Multi-angle face recognition method of focus-adjustable code scanning camera

Abstract

The invention provides a multi-angle face recognition method of a focus-adjustable scanning camera, and relates to the technical field of face recognition, and the method comprises the steps: obtaining a scanning activation instruction, and carrying out the collection of a face video stream within a preset range through the focus-adjustable scanning camera; tracking and recognizing a face image in the face video stream, and judging whether a preset recognition target is reached or not; if not, adjusting the acquisition parameters of the focusable code scanning camera according to the deviation value of the face recognition image and the preset recognition target until the preset recognition target is met; and rotating and zooming the face recognition image, projecting the face recognition image to a standard posture for face comparison, and obtaining a face recognition result. The technical problems that in the prior art, due to the fact that the prior art generally depends on a fixed-focal-length camera and lacks self-adaptive adjustment of the focal length, under different distances, angles or environment changes, the image definition is insufficient, clear facial features cannot be extracted, and the recognition precision is affected are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of face recognition, and particularly to a multi-angle face recognition method for an adjustable-focus code-scanning camera. Background Art

[0002] As a non-contact identity authentication method, face recognition technology has been widely applied in fields such as access control systems, financial payments, and security monitoring. Although traditional fixed-focus cameras can provide relatively clear images in some fixed scenarios, since the focal length of the camera cannot be adjusted, if the face is at different distances or angles, the camera may not be able to provide appropriate image quality. This limitation is particularly evident in multiple application scenarios, such as security monitoring and access control management. Especially when the face is at an irregular angle, the camera often cannot focus on a clear image. Blurred images or incorrect focal lengths will lead to inaccurate feature extraction, thereby affecting the accuracy of face recognition. Summary of the Invention

[0003] The present application provides a multi-angle face recognition method for an adjustable-focus code-scanning camera, aiming to solve the technical problem that the prior art usually relies on a fixed-focus camera and lacks adaptive adjustment of the focal length, resulting in insufficient image clarity under different distances, angles, or environmental changes, thereby being unable to extract clear facial features and affecting the recognition accuracy.

[0004] The multi-angle face recognition method for an adjustable-focus code-scanning camera disclosed in the present application includes: obtaining a code-scanning activation instruction, and collecting a face video stream within a preset range through the adjustable-focus scanning camera; tracking and recognizing the face images in the face video stream, and determining whether the face recognition image reaches a preset recognition target; when it does not reach, adjusting the acquisition parameters of the adjustable-focus code-scanning camera according to the deviation amount between the face recognition image and the preset recognition target until the preset recognition target is met; rotating and scaling the face recognition image, and projecting it to a standard pose for face comparison to obtain a face recognition result.

[0005] One or more technical solutions provided in the present application have at least the following beneficial effects: Through the scanning code activation instruction, it is possible to collect the face video stream within a preset range. By using an adjustable focus camera, the quality of image collection is effectively ensured. Especially in different environments and distances, the camera can automatically adjust the focus according to needs, improving the clarity and details of the image; by tracking and recognizing the face images in the face video stream, it is possible to dynamically determine whether the target meets the preset recognition criteria, which not only ensures the adaptability of the system to dynamic changes, but also adjusts the recognition strategy according to the real-time image, improving the recognition accuracy and efficiency; when the image quality does not reach the preset target, the acquisition parameters of the adjustable focus camera are automatically adjusted based on the deviation amount between the image and the target. This adjustment ensures that the image can reach the best clarity, thereby reducing the risk of misrecognition caused by improper focus or blurred image, and significantly improving the accuracy of the recognition result; the recognized image is rotated and scaled, and projected to a standard pose. This standardization process ensures that even when the face appears at different angles or poses, the image can still be aligned with the preset standard face model, thus improving the consistency and accuracy of recognition; after the image is projected to the standard pose, precise face comparison is performed. In this way, the inconsistencies caused by angles and poses can be effectively eliminated, further improving the accuracy and stability of face recognition.

[0006] The above description is only an overview of the technical solution of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the specific embodiments of this application are specifically given below. Brief Description of the Drawings

[0007] Figure 1 It is a schematic flowchart of the multi-angle face recognition method of the adjustable focus scanning code camera provided by the embodiment of this application.

[0008] Figure 2 It is a schematic flowchart of the face image tracking and recognition in the multi-angle face recognition method of the adjustable focus scanning code camera provided by the embodiment of this application. Detailed Description of the Embodiments

[0009] By providing a multi-angle face recognition method for an adjustable focus scanning code camera in the embodiment of this application, the technical problem in the prior art that usually relies on a fixed focus camera and lacks adaptive adjustment of the focal length, resulting in insufficient image clarity under different distances, angles or environmental changes, thus unable to extract clear facial features and affecting the recognition accuracy, is solved.

[0010] After introducing the basic principle of this application, the various non-restrictive embodiments of this application will be specifically introduced below in conjunction with the drawings of the specification.

