A human eye-based identity recognition method and system

Abstract

The application relates to an eye-based identity recognition method and system, and belongs to the technical field of face recognition. The method comprises the following steps: inputting an eye image into a target detection model to generate a prediction frame and visual features of the eye image; judging whether the prediction frame is correctly recognized; if yes, obtaining basic features of the eye based on the prediction frame; fusing the visual features, the prediction frame and the basic features of the eye to obtain eye fusion features; recognizing corresponding identity information according to the eye fusion features; and using the eye fusion features obtained by fusing the visual features, the prediction frame and the basic features of the eye, so that the eye fusion features have a combination of multiple features, thereby improving the accuracy of recognizing the corresponding identity by using the eye fusion features. The application has the effect of recognizing the identity of a person by using the eye information in the face information when the person wears a mask.

Description

Technical Field

[0001] This invention relates to the field of facial recognition technology, and in particular to a method and system for identity recognition based on human eyes. Background Technology

[0002] Identity verification refers to the authentication of a person's identity information. With the continuous maturation of artificial intelligence technology, intelligent biometric technology has become a common method of identity verification. Intelligent biometric technology can accurately and quickly identify individuals by analyzing and comparing biometric information such as fingerprints, voiceprints, irises, and facial features.

[0003] When using facial information for identity verification, it is usually necessary to obtain a person's complete face in order to accurately identify the corresponding identity information. However, currently, it is often necessary to wear masks when traveling, which prevents the complete face from being exposed, making it difficult to use facial information to identify a person's identity. Summary of the Invention

[0004] To facilitate the identification of individuals using facial information when they are wearing masks, this application provides an identification method and system based on human eyes.

[0005] Firstly, this application provides an identity recognition method based on human eyes, which adopts the following technical solution: A human eye-based identity recognition method includes inputting a human eye image into a target detection model to generate a prediction box and visual features of the human eye image; Determine whether the predicted bounding box is correctly identified. If so, obtain the basic features of the human eye based on the predicted bounding box. By fusing visual features, prediction boxes, and basic features of the human eye, we obtain the human eye fused features; Based on the fusion characteristics of human eyes, the corresponding identity information is identified.

[0006] By adopting the above technical solution, a target detection model is used to generate a prediction box and visual features. The accuracy of the generated prediction box is then determined. If it is correct, the basic features of the human eye are obtained based on the prediction box. If the prediction box is inaccurate, there is no need to obtain the basic features of the human eye, thus saving the program. The visual features, prediction boxes, and basic features of the human eye are fused to obtain the human eye fusion feature. This human eye fusion feature has a combination of multiple features, which results in high accuracy when using the human eye fusion feature to identify the corresponding identity. This makes it easier to identify a person's identity by using the human eye information in the facial information when the person is wearing a mask.

[0007] Optionally, the target detection model includes an average pooling layer and a fully connected layer; The average pooling layer is connected to the backbone network of the target detection model and is used to compress the features extracted by the backbone network to obtain multiple feature points. The fully connected layer, connected to the average pooling layer, is used to combine all feature points output by the average pooling layer to obtain visual features.

[0008] By adopting the above technical solution, multiple feature points are extracted from the backbone network of the target detection model using an average pooling layer, and then the extracted feature points are combined using a fully connected layer to obtain visual features.

[0009] Optionally, the predicted bounding box includes the coordinate information of the predicted bounding box and the category information of the predicted bounding box. The step of determining whether the predicted bounding box is correctly identified specifically includes: Determine whether the category information of the predicted bounding box is a human eye category; If it is not a human eye category, then the predicted bounding box is determined to be incorrect; If it is a human eye category, then based on the coordinate information of the predicted bounding box, it is determined whether the visual feature is located within the predicted bounding box. If it is, the predicted bounding box is considered correct; otherwise, the predicted bounding box is considered incorrect.

[0010] By adopting the above technical solution, it is determined whether the category information of the prediction box is human eye category. If it is, it means that there is a human eye part in the human eye image. Then, it is determined whether the visual feature is located within the prediction box to determine whether the position and range of the prediction box are correct. If the visual feature is located within the prediction box, the prediction box is determined to be correct.

