WO2020215283A1

WO2020215283A1 - Facial recognition method, processing chip and electronic device

Info

Publication number: WO2020215283A1
Application number: PCT/CN2019/084338
Authority: WO
Inventors: 吴勇辉; 刘川熙; 詹洁琼
Original assignee: 深圳市汇顶科技股份有限公司
Priority date: 2019-04-25
Filing date: 2019-04-25
Publication date: 2020-10-29
Also published as: CN110199296A

Abstract

Disclosed are a facial recognition method, a processing chip and an electronic device. The facial recognition method comprises: respectively collecting, by means of a plurality of cameras, at least one two-dimensional image including a human face, and selecting any two cameras to respectively collect a first two-dimensional image including a human face and a second two-dimensional image including a human face (101); extracting two-dimensional coordinates of a plurality of feature points of the human face in the first two-dimensional image, and extracting two-dimensional coordinates of a plurality of feature points in the second two-dimensional image (102); obtaining, according to preset parameters of any two cameras and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image, a three-dimensional mesh model corresponding to the human face (103); and determining whether the human face is a three-dimensional human face of a preset user according to the three-dimensional mesh model and a three-dimensional human face model of the preset user (104). The method can prevent a two-dimensional picture from being verified by means of facial recognition, thereby improving the security of facial recognition.

Description

Face recognition method, processing chip and electronic equipment

Technical field

This application relates to the field of image recognition technology, and in particular to a face recognition method, processing chip and electronic equipment.

Background technique

With the rapid development of smart terminals represented by mobile phones, the use of facial recognition technology on various smart terminals has become more and more widespread. Users can complete operations such as unlocking the terminal and making payments through facial recognition.

The inventor found that the prior art has at least the following problems: At present, when the terminal performs face recognition, it is generally unable to recognize whether the current face image is from the user's photo or the user himself, so that the user's image may pass the face recognition authentication. , Security is low.

Summary of the invention

The purpose of some embodiments of this application is to provide a face recognition method, processing chip and electronic device that can recognize whether a human face is a preset user, and determine whether the human face is a three-dimensional face, so as to prevent a two-dimensional picture from passing face recognition Verification improves the security of face recognition.

An embodiment of the application provides a face recognition method, including: acquiring at least one two-dimensional image containing a human face through a plurality of cameras, and selecting the first two-dimensional image containing the human face separately acquired by any two cameras And a second two-dimensional image containing a human face; extracting the two-dimensional coordinates of multiple feature points of the human face in the first two-dimensional image, and extracting the two-dimensional coordinates of multiple feature points in the second two-dimensional image; The preset parameters of any two cameras, as well as the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image, obtain the three-dimensional mesh model corresponding to the face; according to the three-dimensional mesh model and the preset user’s The three-dimensional face model determines whether the face is a preset user's three-dimensional face.

The embodiment of the application also provides a processing chip, the processing chip is connected to at least one memory, the processing chip is connected to multiple cameras, the memory stores instructions that can be executed by the processing chip, and the instructions are executed by the processing chip so that the processing chip can Perform the aforementioned face recognition method.

An embodiment of the present application also provides an electronic device, including the aforementioned processing chip, at least one memory, and multiple cameras.

For the prior art, the embodiments of the present application collect at least one two-dimensional image containing a human face through multiple cameras. Each two-dimensional image includes multiple feature points of the human face. Any two cameras are selected respectively. Collect the first two-dimensional image containing the human face and the second two-dimensional image containing the human face, and extract the two-dimensional coordinates of multiple feature points on the first two-dimensional image and the second two-dimensional image; The preset parameters of the camera, and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image, to obtain the three-dimensional mesh model corresponding to the face, so that the three-dimensional mesh model can be compared with the preset user's three-dimensional The face model is compared to determine whether the face is the three-dimensional face of the preset user, that is, the three-dimensional mesh model corresponding to the two-dimensional image of the face is used for face recognition, which can identify whether the face belongs to the preset user, and It is determined whether the face is a three-dimensional face, and the two-dimensional picture is prevented from passing the face recognition verification, which improves the security of face recognition.

For example, after determining whether the face is the preset user's three-dimensional face according to the three-dimensional mesh model and the preset user's three-dimensional face model, it also includes: determining whether the face is the preset user's face according to the multi-frame three-dimensional mesh model The user's live face. In this embodiment, performing live detection on the face can verify whether the face is the live face of the preset user, and avoid the face model of the preset user from being verified by face recognition.