[0011] Such as Figure 1As shown in the figure, the embodiment of the present application provides a multi-angle face recognition method for an adjustable focus barcode scanning camera, and the method includes: Obtain a barcode scanning activation instruction, and collect a face video stream within a preset range through an adjustable focus scanning camera.

[0012] Obtaining a barcode scanning activation instruction means that the system starts the recognition process by scanning a two-dimensional code or other forms of barcodes. The barcode scanning activation instruction can come from information input by the user through a mobile phone, device or computer. For example, the barcode scanning instruction can be to scan a two-dimensional code through an application program to start the system and activate the face recognition function of the adjustable focus barcode scanning camera.

[0013] After obtaining the barcode scanning activation instruction, enable the adjustable focus camera. The camera has the function of adjustable focal length and can adjust the shooting distance and clarity according to needs. Such a camera is usually set within a preset range, such as a specific space or area, so that it can capture face images within this range. The adjustable focus function allows the camera to adapt to different face distances and environmental changes, ensuring the clarity of the image and the accuracy of capture.

[0014] After activation, the adjustable focus camera starts to capture a face video stream. This is achieved by a video capture device continuously shooting and transmitting a real-time video stream, usually recording dynamic images in the target area at a speed of multiple frames per second. The face images contained in the video stream are transmitted to the recognition system in real time for processing. Through the real-time video stream, the target face can be dynamically detected and recognized.

[0015] Track and recognize the face images in the face video stream, and determine whether the face recognition image reaches a preset recognition target.

[0016] Analyze each frame of the image in the video stream in real time. First, perform face detection to find the possible face areas in the video stream. Then, use face tracking algorithms, such as region-based convolutional neural networks, deep learning techniques, etc., to track the faces in each frame, ensuring that even if the person is moving, the camera can still accurately locate and recognize the face. This tracking process not only detects the presence of the face, but also can accurately locate the position and angle of the face through facial key points, such as eyes, nose, mouth, etc.

[0017] After tracking and recognition, determine whether the currently recognized face image meets the preset recognition target. The preset target includes multiple criteria, such as the clarity of the face, the integrity of facial features, recognition accuracy, etc. For example, set a specific recognition accuracy or accuracy standard, such as a 99% matching degree, and compare the current recognition result with the target recognition standard for judgment. If the recognition image does not reach the target standard, the image processing will continue or new images will be collected for comparison.

[0018] When it is not achieved, adjust the acquisition parameters of the focus-adjustable barcode scanning camera according to the deviation amount between the face recognition image and the preset recognition target until the preset recognition target is met.

[0019] If the current image fails to meet the preset recognition target, for example, the image clarity is insufficient, the recognition accuracy is not high, or insufficient feature information can be obtained from the current image, then the acquisition parameters need to be adjusted. The deviation amount refers to the gap between the current image and the preset target, including deviations in position, such as the distance between the face and the camera being too far or too close; deviations in focal length, where the focal length fails to be adjusted to the optimal value; and other deviations such as image rotation or angle. By comparing the matching degree between the features of the recognition image and the preset target features, the value of the deviation is obtained, and the deviation amount is calculated.

[0020] Based on the calculated deviation amount, automatically adjust the acquisition parameters of the focus-adjustable barcode scanning camera, such as focal length, the angle of the camera, the direction of the lens, etc. For example, if the deviation amount indicates insufficient focal length, the focal length will be increased to re-align with the target face. If the deviation amount indicates an incorrect angle, the rotation angle of the camera will be adjusted to align it to the correct position. This adjustment process is carried out automatically, and the adjustment amplitude is determined according to the current deviation amount until the image quality and recognition accuracy reach the preset target. This adjustment mechanism ensures that the system can achieve the best effect in face recognition under different environments and different recognition target conditions.

[0021] Rotate and scale the face recognition image, project it to a standard pose for face comparison, and obtain the face recognition result.

[0022] Since the angle and position of the face in the camera may change, for example, the user may turn their head or stand too close or too far away, before face comparison, it is necessary to process the image through rotation and scaling operations to make its pose close to the standard pose. The rotation operation can adjust the angle of the face image to ensure it faces the camera directly. The scaling operation scales according to the size of the face to make it meet the preset size.

[0023] Projecting the rotated and scaled face image to a standard pose means standardizing the image. The standard pose is usually a widely adopted unified pose, such as a standard face image facing the camera directly. Through this operation, face images of different angles and sizes can be converted into a unified standardized form, reducing errors caused by shooting angles and distances, and making the comparison more accurate.

[0024] After being projected to the standard pose, the image is compared with the standard face template in the database. Using feature-based comparison algorithms such as convolutional neural networks, face recognition algorithms, etc., the similarity between the features of the input image and the standard template is compared. Finally, the face recognition result is output according to the comparison result. If the recognition result is a match, the identity information of the recognized person is returned; if the recognition result is a non-match, it means that the corresponding person cannot be recognized.

