proportion of the positive sample and the negative sample;

Performing a third preprocessing on the training set and the test set to obtain a target training set and a target test set;

Perform a training operation based on the target training set and the target test set to obtain a first detection model.
The method according to claim 9, wherein said performing third preprocessing on said training set and said test set to obtain a target training set and a target test set comprises:

Determine a one-to-one correspondence between each image in the training set and the camera;

Acquiring images corresponding to each of the cameras from the training set according to the determined correspondence relationship to obtain a corresponding training subset;

Determine the conversion parameters of the corresponding camera according to the images included in each of the training subsets;

According to the conversion parameters, preset conversion processing is performed on the images corresponding to the corresponding cameras in the training set and the test set to obtain a target training set and a target test set.
A risk detection device, including:

An acquiring module that acquires a combination of biometric images of the user to be detected, wherein the combination of biometric images includes: multiple images obtained by a single shot of a designated body part of the user to be detected by a multi-lens camera;

A first training module, which detects the consistency of the plurality of images through a pre-trained first detection model to obtain a first detection result; and, uses a pre-trained second detection model to perform the consistency detection on the plurality of images Live body test, get the second test result;

A determining module, which determines whether there is an attack risk based on the first detection result and the second detection result.
The device according to claim 11,

The first training module performs first preprocessing on the multiple images to obtain processed images; and,

The processed image is input to the first detection model to perform consistency detection on the processed image based on the first detection model.
The device according to claim 11 or 12, further comprising: a training module;

The training module obtains the biometric image combination to be trained collected by the multi-camera; and,

Performing second preprocessing on the biometric image combination to be trained to obtain a sample set to be trained;

Training the first detection model based on the sample set.
The device according to claim 13, further comprising: a calibration module;

The calibration module, which performs calibration processing on a plurality of cameras in the multi-lens camera to obtain a conversion matrix;

The training module performs spatial alignment processing on the images in the biometric image combination to be trained according to the conversion matrix; and,

Taking the biometric image combination after the spatial alignment processing as a sample, and dividing the sample into a positive sample and a negative sample; wherein each image in the positive sample is consistent, and each image in the negative sample is not consistent;

The positive sample and the negative sample are determined as a sample set to be trained.
A risk detection equipment, including:

Processor; and,

A memory arranged to store computer-executable instructions which, when executed, cause the processor to:

Acquiring a biometric image combination of the user to be detected, where the biometric image combination includes: multiple images obtained by a single shot of a designated body part of the user to be detected by a multi-lens camera;

Perform consistency detection on the multiple images through a pre-trained first detection model to obtain a first detection result; and, through a pre-trained second detection model, perform a living detection on the multiple images to obtain a second Test results;

According to the first detection result and the second detection result, it is determined whether the live detection of the to-be-detected user is at risk of being attacked.
A storage medium used to store computer-executable instructions that, when executed, implement the following processes:

Acquiring a biometric image combination of the user to be detected, where the biometric image combination includes: multiple images obtained by a single shot of a designated body part of the user to be detected by a multi-lens camera;

Perform consistency detection on the multiple images through a pre-trained first detection model to obtain a first detection result; and, through a pre-trained second detection model, perform a living detection on the multiple images to obtain a second Test results;

According to the first detection result and the second detection result, it is determined whether the live detection of the to-be-detected user is at risk of being attacked.