For example, according to the multi-frame 3D mesh model, determining whether the human face is a preset user's live human face includes: selecting two three-dimensional mesh models, and calculating the corresponding mesh of the multiple meshes on the two-frame three-dimensional mesh model. If at least one of the deformation variables is greater than or equal to the preset deformation threshold, it is determined that the face is the live face of the preset user. This embodiment provides a specific implementation manner for determining whether a human face is a preset user's live human face according to a three-dimensional mesh model of a human face in consecutive multiple frames.

For example, before obtaining the three-dimensional mesh model corresponding to the face according to the preset parameters of any two cameras and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image, the method further includes: A two-dimensional image and a two-dimensional face image of a preset user to determine whether the two-dimensional image is from the preset user; if it is determined that the face is from the preset user, enter the preset parameters according to any two cameras, and the second The step of obtaining the three-dimensional mesh model corresponding to the human face by the two-dimensional coordinates of each feature point on the one-two-dimensional image and the second two-dimensional image. In this embodiment, the use of a two-dimensional image for face recognition is added, and the face recognition speed of the two-dimensional image is faster, thereby improving the face recognition speed of non-default users.

For example, according to the three-dimensional mesh model and the preset user's three-dimensional face model, determining whether the face is the preset user's three-dimensional face includes: calculating the relative position parameters between the meshes on the three-dimensional mesh model; determining the three-dimensional Whether each relative position parameter on the mesh model matches the corresponding relative position parameter on the preset 3D face model; if the relative position parameters on the 3D mesh model are relative to the preset 3D face model The location parameters are matched, and it is determined that the human face is a three-dimensional human face of the preset user. This embodiment provides a specific implementation method for determining whether the human face is the three-dimensional face of the preset user according to the three-dimensional mesh model and the three-dimensional face model of the preset user.

For example, according to the preset parameters of any two cameras and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image, the three-dimensional mesh model corresponding to the face is obtained, including: according to any two cameras Calculate the three-dimensional coordinates of each feature point according to the preset parameters of the first two-dimensional image and the two-dimensional coordinates of each feature point on the second two-dimensional image; construct a three-dimensional network corresponding to the face image according to the three-dimensional coordinates of each feature point Lattice model. This embodiment provides a specific implementation of obtaining a three-dimensional mesh model corresponding to a human face according to the preset parameters of any two cameras and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image the way.

For example, before calculating the three-dimensional coordinates of each feature point according to the preset parameters of any two cameras and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image, it also includes: using a window matching function And preset the user’s two-dimensional face image to correct the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image; according to the preset parameters of any two cameras, and the first two-dimensional image and The two-dimensional coordinates of each feature point on the second two-dimensional image are calculated, and the three-dimensional coordinates of each feature point are calculated, specifically: according to the preset parameters of any two cameras, and the corrected first and second two-dimensional images On the two-dimensional coordinates of each feature point, obtain the three-dimensional coordinates of each feature point. In this embodiment, the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image are corrected to make the acquired three-dimensional coordinates of each feature point more accurate, thereby obtaining a more accurate three-dimensional mesh model , Improve the accuracy of face recognition.

For example, using a window matching function and a preset user’s two-dimensional face image to correct the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image includes: Set the window area including the feature points on the one-dimensional image and the second two-dimensional image; search for the feature on the two-dimensional face image in the window area of the first two-dimensional image and the window area of the second two-dimensional image respectively The point matched by the point is recorded as the exact feature point; the two-dimensional coordinates of the precise feature point on the first two-dimensional image are taken as the two-dimensional coordinates of the feature point on the corrected first two-dimensional image, and the second two-dimensional The two-dimensional coordinates of the precise feature point on the image are the two-dimensional coordinates of the feature point on the second two-dimensional image after correction. This embodiment provides a specific implementation method for correcting the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image by using a window matching function and a preset user's two-dimensional face image.

For example, searching for points matching the feature points on the two-dimensional face image in the window area of the first two-dimensional image and the window area of the second two-dimensional image respectively includes: Set the gray value of the two-dimensional face image and the gray value of each point in the window area to obtain the matching degree between each point in the window area and the feature point on the two-dimensional face image; The matching degree between each point of and the feature point on the two-dimensional face image is obtained, and the point in the window area that matches the feature point on the two-dimensional face image is obtained. This embodiment provides a specific implementation of searching for points matching the feature points on the two-dimensional face image in the window area of the first two-dimensional image and the window area of the second two-dimensional image respectively.

For example, determining whether the face is the preset user's 3D face based on the 3D mesh model and the preset user's 3D face model includes: correcting the face pose corresponding to the 3D mesh model to the preset pose; The three-dimensional mesh model of the preset posture and the three-dimensional face model of the preset user determine whether the face is the three-dimensional face of the preset user. In this embodiment, the face posture corresponding to the three-dimensional mesh model is corrected to a preset posture, so as to facilitate comparison with the three-dimensional face model and improve the accuracy of face recognition.