[0025] Furthermore, as Figure 2 shown, the face images in the face video stream are tracked and recognized, including: Performing feature analysis on the collected face data to obtain multi-dimensional analysis features; respectively performing recognition sensitivity comparison for each dimension based on the multi-dimensional analysis features to obtain sensitive features for each dimension; determining core recognition features and auxiliary recognition features according to the dimension recognition response coefficients of the sensitive features for each dimension; and performing tracking and recognition on the face images in the face video stream based on the core recognition features and auxiliary recognition features.

[0026] Face images contain a large amount of feature information, and these features can help the system identify individuals. Feature analysis is to analyze the collected image data and extract a series of recognition-significant features. These features can be divided into multiple dimensions, and each dimension represents different facial features or information. Specifically, it includes facial geometric features: such as the facial contour, the positions and proportions of the eyes, nose, and mouth; texture features: including the fine textures on the skin surface, which are very helpful for recognition; motion features: if the video stream is dynamic, motion features can help identify the dynamic changes of the face, such as eye blinks and mouth movements; skin features: the color and texture of the skin; personalized features: such as the uniqueness of facial features, which can be certain specific facial marks, such as freckles, scars, and the shape of the eyes. By extracting features from the collected image data, multi-dimensional analysis features are obtained, and these features reflect different aspects and information from different angles of the face.

[0027] Further analyze the multi-dimensional analysis features extracted from the face image. Specifically, analyze the contribution degree of the features in each dimension to face recognition, calculate the sensitivity of the features in each dimension. The purpose of the recognition sensitivity comparison is to evaluate the influence of the features in each dimension on face recognition under different conditions, especially to compare the differences between different individuals and the stability of the same individual under different conditions. Through comparison and analysis, features with higher recognition sensitivity in each dimension are obtained. Sensitive features are those that have a greater impact on recognition. For example, facial geometric features (such as the distance between the eyes and the nose) are more sensitive than skin features because they have greater differences between different individuals.

[0028] Based on the obtained sensitive features in each dimension, calculate the recognition response coefficient for each feature. The recognition response coefficient represents the importance and response ability of the feature in recognition. For example, some features can better help distinguish different people during recognition, while other features may only work in certain specific scenarios. According to the response coefficients of the sensitive features in each dimension, screen out the core recognition features and auxiliary recognition features. Among them, the core recognition features contribute the most to the recognition result and are the most sensitive and reliable features in the recognition process, usually facial geometric features, skin features, etc.; the auxiliary recognition features are helpful for improving the recognition accuracy and enhancing the recognition stability, but are not decisive. For example, personalized features and some subtle texture features may be used as auxiliary features to help the recognition system make better decisions.

[0029] Use the determined core recognition features and auxiliary recognition features to continue tracking and recognizing the face images in the video stream. In this process, achieve high-precision recognition through the core recognition features, improve the robustness and stability of recognition through the auxiliary recognition features. Through continuous tracking and analysis, it is possible to judge the identity of the face image or verify whether the target face is the expected person, and obtain the face recognition result in real time.

[0030] Furthermore, the feature parsing of the collected face data to obtain multi-dimensional parsing features includes: Take facial geometric features, texture features, motion features, skin features, and personalized features as parsing dimensions; extract features from the collected face data item by item according to the parsing dimensions to obtain the multi-dimensional parsing features.

[0031] Facial geometric features refer to the structural features of the human face, mainly including the relative positions and proportions of key parts such as the facial contour, eyes, nose, and mouth. These features are usually unique marks of each face and have significant differences among different individuals; texture features refer to the fine surface structure features of the facial skin, usually including details such as facial wrinkles, skin color, spots, and pimples. Texture features are very important for face recognition, especially in different lighting or poses, and they can still provide useful information; motion features refer to the change patterns of the human face in dynamic videos, such as eye blinking, mouth opening and closing, etc. These actions are often important features in face dynamic recognition; skin features refer to the characteristics of the facial skin color, texture, etc. Skin features are often related to a person's age, gender, health status, etc.; personalized features refer to the unique marks on each person's face, and these features may be difficult to identify through conventional geometric features or texture features, such as facial scars, freckles, wrinkles, eye shapes, etc.

[0032] According to the parsing dimensions, feature extraction is performed on the collected face data, that is, each frame of image in the video stream. Each parsing dimension corresponds to a specific feature extraction process, and each dimension is processed one by one. For example, facial geometric features are extracted through facial key point detection technology. By identifying and calibrating the positions of eyes, nose, mouth, etc., the proportions and positions between various parts of the face can be calculated. These geometric features are extracted and converted into digital data, becoming the extraction results of facial geometric features; in the process of texture feature extraction, the skin surface details of the face image are analyzed, and image processing algorithms such as texture analysis and gray-level co-occurrence matrix are used to extract minute texture differences, thereby generating vector data representing facial texture features. By extracting the features of each dimension item by item, a feature vector composed of multi-dimensional features is finally obtained, that is, multi-dimensional parsing features. These features contain information about the geometric shape, texture details, dynamic changes, skin characteristics, and personalized identifiers of the face, etc.