[0011] Optionally, after determining whether the prediction box is correctly identified, the method further includes: If the predicted bounding box is incorrect, a loss function is selected to train the object detection model.

[0012] By adopting the above technical solution, when the predicted bounding box is incorrect, it indicates that there is an error in the output of the target detection model. At this time, it is necessary to train the target detection model to improve the accuracy of target detection model recognition.

[0013] Optionally, the basic features of the human eye are obtained based on the prediction box, specifically including: Within the predicted bounding box, obtain the grayscale of the human eye and the area of ​​the human eye; Within the predicted bounding box, capture the key point features of the human eye; The basic features of the human eye are composed of the grayscale, area, and key features of the human eye.

[0014] By adopting the above technical solution, the grayscale, area, and key features of the human eye are obtained within the prediction box, which reduces the area of ​​the image to be processed and the basic features of the human eye are more accurate.

[0015] Optionally, acquiring the key point features of the human eye specifically includes: Position the preset standard human eye feature points into the preset box within the prediction frame; Using the coordinates of each standard human eye feature point as the center, the point with the largest response value within a preset distance is obtained as the key point in the human eye image; Combine all key points to obtain key point features.

[0016] By adopting the above technical solution, standard human eye feature points are located in the prediction box. Taking the coordinates of each standard human eye feature point as the center, the point with the largest response value within a preset distance is obtained as the key point in the human eye image. This can accurately obtain each key point in the human eye image. By combining all the key points, a more accurate key point feature is obtained.

[0017] Secondly, this application provides an identity recognition system based on human eyes, employing the following technical solution: An identity recognition system based on human eye recognition method, comprising: The feature extraction unit is used to input the human eye image into the target detection model and generate the prediction box and visual features of the human eye image; The feature verification unit is used to determine whether the predicted bounding box is correctly identified. If so, it obtains the basic features of the human eye based on the predicted bounding box. The feature fusion unit is used to fuse visual features, prediction boxes, and basic features of the human eye to obtain human eye fused features; The identity recognition unit is used to identify corresponding identity information based on the fusion features of human eyes.

[0018] By adopting the above technical solution, the feature extraction unit generates a prediction box and visual features of the human eye image, the feature verification unit determines whether the prediction box is correctly identified, and if so, the basic features of the human eye are obtained based on the prediction box. The feature fusion unit fuses the visual features, the prediction box and the basic features of the human eye to obtain the human eye fusion features. The identity recognition unit identifies the corresponding identity information based on the human eye fusion features.

[0019] Optionally, the feature verification unit includes: The category check sub-unit is used to determine whether the category information of the predicted box is the human eye category; if it is not the human eye category, the predicted box is determined to be incorrect. The feature verification subunit is used to determine whether all feature points in the visual features are located within the predicted box when the preset box is the human eye category, based on the coordinate information of the predicted box. If so, the predicted box is considered correct; otherwise, the predicted box is considered incorrect.

[0020] By adopting the above technical solution, the category checking unit determines whether the category information of the prediction box is human eye category, and the feature checking subunit determines whether the feature points in the visual features are all located within the prediction box based on the coordinate information of the prediction box when the preset box is human eye category. That is, the prediction box is judged from two perspectives: category and feature position, making the judgment result more accurate.

[0021] Thirdly, this application provides a computer device that adopts the following technical solution: A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor performing an eye-based identification method as described in any of the first aspects.

[0022] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium includes a computer program that can be loaded by a processor and executed as described in any of the first aspects, a human eye-based identification method. Attached Figure Description

[0023] Figure 1 This is a flowchart of an identity recognition method according to one embodiment of this application.

[0024] Figure 2 This is a flowchart of a method for determining whether a prediction box is correct, according to one embodiment of this application.

[0025] Figure 3 This is a flowchart of a method for obtaining basic features of the human eye according to one embodiment of this application.

[0026] Figure 4 This is a flowchart of a method for obtaining key feature points according to one embodiment of this application.