Description of the drawings

One or more embodiments are exemplified by the pictures in the corresponding drawings. These exemplified descriptions do not constitute a limitation on the embodiments. Elements with the same reference numbers in the drawings are represented as similar elements. Unless otherwise stated, the figures in the attached drawings do not constitute a limitation of scale.

Fig. 1 is a specific flowchart of a face recognition method according to the first embodiment of the present application;

Fig. 2 is a specific flowchart of a face recognition method according to a second embodiment of the present application;

Fig. 3 is a specific flowchart of a face recognition method according to a third embodiment of the present application;

Fig. 4 is a specific flow chart of a face recognition method according to a fourth embodiment of the present application;

Fig. 5 is a specific flow chart of the face recognition method according to the fifth embodiment of the present application;

Fig. 6 is a specific flow chart of a face recognition method according to a sixth embodiment of the present application;

Fig. 7 is a specific flowchart of a face recognition method according to a seventh embodiment of the present application;

Fig. 8 is a schematic diagram of a window area in a seventh embodiment according to the present application;

Fig. 9 is a specific flow chart of a face recognition method according to the eighth embodiment of the present application;

Fig. 10 is a schematic diagram of the processing chip in the ninth embodiment and the main control device in the tenth embodiment according to the present application.

Specific embodiment

In order to make the objectives, technical solutions, and advantages of the present application clearer, some embodiments of the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the application, and are not used to limit the application.

The first embodiment of the present application relates to a face recognition method, which is applied to an electronic device containing multiple cameras, such as a mobile phone, a tablet computer, and the like.

The specific process of the face recognition method in this embodiment is shown in FIG. 1.

Step 101: Collect at least one two-dimensional image containing a human face through a plurality of cameras, and select a first two-dimensional image containing a human face and a second two-dimensional image containing a human face separately collected by any two cameras.

Specifically, in the process of using the electronic device, the user needs to perform face recognition verification in scenes such as unlocking the screen and payment verification. At this time, at least one two-dimensional image containing a human face is collected through multiple cameras of the electronic device. Multiple cameras can be collected simultaneously, that is, multiple face images can be collected simultaneously by multiple cameras at the same time. When selecting any two of the multiple cameras, the two two-dimensional images collected by any two cameras are the first two dimensions. Image and the second two-dimensional image; when two two-dimensional images collected by any two of the multiple cameras are selected, two two-dimensional images collected by any two cameras at the same time can be selected.

Step 102: Extract the two-dimensional coordinates of multiple feature points of the human face in the first two-dimensional image, and extract the two-dimensional coordinates of the multiple feature points in the second two-dimensional image.

Specifically, the electronic device presets which positions on the face are used as feature points, and the face is set with multiple feature points corresponding to multiple locations, such as eyes, corners of eyes, nose, corners of mouth, etc.; , The two-dimensional coordinates of multiple feature points of the face can be extracted on the first two-dimensional image and the second two-dimensional image respectively. Taking the first two-dimensional image as an example, a corner point of the first two-dimensional face image is used as the origin to establish a coordinate system, and the pixel coordinates of each feature point on the first two-dimensional face image are obtained as the pixel coordinates of each feature point. Two-dimensional coordinates.

Step 103: According to the preset parameters of any two cameras and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image, a three-dimensional mesh model corresponding to the human face is obtained.

Specifically, the preset parameters of the two cameras include the internal parameters of the two cameras and the external parameters between the two cameras. The internal parameters include the focal length of the camera itself, imaging parameters, and the center of the principal point. The external parameters include the calibration of the two cameras. The coordinate relationship of the, including the translation matrix and the rotation matrix, etc., according to the preset parameters of the two cameras, and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image, it can be obtained that the face corresponds to a frame of three-dimensional Grid model.

Step 104, according to the three-dimensional mesh model and the three-dimensional face model of the preset user, determine whether the face is the three-dimensional face of the preset user.

Specifically, the electronic device stores the three-dimensional face model entered by the preset user, compares the three-dimensional mesh model corresponding to the face with the three-dimensional face model of the preset user, and judges the three-dimensional mesh model and the three-dimensional face model Whether it matches, when the two match, it means that the face is the preset user's three-dimensional face; when the two do not match, it means that the face is not the preset user's three-dimensional face, that is, the face may be non-predicted. Set the user's face and/or two-dimensional picture.