[0033] Furthermore, based on the multi-dimensional parsing features, sensitivity comparisons for each dimension are respectively carried out to obtain sensitive features for each dimension, including: Obtain the inter-class differences and intra-class stabilities of each feature in the multi-dimensional parsing features. The inter-class differences describe the difference degrees of the same feature among different individuals, and the intra-class stabilities describe the feature fluctuations of the same individual under different conditions; quantify the recognition sensitivities of each feature in the multi-dimensional parsing features according to the inter-class differences and intra-class stabilities to obtain the sensitivity coefficients of each feature in the multi-dimensional parsing features; perform sensitive feature comparison and determination according to the sensitivity coefficients to obtain the sensitive features for each dimension.

[0034] The inter-class differences are used to measure the difference degrees between different individuals in the same feature. For example, if the inter-pupillary distance is used as a feature, the inter-class differences refer to the differences in the inter-pupillary distances between different individuals. The calculation method of the inter-class differences can be based on variance or standard deviation in statistics, especially by calculating the difference between the average value of the feature values of each individual and the overall feature mean value.

[0035] The intra-class stabilities describe the feature fluctuations of the same person under different conditions, referring to the changes in facial features of the same person at different times, different angles, or different environmental conditions. Assuming there is data of the same person collected multiple times in the database, the intra-class stabilities can be quantified by calculating the variance or standard deviation of the same feature under different collection conditions. For example, for the feature of inter-pupillary distance, the intra-class stability represents the fluctuation range of the inter-pupillary distance in multiple collections of the same person.

[0036] Suppose we want to evaluate the sensitivity of the eye distance feature. First, calculate the between-class difference of this feature (i.e., the difference in eye distance among different individuals) and the within-class stability (i.e., the change in eye distance of the same person under different conditions). If the between-class difference is large and the within-class stability is small, it indicates that the eye distance has a relatively high sensitivity for distinguishing different individuals. However, if the within-class stability is also large, it means that the eye distance of the same person varies greatly under different conditions, thus reducing the stability and reliability of this feature.

[0037] Quantifying recognition sensitivity means combining the between-class difference and the within-class stability to comprehensively evaluate each feature and obtain the sensitivity coefficient of this feature. The sensitivity coefficient represents the importance of this feature in the entire recognition process. For example, it can be calculated by weighting the combination of the between-class difference and the within-class stability. After quantifying each feature, the sensitivity coefficient of each feature is obtained. A feature with a high sensitivity coefficient indicates a greater contribution to the recognition result. For example, when evaluating the eye distance feature, if the between-class difference of the eye distance is large and the within-class stability is small, then the sensitivity coefficient of the eye distance will be relatively high, which means that the eye distance is a feature that is very sensitive for distinguishing individuals and should play an important role in subsequent recognition algorithms.

[0038] After the sensitivity coefficients of all features are calculated, compare and screen all features according to these coefficients. Select the most critical features for face recognition according to the sensitivity coefficients. Specifically, according to the sensitivity coefficient of each feature, determine which features have sufficient discrimination ability and regard these features as sensitive features. Sensitive features usually have a greater impact on the final recognition result.

[0039] Furthermore, according to the dimension recognition response coefficients of the sensitive features in each dimension, determine the core recognition features and auxiliary recognition features, including: By identifying sample data, calculate the recognition contribution degree and recognition speed of the sensitive features in each dimension; configure the evaluation weight values of the recognition contribution degree and recognition speed, perform recognition response coefficient operations according to the recognition contribution degree and recognition speed of the sensitive features in each dimension, and obtain the recognition response coefficients of the sensitive features in each dimension; compare and screen the recognition response coefficients of the sensitive features in each dimension according to the feature screening threshold to determine the core recognition features and auxiliary recognition features.

[0040] The recognition contribution measures the importance of each feature in the actual recognition process. Specifically, the recognition contribution refers to the degree of influence of each feature on the final recognition result. Features with high contribution usually have a greater impact on improving the recognition accuracy. To calculate the recognition contribution, it can be tested through experimental data or training data. Each time a specific feature is used for recognition, the accuracy rate of the recognition result is calculated. For example, when only using facial geometric features for recognition, the recognition accuracy is 90%; when using multiple features including texture, skin, etc., the accuracy rate increases to 95%. Through this comparison, the size of the contribution of each feature to the recognition result can be obtained.

[0041] The recognition speed measures the computational efficiency of a certain feature in the recognition process, that is, the response time or computational time of the system when using this feature for recognition. A higher recognition speed indicates that the calculation of this feature is relatively simple, so it is very important in a real-time face recognition system. The evaluation of the recognition speed can usually be obtained by measuring the processing time each time a specific feature is used. For example, geometric features may only require simple key-point detection of the image, while skin texture features may require more complex image processing. Therefore, geometric features may be superior in terms of recognition speed.