[0027] Figure 5 This is a block diagram of an identity recognition system according to one embodiment of this application. Detailed Implementation

[0028] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figure 1-5 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0029] This application discloses an identity recognition method based on human vision. (Refer to...) Figure 1 One method for identity recognition based on human eyes includes: Step S101: Input the human eye image into the object detection model to generate the prediction box and visual features of the human eye image; The target detection model can adopt the YOLOv3 model, which consists of 6 convolutional layers with a kernel size of 3×3 and 23 residual modules. The stride of the convolutional layers is set to 2, which is used to downsample the human eye image.

[0030] Furthermore, to improve the accuracy of the predicted bounding boxes, a feature pyramid is connected after the backbone network of the object detection model. Features from the smallest classification layer in the feature pyramid are then added to other classification layers within the pyramid using a channel attention mechanism. Specifically, the backbone network of the object detection model also connects a first convolutional layer group with an output of 13×13×21, a second convolutional layer group with an output of 26×26×21, and a third convolutional layer group with an output of 52×52×21. The first convolutional layer group is then connected to the second and third convolutional layer groups using a channel attention mechanism. Finally, the outputs of the first, second, and third convolutional layer groups are fused, resulting in an object detection model with both higher accuracy and a larger receptive field (the receptive field refers to the area mapped onto the input image by pixels on the feature map output by each layer of the convolutional neural network).

[0031] The predicted bounding box includes at least its coordinate information and category information. The coordinate information includes four coordinates: x, y, w, and h. The x and y coordinates represent the top-left corner of the predicted bounding box, the w coordinate represents its width, and the h coordinate represents its height. The category information includes the category and a confidence score (C). The category represents the identified type of the object in the predicted bounding box; for example, the "eye" category indicates that the object in the predicted bounding box is an eye. The confidence score (C) represents the degree of matching of the identified category. For example, a confidence score of 0.9 for the "eye" category means that the probability of the object in the predicted bounding box being an eye within the allowable error range is 90%.

[0032] Step S102: Determine whether the predicted bounding box is correctly identified. If so, proceed to step S103. It should be noted that if the predicted bounding box is incorrect, a loss function is selected to train the object detection model. The loss function can be one or more of the following: confidence loss function, classification loss function, and localization loss function. If an incorrect predicted bounding box is identified, subsequent steps are stopped.

[0033] Step S103: Based on the predicted bounding box, obtain the basic features of the human eye; It should be understood that the basic features of the human eye are only obtained after the predicted bounding box is determined to be correct. Once the predicted bounding box is determined to be correct, it means that the human eye is included in the predicted bounding box. At this time, it is only necessary to obtain the basic features of the human eye within the predicted bounding box. No processing is required for the area outside the predicted bounding box, which improves processing efficiency.

[0034] Step S104: Fuse the visual features, prediction boxes, and basic features of the human eye to obtain the human eye fused features; Among them, visual features, prediction boxes, and basic features of the human eye can be fused according to preset weights.

[0035] Step S105: Identify the corresponding identity information based on the human eye fusion features.

[0036] It should be noted that a database containing identity information and human eye fusion features can be established. After the human eye fusion features are generated, they are matched in the database to select the identity information corresponding to the human eye fusion features, thereby completing the identification of the person's identity.

[0037] It should be understood that since human eye fusion features integrate visual features, prediction boxes, and basic features of the human eye from multiple perspectives, relatively accurate recognition results can be obtained by using human eye fusion features alone.

[0038] It should also be noted that, as one use case of this application, it can be used in epidemic prevention electronic sentinels. Current epidemic prevention electronic sentinels obtain a user's health status by recognizing their health code, ID card, etc., thereby determining the user's access rights. However, the technical solution of this application obtains the user's eye information to get human eye fusion features, and then identifies the user's identity information based on these features. Based on the identity information, the corresponding health code information is obtained; it is then determined whether the corresponding health code is green. If so, access is permitted. This achieves the ability to determine a user's access rights based on eye information, without requiring the removal of a mask throughout the process, making the verification process convenient and quick.