Compared with the prior art, this embodiment collects at least one two-dimensional image containing a human face through multiple cameras, and each two-dimensional image includes multiple feature points of the human face, and any two cameras are selected to collect them separately. The first two-dimensional image containing the human face and the second two-dimensional image containing the human face, and the two-dimensional coordinates of multiple feature points are extracted from the first two-dimensional image and the second two-dimensional image; then according to the two cameras The preset parameters of the user’s face, and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image, to obtain the three-dimensional mesh model corresponding to the face, so that the three-dimensional mesh model can be combined with the preset user’s three-dimensional human The face model is compared to determine whether the face is a three-dimensional face of the preset user. That is, the three-dimensional mesh model corresponding to the two-dimensional image of the face is used for face recognition. It can identify whether the face belongs to the preset user and determine Whether the face is a three-dimensional face, to prevent the two-dimensional picture from passing the face recognition verification, which improves the security of face recognition.

The second embodiment of the present application relates to a face recognition method. This embodiment is an improvement on the basis of the first embodiment. The main improvement lies in the addition of live detection of a human face.

The specific process of the face recognition method in this embodiment is shown in FIG. 2.

Among them, step 201 to step 204 are roughly the same as step 101 to step 104, which will not be repeated here. The main difference lies in the addition of step 205, which is specifically as follows:

Step 205: Determine whether the human face is a live human face of the preset user according to the multi-frame three-dimensional mesh model.

Specifically, two face images at the same time collected by any two cameras can be selected to obtain a three-dimensional mesh model, so that multiple frames of three-dimensional mesh models at multiple times can be obtained, and the The comparison of multi-frame 3D mesh models can determine whether the human face is the live face of the preset user; because live human faces generally have subtle expression changes, such as blinking, changes in the corners of the mouth, etc., when the multi-frame 3D mesh model is completely When they are consistent, it means that the two-dimensional image may be derived from the face model of the preset user; when the multiple three-dimensional mesh models are not completely consistent, it means that the face is the live face of the preset user.

Compared with the first embodiment, this embodiment performs live detection on the face to verify whether the face is the live face of the preset user, and avoids the face model of the preset user from being verified by the face recognition.

The third embodiment of the present application relates to a face recognition method. This embodiment is a refinement on the basis of the second embodiment. The main refinement is that it provides a method for determining a person based on a multi-frame three-dimensional mesh model. Whether the face is a specific implementation method of the preset user's live face.

The specific process of the face recognition method in this embodiment is shown in FIG. 3.

Among them, step 301 to step 304 are roughly the same as step 201 to step 204, and will not be repeated here. The main difference is that step 305 includes the following sub-steps:

In sub-step 3051, two frames of three-dimensional grid models are selected, and the deformation between corresponding grids in the plurality of grids on the two frames of three-dimensional grid models is calculated.

Specifically, the three-dimensional mesh model divides the human face into multiple meshes, and selects two frames of three-dimensional mesh models arbitrarily from the multi-frame three-dimensional mesh models, and calculates the correspondence between the two frames of the three-dimensional mesh model. The deformation between the two grids, specifically, the deformation between the two grids can be calculated according to the coordinates of the vertices of the corresponding two grids, so that multiple deformations can be obtained.

In sub-step 3052, if at least one deformation variable is greater than or equal to a preset deformation threshold, it is determined that the human face is a living human face of the predetermined user.

Specifically, it is determined whether there is at least one deformation greater than or equal to the preset deformation threshold. If there is at least one deformation greater than or equal to the preset deformation threshold, it means that the two three-dimensional mesh models are not completely consistent. It is the live face of the preset user. If all the deformation variables are less than the preset deformation threshold, it means that the two three-dimensional mesh models are completely consistent, and the face may be the preset user's face model.

It should be noted that in this embodiment, two frames of face mesh models are selected for comparison, but it is not limited to this, and multiple frames of face mesh models can be compared separately to determine whether the face is a preset The result of the user's live face is more accurate.

Compared with the second embodiment, this embodiment provides a specific implementation manner for determining whether a human face is a preset user's live human face according to a multi-frame three-dimensional mesh model.

The fourth embodiment of the present application relates to a face recognition method. This embodiment is an improvement on the basis of the first embodiment. The main improvement lies in the addition of a preset user person based on a two-dimensional face image and at least one frame. The step of comparing face images to determine whether the face is a preset user.

The specific process of the face recognition method in this embodiment is shown in FIG. 4.

Among them, step 401 and step 402 are roughly the same as step 101 and step 102, and step 404 and step 405 are roughly the same as step 103 and step 104, and will not be repeated here. The main difference is that step 403 is added, which is specifically as follows:

Step 403: Determine whether the two-dimensional image comes from the preset user based on at least one two-dimensional image and the two-dimensional face image of the preset user. If yes, go to step 403; if no, then directly end.