[0042] The evaluation weight is a parameter used to determine the importance of the recognition contribution and the recognition speed in the comprehensive evaluation. These weights are configured according to the requirements of the actual application. For scenarios with higher real-time requirements, a higher weight will be given to the recognition speed; while for scenarios that require high-precision recognition, the weight of the recognition contribution will be higher. The recognition response coefficient is a comprehensive index calculated based on the recognition contribution and the recognition speed of the feature. The specific calculation method is to sum the recognition contribution and the recognition speed of each feature after weighting according to the evaluation weight to obtain the response coefficient of this feature. An exemplary calculation formula is as follows: , where, is the recognition response coefficient of the i-th feature, is the recognition contribution, is the evaluation weight of the recognition contribution, is the recognition speed, is the evaluation weight of the recognition speed. The obtained recognition response coefficient represents the comprehensive utility of the feature in the recognition process. Features with higher response coefficients indicate that they have higher importance and efficiency in recognition. After calculating the recognition response coefficients of each dimension feature, the recognition response coefficients of all features are obtained. These coefficients help determine which features should be emphasized in subsequent recognition.

[0043] The feature screening threshold is the threshold value used to determine which features can be selected as core features. This threshold is usually determined through experiments, aiming to screen out the features that make the greatest contribution to recognition. For example, setting the threshold to 0.8 means that features with a recognition response coefficient higher than 0.8 will be considered core features.

[0044] Compare the calculated recognition response coefficients with a preset feature screening threshold. Features higher than the threshold will be selected as core recognition features, while features lower than the threshold may be regarded as auxiliary features. The feature screening process is not limited to selecting a single core feature. Instead, it can build an optimized recognition model by combining multiple features. The combination of multiple features may enhance the recognition accuracy of the system. In some cases, the combination of multiple features can jointly form a more efficient recognition model. Flexibly select the optimal combination of core and auxiliary features according to the response coefficients of each feature to ensure the best recognition effect in different scenarios.

[0045] Furthermore, determining the core recognition features and auxiliary recognition features includes: Determine the feature screening threshold according to the recognition features of the recognition scenario. The feature screening threshold includes a quantity threshold and a recognition response coefficient threshold. Screen and sort the sensitive features in each dimension according to the recognition response coefficient thresholds of the core recognition features and auxiliary recognition features respectively, and then perform sequential screening based on the quantity threshold according to the sorting to obtain the core recognition features and auxiliary recognition features.

[0046] The recognition scenario refers to the actual application scenario or task-specific requirements. For example, some scenarios may require a fast response time, while other scenarios may place more emphasis on high-precision recognition. In different recognition scenarios, the selection and screening criteria for features will vary. For example, in a security monitoring system, high speed and real-time performance are required, so features with faster response speeds will be preferred; while in scenarios with high security requirements such as banks, more attention is paid to recognition accuracy, so features with higher recognition contributions will be selected.

[0047] The quantity threshold determines how many core features and auxiliary features the system selects for final recognition. The quantity threshold sets the upper limit of the number of features, that is, at most how many features are used in the final recognition. By setting the quantity threshold, the use of too many redundant features can be avoided, the computational amount can be reduced, and the efficiency can be improved; the recognition response coefficient threshold is used to screen out features with higher response coefficients. Features with higher response coefficients have a greater impact on the final recognition. Usually, these features have strong recognition capabilities in different scenarios. Features lower than this threshold will be considered to make little contribution to the recognition. Adjust the setting of the feature screening threshold according to the specific application scenario and system requirements.

[0048] For each feature, compare its recognition response coefficient with the set response coefficient threshold to screen out the features that meet the requirements. Among them, the core recognition features usually have higher response coefficients, so these features will pass the screening first. The response coefficients of the auxiliary recognition features may be lower, but they still help to improve the stability of recognition, so they will also participate in the screening process.

[0049] After screening out the features that meet the response coefficient threshold, sort them according to their response coefficients, that is, arrange the features in descending order of the recognition response coefficient. The purpose of sorting is to preferentially select those features that contribute more to recognition. After the feature sorting, screen according to the set quantity threshold. The quantity threshold determines how many features will be finally selected for recognition. Finally, obtain the core recognition features and auxiliary recognition features through the above screening process.

[0050] Furthermore, judging whether the face recognition image reaches the preset recognition target includes: Extract the recognition requirement parameters of the core recognition features and auxiliary recognition features respectively; determine the recognition constraint coefficients according to the recognition influence of the core recognition features and auxiliary recognition features; set the preset recognition target according to the recognition requirement parameters and their recognition constraint coefficients.

[0051] Extract the recognition requirement parameters of the core recognition features and auxiliary recognition features respectively. These parameters are used to describe how to meet the requirements of the feature for recognition accuracy in actual applications, including: clarity requirements: some features have higher requirements for the clarity of the image. For example, fine features on the face such as wrinkles, the corners of the eyes, and the corners of the mouth require higher clarity to be accurately recognized. If the image is not clear enough, it may affect the recognition accuracy of these features, thus affecting the overall face recognition result; recognition granularity: some features require finer recognition granularity. For example, details such as wrinkles and skin color differences may require higher precision and granularity in recognition to ensure that the system can correctly distinguish different individuals or judge a specific state; light and environmental adaptability: for example, skin features may have different sensitivities to different lighting conditions. Environmental adaptability refers to whether the features can still effectively participate in the recognition process under different backgrounds or light conditions.