[0039] In the above implementation, a target detection model is used to generate a prediction box and visual features. The accuracy of the generated prediction box is then determined. If the prediction box is correct, the basic features of the human eye are obtained based on the prediction box. If the prediction box is inaccurate, there is no need to obtain the basic features of the human eye, thus saving the program. The visual features, prediction box, and basic features of the human eye are fused to obtain the human eye fusion feature. This human eye fusion feature has multiple features, which results in higher accuracy when using the human eye fusion feature to identify the corresponding identity. This makes it easier to identify a person's identity by using the human eye information in the facial information when the person is wearing a mask.

[0040] As one implementation of the object detection model, the object detection model includes an average pooling layer and a fully connected layer; The average pooling layer, connected to the backbone network of the object detection model, is used to compress the features extracted by the backbone network to obtain multiple feature points; A fully connected layer, connected to the average pooling layer, is used to combine all feature points output by the average pooling layer to obtain visual features.

[0041] The fully connected layer outputs its results using the softmax function.

[0042] In this process, multiple feature points are extracted from the backbone network of the object detection model using an average pooling layer, and then the extracted feature points are combined using a fully connected layer to obtain visual features.

[0043] Reference Figure 2 As one implementation of step S102, determining whether the predicted bounding box is correctly identified specifically includes: Step S1021: Determine whether the category information of the predicted box is the human eye category; if it is not the human eye category, proceed to step S1024; if it is the human eye category, proceed to step S1022.

[0044] Specifically, it is determined whether the category in the category information of the predicted bounding box is the human eye category. If not, step S1024 is executed. If yes, it is determined whether the confidence level in the category information of the predicted bounding box is greater than the preset confidence level. If it is greater, it is determined to be the human eye category. The preset confidence level can be set to 0.5 or 0.6.

[0045] Step S1022: Based on the coordinate information of the prediction box, determine whether the visual feature is located within the prediction box. If yes, proceed to step S1023; otherwise, proceed to step S1024.

[0046] It should be understood that visual features are extracted from the human eye region, and the predicted bounding box should cover the entire human eye region. If the visual features and the predicted bounding box are not in the same region, it indicates that there may be an error in the generation of the predicted bounding box.

[0047] Step S1023: Determine that the predicted bounding box is correct.

[0048] It should be understood that a correct prediction box indicates that the prediction box contains a human eye category with high confidence and that the location of the visual feature is also within the prediction box.

[0049] Step S1024: Determine if the predicted bounding box is incorrect.

[0050] It should be noted that when the prediction box is incorrect because the category information of the prediction box is not the human eye category, the classification loss function is selected to train the object detection model; when the prediction box is incorrect because the confidence of the category information of the prediction box is low, the confidence loss function is selected to train the object detection model; and when the prediction box is incorrect because the position of the visual feature is not within the prediction box, the localization loss function is selected to train the object detection model.

[0051] In the above implementation, by determining whether the category information of the prediction box is a human eye category, if so, it means that there is a human eye part in the human eye image. Then, it is determined whether the visual feature is located within the prediction box to determine whether the position and range of the prediction box are correct. If the visual feature is located within the prediction box, the prediction box is determined to be correct.

[0052] Reference Figure 3 As one implementation of S103, based on the prediction box, the basic features of the human eye are obtained, specifically including: Step S1031: Within the predicted bounding box, obtain the grayscale of the human eye and the area of ​​the human eye; Specifically, the area of ​​the human eye can be obtained using contour detection, which involves selecting an appropriate threshold to perform grayscale processing on the human eye image; performing morphological processing on the grayscale-processed human eye image to eliminate residual parts such as eyebrows and nose; locating the contour of the eye in the morphologically processed human eye image and obtaining the area of ​​the human eye based on the eye contour; and then obtaining the grayscale of the corresponding region of the human eye image within the contour, which is the grayscale of the human eye.

[0053] Step S1032: Within the predicted bounding box, obtain the key point features of the human eye; It should be understood that extracting key features of the human eye from the predicted bounding box reduces the image size that needs to be processed, thereby speeding up the process. At the same time, the predicted bounding box essentially provides prior knowledge of the key features of the human eye, facilitating more accurate extraction of these features based on the predicted bounding box.

[0054] Step S1033: Use the grayscale of the human eye, the area of ​​the human eye, and the key features of the human eye to form the basic features of the human eye.