Specifically, at least one of the two-dimensional images collected by multiple cameras containing human faces is selected, and the two-dimensional image of the two-dimensional image is calculated according to the two-dimensional coordinates of each feature point in the two-dimensional image. The feature value is, for example, a feature vector. A preset user’s two-dimensional face image is pre-recorded in the electronic device, so that the feature value of the two-dimensional face image can be calculated and stored, and the feature value of the two-dimensional image is compared with The feature values of the two-dimensional face images of the preset user are compared. When the two are compared, it is determined that the two-dimensional image comes from the preset user, and step 404 is entered according to the preset parameters of any two cameras, and the first and second The two-dimensional coordinates of each feature point on the two-dimensional image and the second two-dimensional image are obtained to obtain a three-dimensional mesh model corresponding to the human face; when the two do not match, it is determined that the two-dimensional image does not originate from a preset user and ends directly. Wherein, multiple two-dimensional images collected by multiple cameras may also be subjected to the above-mentioned judgment whether it is derived from a preset user.

Compared with the first embodiment, this embodiment increases the use of two-dimensional images for face recognition, and the speed of face recognition of the two-dimensional images is faster, thereby increasing the speed of face recognition for non-predetermined users. This embodiment can also be used as an improvement on the basis of the second or third embodiment, and the same technical effect can be achieved.

The fifth embodiment of the present application relates to a face recognition method. This embodiment is a refinement on the basis of the first embodiment. The main refinement is that it provides a method based on a three-dimensional mesh model and a preset user The three-dimensional face model determines whether the face is a specific realization method of the preset user's three-dimensional face.

The specific process of the face recognition method in this embodiment is shown in FIG. 5.

Among them, step 501 to step 503 are roughly the same as step 101 to step 103, and will not be repeated here. The main difference is that step 504 includes the following sub-steps:

In sub-step 5041, the relative position parameters between the grids on the three-dimensional grid model are calculated.

Specifically, the three-dimensional mesh model includes N meshes, where N is a positive integer greater than 1. Taking N=5 as an example, the three-dimensional mesh model includes 5 meshes, namely mesh 1, mesh 2. , Grid 3, Grid 4, Grid 5, calculate the relative position parameters between each grid on the three-dimensional grid model, that is, calculate the relative position parameters between any two grids on the three-dimensional grid model. In an example, only the relative position parameters between two adjacent grids on the three-dimensional grid model can be calculated. The relative position parameters are, for example, the angle between the two grids, the distance between the center points, and the like.

In sub-step 5042, if each relative position parameter on the three-dimensional mesh model matches the corresponding relative position parameter on the preset three-dimensional face model, it is determined that the face is the preset user's three-dimensional face.

Specifically, the electronic device prestores the preset relative position parameters between the grids on the user's three-dimensional face model, and determines that each relative position parameter on the three-dimensional grid model corresponds to the preset three-dimensional face model. Whether the relative position parameters are matched, for example, the relative position parameter between grid 1 and grid 2 on the three-dimensional mesh model is A, and the relative position parameter between grid 1 and grid 2 on the three-dimensional face model is B , If the absolute value of the difference between A and B is within the preset range, it is determined that the relative position parameter A matches the relative position parameter B. When the relative position parameters on the three-dimensional mesh model correspond to those on the three-dimensional face model When the relative position parameters are matched, it is determined that the human face is a three-dimensional human face of the preset user. When there is at least one relative position parameter on the 3D mesh model that does not match the corresponding relative position parameter on the 3D face model, it is determined that the face is not the 3D face of the preset user, and the face image may be of a non-preset user Human faces and/or two-dimensional pictures.

In this embodiment, a matching number threshold can also be set. When the number of relative position parameters that do not match the three-dimensional face model on the three-dimensional mesh model is less than the matching number threshold, it is determined that the face is the default user’s three-dimensional person Face: When the number of relative position parameters that do not match on the three-dimensional mesh model and the three-dimensional face model is greater than or equal to the matching number threshold, it is determined that the face image is not the preset user's three-dimensional face.

Compared with the first embodiment, this embodiment determines whether the human face is a specific implementation manner of the predetermined user's three-dimensional face according to the three-dimensional mesh model and the preset user's three-dimensional face model. This embodiment can also be used as a refinement on the basis of the second to fourth embodiments, and the same technical effect can be achieved.

The sixth embodiment of this application relates to a face recognition method. This embodiment is a refinement on the basis of the first embodiment. The main refinement is that it provides a preset parameter based on any two cameras, And the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image to obtain a specific implementation of the three-dimensional mesh model corresponding to the human face.

The specific process of the face recognition method in this embodiment is shown in FIG. 6.