[0052] If the quality of the image does not meet these requirement parameters, such as insufficient resolution, poor lighting conditions, or unclear features, these features cannot be accurately recognized. For the core recognition features, their recognition requirements are usually higher, requiring higher resolution and clarity in the image, while for the auxiliary recognition features, the requirements are relatively lower.

[0053] The impact of each feature on the recognition result is different. Core recognition features usually have a higher impact because they are directly related to the accuracy of identity recognition. For such features, if they do not meet the expected recognition requirement parameters, the recognition accuracy will be greatly reduced; while auxiliary recognition features, although they have an impact on the final recognition result, their impact is relatively small. Therefore, the requirements for these features can be slightly relaxed.

[0054] According to the impact of each feature, a corresponding recognition constraint coefficient is set for each feature, that is, the priority for the feature to meet the recognition requirement parameters. For core recognition features with high impact, if the image does not meet their requirement parameters, strict requirements will be imposed on the image to achieve high precision to avoid affecting the recognition result. At this time, the constraint coefficient will be high; for auxiliary recognition features with less impact, even if the image quality does not fully meet the requirements, a certain degree of error can still be tolerated. At this time, the constraint coefficient will be relatively low, allowing recognition errors or ambiguities under certain conditions.

[0055] The preset recognition target is a standard set according to the recognition requirement parameters and recognition constraint coefficients of each feature. During the recognition process, this target will be used as a reference to determine whether the current image meets the recognition requirements. For example, for core recognition features, the set target will include high clarity, resolution, and less environmental interference; for auxiliary recognition features, the target includes some acceptable error ranges, such as lower clarity or resolution requirements, because these features have less impact on the final recognition. The setting of the preset recognition target ensures that the system can flexibly adapt to different recognition conditions and maximally guarantee the recognition accuracy.

[0056] Furthermore, when the requirements are not met, the acquisition parameters of the adjustable-focus barcode scanning camera are adjusted according to the deviation amount between the face recognition image and the preset recognition target, including: Based on the face recognition image and the preset recognition target, obtain the maximum parameter deviation amount; use the maximum parameter deviation amount as the adjustment target, and perform adaptive focusing based on the acquisition parameter adjustment relationship of the adjustable-focus barcode scanning camera.

[0057] Compare the currently acquired face recognition image with the preset recognition target set in advance. The deviation amount represents the difference between the actual features of the image and the preset target. The calculation of this deviation amount is usually based on various factors such as the clarity, focus level, and feature matching degree of the image. For example, if the image is blurred or out of focus, the deviation amount will be large; if the image is clear and in accurate focus, the deviation amount will be small. The maximum parameter deviation amount refers to the maximum value among all the deviation amounts in the current image, which usually reflects the most serious deviation and indicates the error in certain features or clarity of the current image.

[0058] The maximum deviation is used as the adjustment target during the focusing process. That is, the maximum deviation between the current image and the preset target is used as the standard for adaptive focusing adjustment. The goal is to minimize the deviation as much as possible so that the clarity and feature accuracy of the image reach the preset standard. Among them, adaptive focusing means automatically adjusting the acquisition parameters of the focusable barcode scanning camera according to the deviation between the current image and the preset target. For example, if the deviation indicates insufficient clarity of the image, the focus state of the image can be adjusted by increasing or decreasing the focal length. The adjustment of the focal length helps to optimize the clarity of the image, especially for targets at close or far distances. This adjustment process is automated. According to the calculated deviation and the adjustment relationship of the camera, the focusing is gradually carried out until the deviation reaches an acceptable range or the image quality meets the preset standard.

[0059] Furthermore, adjusting the acquisition parameters of the focusable barcode scanning camera according to the deviation between the face recognition image and the preset recognition target includes: Obtaining the parameter deviation of each recognition feature according to the position distribution of the core recognition feature and the auxiliary recognition feature; performing compensation partitioning according to the position distribution and the parameter deviation of the corresponding recognition feature; respectively performing acquisition parameter adjustment relationship operations based on the parameter deviation of each compensation partition to obtain the acquisition parameters of each compensation partition for partition adjustment.

[0060] The positions and distributions of each recognition feature in the image are different. By analyzing the position distribution of these features, it is possible to understand the spatial positions of each feature in the image and their relative relationships. In the image, different recognition features may deviate, and these deviations are usually reflected as changes in aspects such as position, size, and shape. The parameter deviation is a measure of the impact of these changes on the image quality and recognition accuracy. The parameter deviation calculation can be carried out by comparing the difference between the actual position and the standard position of each recognition feature. For example, the distance or angle change of the eye position deviation will be calculated as the deviation of this feature. Among them, the standard position is determined based on the statistical analysis of large-scale training data. Specifically, based on a large-scale face database, a reference distribution of facial features is generated through statistical learning. For example, the key points (such as the center of the eyes, the tip of the nose, and the corners of the mouth) of all faces in the database are normalized, and their average coordinates are calculated as the standard position.