[0055] Reference Figure 4 As one implementation of step S1032, step S1032 specifically includes: Step S10321: Position the preset standard human eye feature points into the preset box in the prediction box; There are multiple standard human eye feature points, typically set at the outline of each eye, with eight to twelve feature points per eye. Additionally, standard human eye feature points can also be set at the eyebrows and the outer edges of the face to aid in determining the eye's position.

[0056] All standard human eye feature points form a human eye feature point group, which constitutes the outline shape of the human eye. When positioning the standard human eye feature point group within the prediction box, it should be positioned as a whole within the prediction box. For example, the center of the human eye feature point group can be positioned at the center of the prediction box.

[0057] It should be understood that since the human eye that needs to be located is already within the prediction box, the human eye position in the image can be found relatively accurately by directly positioning the preset standard human eye feature points within the preset box.

[0058] Step S10322: Using the coordinates of each standard human eye feature point as the center, obtain the point with the largest response value within a preset distance as the key point in the human eye image; The preset distance refers to the area covered by a pixel grid of 5*5, 7*7, or 9*9, and the center of the area covered by the pixel grid is a standard human eye feature point.

[0059] The response value refers to the change in pixel grayscale value. The larger the response value, the more obvious the grayscale change, which indicates that the point is most likely the boundary point of the eye. Therefore, the point with the largest response value is used as the key point of the human eye image to facilitate the location of the edge contour of the human eye.

[0060] Step S10323: Combine all key points to obtain key point features.

[0061] In this process, key points should be combined in the order of preset standard human eye feature points. That is, key points are used to replace corresponding standard human eye feature points, so that the key point features can retain the combined shape of standard human eye feature points, and can be adjusted according to the differences of each person's eyes, so that the obtained key point features are closer to the shape of the human eye being identified.

[0062] In the above implementation, standard human eye feature points are located in the prediction box. Using the coordinates of each standard human eye feature point as the center, the point with the largest response value within a preset distance is obtained as the key point in the human eye image. This can obtain each key point in the human eye image more accurately. By combining all the key points, a more accurate key point feature is obtained.

[0063] This application discloses an identity recognition system based on a human eye-based identity recognition method. (Refer to...) Figure 5 An identity recognition system based on human eye recognition includes: The feature extraction unit is used to input the human eye image into the target detection model and generate the prediction box and visual features of the human eye image; The feature verification unit is used to determine whether the predicted bounding box is correctly identified. If so, it obtains the basic features of the human eye based on the predicted bounding box. The feature fusion unit is used to fuse visual features, prediction boxes, and basic features of the human eye to obtain human eye fused features; The identity recognition unit is used to identify corresponding identity information based on the fusion features of human eyes.

[0064] In the above embodiments, a feature extraction unit generates a prediction box and visual features of a human eye image, a feature verification unit determines whether the prediction box is correctly identified, and if so, the basic features of the human eye are obtained based on the prediction box. A feature fusion unit fuses the visual features, the prediction box, and the basic features of the human eye to obtain the human eye fusion features. An identity recognition unit identifies the corresponding identity information based on the human eye fusion features.

[0065] As one implementation of the feature verification unit, the feature verification unit includes: The category check sub-unit is used to determine whether the category information of the predicted box is the human eye category; if it is not the human eye category, the predicted box is determined to be incorrect. The feature verification subunit is used to determine whether all feature points in the visual features are located within the predicted box when the preset box is the human eye category, based on the coordinate information of the predicted box. If so, the predicted box is considered correct; otherwise, the predicted box is considered incorrect.

[0066] In the above embodiments, the category checking unit is used to determine whether the category information of the prediction box is human eye category. When the preset box is human eye category, the feature checking subunit is used to determine whether the feature points in the visual features are all located within the prediction box based on the coordinate information of the prediction box. That is, the prediction box is judged from two perspectives: category and feature position, so that the judgment result is more accurate.

[0067] The identity recognition method and system based on human eye provided in this application can realize the above-mentioned identity recognition method based on human eye, and the specific working process of the identity recognition system based on human eye can be referred to the corresponding process in the above-mentioned method embodiment.

[0068] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0069] Based on the same technical concept, the present invention also discloses a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes any of the above-described methods for human eye-based identity recognition.