Among them, step 601, step 602, and step 604 are roughly the same as step 101, step 102, and step 104, and will not be repeated here. The main difference is that step 603 includes the following sub-steps:

In sub-step 6031, the three-dimensional coordinates of each feature point are calculated according to the preset parameters of any two cameras and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image.

Specifically, the preset parameters of the two cameras include the internal parameters of the two cameras and the external parameters between the two cameras. The internal parameters include the focal length of the camera itself, imaging parameters, and the center of the principal point. The external parameters include the calibration of the two cameras. The coordinate relationship of each feature point, including translation matrix and rotation matrix, etc., according to the preset parameters of the two cameras, and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image, the three-dimensional of each feature point can be calculated coordinate.

Sub-step 6032, according to the three-dimensional coordinates of each feature point, construct a three-dimensional mesh model corresponding to the face image.

Specifically, according to the three-dimensional coordinates of each feature point, a point set in space can be obtained, and then a three-dimensional mesh model can be constructed. The mesh can be a triangular mesh, a quadrilateral mesh, etc. The grid is a triangular mesh as an example. The Delaunay method is used to triangulate, so that a triangle can be formed from the nearest three points, and the sides of multiple triangles are not intersected. After the division, a polygonal shell is formed. The triangular mesh model corresponding to the face image is presented.

Compared with the first embodiment, this embodiment provides a way to obtain a human face based on the preset parameters of any two cameras and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image. The specific implementation of the corresponding 3D mesh model. This embodiment can also be used as a refinement on the basis of the second to fifth embodiments, and the same technical effect can be achieved.

The seventh embodiment of the present application relates to a face recognition method. This embodiment is an improvement on the basis of the sixth embodiment. The main improvement lies in the addition of the first two-dimensional image and the second two-dimensional image. Correction of two-dimensional coordinates of feature points.

The specific process of the face recognition method of this embodiment is shown in FIG. 7.

Among them, step 601, step 602, and step 604 are roughly the same as step 101, step 102, and step 104, and will not be repeated here. The main difference is that a sub-step 7031 is added to step 703, which is specifically as follows:

In sub-step 7031, the two-dimensional coordinates of the feature points on the first two-dimensional image and the second two-dimensional image are corrected by using the window matching function and the preset two-dimensional face image of the user.

Specifically, taking the six feature points set on the face and the correction of the two-dimensional coordinates of each feature point on the first two-dimensional image as an example, the six feature points are feature point 1, feature point 2, feature Point 3, Feature Point 4, Feature Point 5, Feature Point 6, when correcting the two-dimensional coordinates of each feature point on the first two-dimensional image, take feature point 1 as an example, set on the first two-dimensional image Given the window area that includes feature point 1, at least one pixel point can be expanded with feature point 1 as the center, so that a window area can be obtained. For example, by expanding one pixel outside, the window area shown in Figure 8 can be obtained. The window area includes 9 points (including feature point 1), and the 9 points in this window area are searched for a point that matches the feature point 1 on the preset user's two-dimensional face image, and it is recorded as accurate feature point 1; Then, the two-dimensional coordinates of the precise feature point 1 are used as the two-dimensional coordinates of the feature point 1 on the corrected first two-dimensional image. The correction methods for other feature points are similar to the above; for each on the second two-dimensional image The method of correcting the two-dimensional coordinates of the feature point is similar to the above, so I won’t repeat it here.

In this embodiment, the first two-dimensional image window area is searched for points that match the feature points on the preset user's two-dimensional face image as an example. The search method is: take feature point 1 as an example, and take the first A window with feature point 1 on a two-dimensional face image, take the window area shown in Figure 8 as an example, take a certain size window in turn with each point in the window as the center; preset the user’s two-dimensional face image Feature point 1 is a window with the same size as the center, and the matching degree between the gray levels of the two windows is calculated, so the matching degree D1 to D9 is obtained. The matching degree can be similarity, ratio, etc., according to each point in the window area and the preset The matching degree between the feature point 1 on the user's two-dimensional face image can be obtained in the window area that matches the feature point 1 on the preset user's two-dimensional face image, and the matching degree from D1 to D9 can be obtained The point with the largest value is used as the point in the window area that matches the feature point 1 on the preset user's two-dimensional face image.

In sub-step 7032, the three-dimensional coordinates of each feature point are obtained according to the preset parameters of any two cameras and the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image after correction.

Specifically, it is roughly the same as the sub-step 6031, and the main difference is that the three-dimensional coordinates of each feature point are calculated by using the two-dimensional coordinates of each feature point on the corrected first two-dimensional image and the second two-dimensional image.

Sub-step 7033, according to the three-dimensional coordinates of each feature point, construct a three-dimensional mesh model corresponding to the face image.