[0061] The compensation partition divides the image into several regions according to the position distribution of the recognition features in the image and the corresponding deviation amounts, so as to compensate and adjust the deviation of each region. By this method, targeted adjustment can be made according to the deviation amounts of the features in different regions. For example, some regions may have larger deviations and require more compensation, while other regions have smaller deviations and lower compensation requirements. According to the position distribution of each feature, the image is divided into multiple regions, and each region corresponds to a specific feature or a group of features. For example, the region where the eyes and nose are located can be used as one partition, and the mouth and chin can be used as another partition. Within each partition, the compensation amount is determined according to the parameter deviation amount of the features in that partition. If the feature deviation in a certain region is large, more compensation will be performed for that region. Conversely, if the deviation is small, the compensation amount will be less.

[0062] The adjustment relationship of the acquisition parameters within each compensation partition refers to how to adjust parameters such as focal length, angle, and aperture to reduce the deviation amount of the features in that partition. The adjustment relationship can be calculated through models and algorithms, and usually determines what adjustments need to be made for each partition based on the feedback of image quality and the magnitude of the deviation amount. For example, the change in focal length can affect the clarity of the image and the positioning of the features, and the adjustment of the lens angle can affect the focusing position of the image.

[0063] For each compensation partition, the acquisition parameters to be adjusted are calculated based on the deviation amount in that partition. For example, if the position of the eyes in a certain partition is shifted, the focal length and angle will be adjusted to ensure that the image of the eyes is optimized. By adjusting the acquisition parameters of each partition, more accurate image acquisition can be achieved and unnecessary errors can be reduced.

[0064] Furthermore, the face recognition image is rotated and scaled, and projected to a standard pose for face comparison to obtain the face recognition result, including: According to the feature distribution position and distribution angle of the multi-dimensional analysis features, three-dimensional feature marking partitions are made for the face to establish a bottom-layer feature partition, where the marking lines of the feature partitions are the feature lines of the analysis features; according to the feature distribution position and distribution angle of the core recognition features and auxiliary recognition features, re-partitioning is performed in the bottom-layer feature partition to construct a face feature distribution map, where the face feature distribution map has core feature, auxiliary feature, and basic feature labels; the face recognition image is rotated and scaled, and aligned and projected with the face feature distribution map to identify the matching feature partitions, and the core recognition features and auxiliary recognition features are used in sequence for feature comparison to obtain the face recognition result.

[0065] Determine the distribution positions and angles of these features in space based on the extracted multi-dimensional analysis features. For example, the positions of eyes, nose, and mouth are usually within specific regions of the human face. Identify the spatial relationships of the features through this information. Based on the distribution positions and angles of the features, divide the face image into multiple three-dimensional feature marking partitions. These partitions correspond to the features of different regions of the face. For example, the eye region, nose region, mouth region, etc. can be used as different partitions. Among them, mark feature lines within each partition. The feature lines identify the main features within each region. For example, the distance between eyes, the height of the nose bridge, etc. The underlying feature partitions refer to the most basic and finest-grained region divisions in the image. These partitions are usually closely related to the detailed features of the face, such as the angle of the eyes, the curvature of the eyebrows, etc. The marking lines of each partition represent the specific features within that region, such as the position and angle of the eyes, etc.

[0066] The core recognition features are crucial for face recognition. They are distributed within specific regions of the human face. According to the distribution position, angle, and other information of these features, map them into the underlying feature partitions. These core features will form the main part of the entire face feature distribution map; although the auxiliary features contribute less to recognition, they still provide additional information. Divide them into corresponding regions according to their positions and angles in the image.

[0067] Based on the distribution of the core features and auxiliary features, construct a face feature distribution map in the underlying feature partitions. This map includes the spatial positions, angles, and markings of feature lines of each feature, and also indicates the category of each feature through labels. These serve as the basic feature labels. The label of each feature helps the system understand its position and role in space. Core features usually have higher weights during the recognition process, while auxiliary features provide additional auxiliary information, which helps to improve the robustness of recognition.

[0068] In order to match the face recognition image with the face feature distribution map, first, rotation and scaling operations need to be performed on the input image. This is because the face in the input image may have different angles, poses, or scaling ratios. The rotation operation makes the face face the standard frontal angle, and the scaling operation adjusts the size of the face to align it with the standard model in the feature distribution map. Among them, the rotation operation adjusts the image within the coordinate system through geometric transformation of the input image to correct the deviation caused by the face during shooting, so that the face features in the image face a unified standard angle. For example, when the face is skewed to one side in the image and not at a vertical angle, the positions of the face contour and feature points in the image will deviate from the standard model. Through the rotation operation, the positions of these feature points are transformed to the positions consistent with the standard frontal angle. These two operations ensure the matching of the face image with the preset standard model, reducing the recognition error caused by different angles and distances.