[0070] The present invention also discloses a computer-readable storage medium, characterized in that it includes a computer program that can be loaded by a processor and executed as any of the above-described human eye-based identification methods.

[0071] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0072] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0073] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for identity recognition based on human eyes, characterized in that, include: The human eye image is input into the object detection model to generate a predicted bounding box and visual features of the human eye image; Determine whether the predicted bounding box is correctly identified. If so, obtain the basic features of the human eye based on the predicted bounding box. The visual features, the prediction box, and the basic features of the human eye are fused to obtain the human eye fusion feature; Based on the aforementioned human eye fusion features, identify the corresponding identity information; The target detection model includes an average pooling layer and a fully connected layer; The average pooling layer is connected to the backbone network of the target detection model and is used to compress the features extracted by the backbone network to obtain multiple feature points. The fully connected layer, connected to the average pooling layer, is used to combine all feature points output by the average pooling layer to obtain visual features. The predicted bounding box includes the coordinate information and category information of the predicted bounding box. The determination of whether the predicted bounding box is correctly identified specifically includes: Determine whether the category information of the predicted bounding box is a human eye category; If it is not a human eye category, then the predicted bounding box is determined to be incorrect; If it is a human eye category, then based on the coordinate information of the prediction box, it is determined whether the visual feature is located within the prediction box. If yes, the prediction box is determined to be correct; otherwise, the prediction box is determined to be incorrect. After determining whether the predicted bounding box is correctly identified, the method further includes: if the predicted bounding box is incorrect, selecting a loss function and training the target detection model; The above method obtains the basic features of the human eye based on the prediction box, specifically including: Within the predicted bounding box, obtain the grayscale value of the human eye and the area of ​​the human eye; Within the predicted bounding box, key point features of the human eye are obtained; The basic features of the human eye are composed of the grayscale, area, and key features of the human eye. Position the preset standard human eye feature points into the preset box within the prediction frame; Using the coordinates of each standard human eye feature point as the center, the point with the largest response value within a preset distance is obtained as the key point in the human eye image; Combine all key points to obtain key point features.

2. An identity recognition system based on human eye recognition method, characterized in that, include: The feature extraction unit is used to input the human eye image into the target detection model and generate the prediction box and visual features of the human eye image; The feature verification unit is used to determine whether the predicted bounding box is correctly identified. If so, it obtains the basic features of the human eye based on the predicted bounding box. The feature fusion unit is used to fuse visual features, prediction boxes, and basic features of the human eye to obtain human eye fused features; The identity recognition unit is used to identify corresponding identity information based on human eye fusion features; The target detection model includes an average pooling layer and a fully connected layer; The average pooling layer is connected to the backbone network of the target detection model and is used to compress the features extracted by the backbone network to obtain multiple feature points. The fully connected layer, connected to the average pooling layer, is used to combine all feature points output by the average pooling layer to obtain visual features. The category check sub-unit is used to determine whether the category information of the predicted box is the human eye category; if it is not the human eye category, the predicted box is determined to be incorrect. The feature verification subunit is used to determine whether all feature points in the visual features are located within the predicted box when the preset box is the human eye category, based on the coordinate information of the predicted box. If so, the predicted box is judged to be correct; otherwise, the predicted box is judged to be incorrect. After determining whether the predicted bounding box is correctly identified, the method further includes: if the predicted bounding box is incorrect, selecting a loss function and training the target detection model; The above method obtains the basic features of the human eye based on the prediction box, specifically including: Within the predicted bounding box, obtain the grayscale value of the human eye and the area of ​​the human eye; Within the predicted bounding box, key point features of the human eye are obtained; The basic features of the human eye are composed of the grayscale, area, and key features of the human eye. Position the preset standard human eye feature points into the preset box within the prediction frame; Using the coordinates of each standard human eye feature point as the center, the point with the largest response value within a preset distance is obtained as the key point in the human eye image; Combine all key points to obtain key point features.

3. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor performing a human eye-based identification method as described in claim 1.

4. A computer-readable storage medium, characterized in that, It includes a computer program that can be loaded by a processor and executed as described in claim 1, which is based on human eye recognition.

Images