Specifically, it is roughly the same as the sub-step 6032, and will not be repeated here.

Compared with the sixth embodiment, this embodiment corrects the two-dimensional coordinates of each feature point on the first two-dimensional image and the second two-dimensional image to make the acquired three-dimensional coordinates of each feature point more accurate, thereby obtaining A more accurate three-dimensional mesh model improves the accuracy of face recognition.

The eighth embodiment of the present application relates to a face recognition method. This embodiment is an improvement on the basis of the first embodiment, and the main improvement lies in: correcting the face pose corresponding to the acquired three-dimensional mesh model.

The specific process of the face recognition method in this embodiment is shown in FIG. 9.

Among them, step 801 to step 803 are substantially the same as step 101 to step 103, which will not be repeated here. The main difference is that step 804 includes the following sub-steps:

In sub-step 8041, the face pose corresponding to the three-dimensional mesh model is corrected to a preset pose.

Specifically, when performing face recognition, the current user may not aim at the camera with a preset pose. Therefore, the face pose corresponding to the obtained 3D mesh model is not the preset pose. The corresponding face pose is corrected to obtain a three-dimensional mesh model of the preset pose. Among them, the preset pose is consistent with the face pose corresponding to the preset 3D face model entered by the user. For example, if the user enters the 3D face model facing the camera, the preset pose is facing the camera. At this time, the 3D mesh model It is consistent with the posture of the three-dimensional face model, making it easier to compare.

In sub-step 8042, according to the three-dimensional mesh model of the preset posture and the three-dimensional face model of the preset user, it is determined whether the face is the three-dimensional face of the preset user.

Specifically, the electronic device stores the three-dimensional face model entered by the preset user, compares the three-dimensional mesh model of the preset pose with the three-dimensional face model of the preset user, and judges the three-dimensional mesh model and the three-dimensional face model Whether it matches, when the two match, it means that the face is the preset user's three-dimensional face; when the two do not match, it means that the face is not the preset user's three-dimensional face, that is, the face may be non-predicted. Set the user's face and/or two-dimensional picture.

Compared with the first embodiment, this embodiment corrects the face posture corresponding to the three-dimensional mesh model to a preset posture, so as to facilitate comparison with the three-dimensional face model and improve the accuracy of face recognition. This embodiment can also be used as an improvement on the basis of the second to seventh embodiments, and the same technical effect can be achieved.

The ninth embodiment of the present application relates to a processing chip. As shown in FIG. 10, the processing chip 110 is connected to at least one memory 111, and the processing chip 110 is also connected to a plurality of cameras 112. The memory 111 stores data that can be executed by the processing chip 110. Instructions, the instructions are executed by the processing chip 110, so that the processing chip 110 can execute the foregoing method embodiments.

The tenth embodiment of the present application relates to a main control device. Please refer to FIG. 10. The main control device includes the aforementioned processing chip 110, at least one memory 111 and a plurality of cameras 112.

A person of ordinary skill in the art can understand that the above-mentioned embodiments are specific embodiments for realizing the present application, and in practical applications, various changes can be made to them in form and details without departing from the spirit and spirit of the present application. range.

Claims

A face recognition method, characterized in that it includes:

Collect at least one two-dimensional image containing the human face through multiple cameras, and select the first two-dimensional image containing the human face and the second two-dimensional image containing the human face collected by any two cameras respectively image;

Extracting the two-dimensional coordinates of the multiple feature points of the human face in the first two-dimensional image, and extracting the two-dimensional coordinates of the multiple feature points in the second two-dimensional image;

According to the preset parameters of any two of the cameras, and the two-dimensional coordinates of each of the feature points on the first two-dimensional image and the second two-dimensional image, a three-dimensional network corresponding to the face is obtained Lattice model

According to the three-dimensional mesh model and the three-dimensional face model of the preset user, it is determined whether the face is the three-dimensional face of the preset user.
The face recognition method of claim 1, wherein in the step of determining whether the face is a three-dimensional face model of the preset user according to the three-dimensional mesh model and a preset user's three-dimensional face model After the face, it also includes:

According to the multiple frames of the three-dimensional mesh model, it is determined whether the human face is a living human face of the preset user.
3. The face recognition method according to claim 2, wherein the determining whether the face is a living face of the preset user according to multiple frames of the three-dimensional mesh model comprises:

Selecting two frames of the three-dimensional grid model, and calculating the deformation between the corresponding grids in the plurality of grids on the two frames of the three-dimensional grid model;

If at least one of the deformation variables is greater than or equal to a preset deformation threshold, it is determined that the face is a living face of the preset user.
The face recognition method according to claim 1, wherein, on the basis of the preset parameters of any two of the cameras, and the first two-dimensional image and the second two-dimensional image Before obtaining the three-dimensional mesh model corresponding to the face, the two-dimensional coordinates of each of the feature points further includes:

Determining whether the two-dimensional image comes from the preset user according to at least one of the two-dimensional image and the two-dimensional face image of the preset user;

If it is determined that the face is from the preset user, enter the preset parameters according to any two of the cameras, and each of the first two-dimensional image and the second two-dimensional image. The step of obtaining the three-dimensional mesh model corresponding to the human face according to the two-dimensional coordinates of the feature points.
The face recognition method of claim 1, wherein the three-dimensional mesh model and a three-dimensional face model of a preset user are used to determine whether the face is a three-dimensional face of the preset user. Face, including:

Calculating the relative position parameters between the grids on the three-dimensional grid model;

Determine whether each of the relative position parameters on the three-dimensional mesh model matches the corresponding relative position parameter on the preset three-dimensional face model;

If each of the relative position parameters on the three-dimensional mesh model matches the corresponding relative position parameter on the preset three-dimensional face model, it is determined that the face is a three-dimensional person of the preset user face.
The face recognition method of claim 1, wherein the preset parameters of any two of the cameras, and each of the first two-dimensional image and the second two-dimensional image The two-dimensional coordinates of the feature points to obtain the three-dimensional mesh model corresponding to the face includes:

Calculate the three-dimensional coordinates of each of the feature points according to the preset parameters of any two of the cameras and the two-dimensional coordinates of each of the feature points on the first two-dimensional image and the second two-dimensional image ；

According to the three-dimensional coordinates of each of the characteristic points, the three-dimensional mesh model corresponding to the face image is constructed.
The face recognition method according to claim 6, characterized in that, based on the preset parameters of any two of the cameras, and the first two-dimensional image and the second two-dimensional image Before calculating the three-dimensional coordinates of each of the feature points, the two-dimensional coordinates of each of the feature points further include:

Correcting the two-dimensional coordinates of each of the feature points on the first two-dimensional image and the second two-dimensional image by using a window matching function and the two-dimensional face image of the preset user;

According to the preset parameters of any two of the cameras, and the two-dimensional coordinates of each of the feature points on the first two-dimensional image and the second two-dimensional image, calculating the value of each feature point Three-dimensional coordinates, specifically:

Obtain each of the characteristic points according to the preset parameters of any two of the cameras, and the corrected two-dimensional coordinates of each of the characteristic points on the first two-dimensional image and the second two-dimensional image The three-dimensional coordinates.
The face recognition method according to claim 7, wherein the use of the window matching function and the two-dimensional face image of the preset user compares the first two-dimensional image and the second two-dimensional image Correction of the two-dimensional coordinates of each of the feature points above includes:

For each of the feature points, respectively setting a window area including the feature point on the first two-dimensional image and the second two-dimensional image;

Searching for points matching the feature points on the two-dimensional face image in the window area of the first two-dimensional image and the window area of the second two-dimensional image respectively, and record them as accurate feature points;

The two-dimensional coordinates of the precise feature point on the first two-dimensional image are taken as the two-dimensional coordinates of the feature point on the first two-dimensional image after correction, and the two-dimensional coordinates on the second two-dimensional image The two-dimensional coordinates of the precise feature point are the two-dimensional coordinates of the feature point on the second two-dimensional image after correction.
8. The face recognition method according to claim 8, wherein said searching for the two-dimensional face in the window area of the first two-dimensional image and the window area of the second two-dimensional image respectively The matching points of the feature points on the image include:

For each window area, according to the gray value of the feature point on the preset two-dimensional face image and the gray value of each point in the window area, each of the points in the window area is obtained The degree of matching between a point and the feature point on the two-dimensional face image;

According to the degree of matching between each point in the window area and the feature point on the two-dimensional face image, obtain a match between the feature point in the window area and the two-dimensional face image Point.
The face recognition method according to claim 1, wherein in the step of determining whether the face is a three-dimensional face model of the preset user according to the three-dimensional mesh model and a preset user's three-dimensional face model Human faces, including:

Correcting the face pose corresponding to the three-dimensional mesh model to a preset pose;

According to the three-dimensional mesh model of the preset posture and the three-dimensional face model of the preset user, it is determined whether the face is the three-dimensional face of the preset user.
A processing chip, characterized in that the processing chip is connected to at least one memory, the processing chip is connected to a plurality of cameras, and the memory stores instructions that can be executed by the processing chip. The processing chip executes, so that the processing chip can execute the face recognition method according to any one of claims 1 to 10.
An electronic device, characterized by comprising the processing chip of claim 11, at least one memory, and multiple cameras.