[0069] After rotation and scaling operations, the input image is aligned and projected with the facial feature distribution map, which means that coordinate transformation will be performed on the image to accurately align it with the feature points in the feature distribution map. After alignment, the feature differences between the two can be directly compared. This process ensures that images taken at different angles and poses can be effectively compared through the standardized feature distribution map.

[0070] After alignment, the core recognition features and auxiliary recognition features are used for comparison in sequence. The core features are compared first. If these features have a high degree of matching, a preliminary recognition judgment can be made. Subsequently, the auxiliary features are used to further verify the recognition result, such as skin texture and facial expressions. By comparing the similarity of these features, the face recognition result is finally obtained and output, such as the matching identity or identity verification information.

[0071] In summary, the multi-angle face recognition method for the adjustable-focus code-scanning camera provided by the embodiments of this application has the following technical effects: Through the code-scanning activation instruction, the collection of face video streams can be carried out within a preset range. By using an adjustable-focus camera, the quality of image collection is effectively ensured. Especially in different environments and distances, the camera can automatically adjust the focal length according to needs, improving the clarity and details of the image; by tracking and recognizing the face images in the face video stream, it can dynamically determine whether the target meets the preset recognition criteria. This not only ensures the adaptability of the system to dynamic changes but also can adjust the recognition strategy according to the real-time image, improving the recognition accuracy and efficiency; when the image quality does not reach the preset target, the acquisition parameters of the adjustable-focus camera are automatically adjusted based on the deviation amount between the image and the target. This adjustment ensures that the image can reach the best clarity, thereby reducing the risk of misrecognition caused by improper focal length or blurred image, and significantly improving the accuracy of the recognition result; the recognition image is rotated and scaled, and projected to a standard pose. This standardization process ensures that even when the face appears at different angles or poses, the image can still be aligned with the preset standard face model, thus improving the consistency and accuracy of recognition; after the image is projected to the standard pose, precise face comparison is carried out. In this way, the inconsistencies caused by angles and poses can be effectively eliminated, further improving the accuracy and stability of face recognition.

[0072] The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-angle face recognition method for an adjustable focus code scanning camera, characterized in that, The method includes: Obtaining a code scanning activation instruction, and collecting a face video stream within a preset range through an adjustable focus scanning camera; Tracking and identifying the face images in the face video stream, and determining whether the face recognition image reaches a preset recognition target; When it does not reach, adjusting the acquisition parameters of the adjustable focus code scanning camera according to the deviation amount between the face recognition image and the preset recognition target until the preset recognition target is met; Performing rotation and scaling operations on the face recognition image, projecting it to a standard pose for face comparison, and obtaining a face recognition result.

2. The multi-angle face recognition method for the adjustable focus code scanning camera according to claim 1, wherein Tracking and identifying the face images in the face video stream includes: Performing feature analysis on the collected face data to obtain multi-dimensional analysis features; Performing recognition sensitivity comparison for each dimension based on the multi-dimensional analysis features to obtain sensitive features for each dimension; Determining core recognition features and auxiliary recognition features according to the dimension recognition response coefficients of the sensitive features for each dimension; Tracking and identifying the face images in the face video stream based on the core recognition features and auxiliary recognition features.

3. The multi-angle face recognition method of the adjustable focus code scanning camera according to claim 2, characterized in that, The performing feature analysis on the collected face data to obtain multi-dimensional analysis features includes: Taking facial geometric features, texture features, motion features, skin features, and personalized features as analysis dimensions; Performing feature extraction on the collected face data item by item according to the analysis dimensions to obtain the multi-dimensional analysis features.

4. The multi-angle face recognition method for a focus-adjustable barcode scanning camera according to claim 2, characterized in that, Performing recognition sensitivity comparison for each dimension based on the multi-dimensional analysis features to obtain sensitive features for each dimension, including: Obtaining the inter-class difference and intra-class stability of each feature in the multi-dimensional analysis features, where the inter-class difference describes the difference degree of the same feature of different individuals, and the intra-class stability describes the feature volatility of the same individual under different conditions; Quantifying the recognition sensitivity of each feature in the multi-dimensional analysis features according to the inter-class difference and intra-class stability to obtain the sensitivity coefficients of each feature in the multi-dimensional analysis features; Performing sensitive feature comparison and determination according to the sensitivity coefficients to obtain the sensitive features for each dimension.

5. The multi-angle face recognition method of the adjustable focus code scanning camera according to claim 4, wherein Determining core recognition features and auxiliary recognition features according to the dimension recognition response coefficients of the sensitive features for each dimension, including: Calculating the recognition contribution degree and recognition speed of the sensitive features for each dimension through recognition sample data; Configuring the evaluation weight values of the recognition contribution degree and recognition speed, and performing recognition response coefficient operations according to the recognition contribution degree and recognition speed of the sensitive features for each dimension to obtain the recognition response coefficients of the sensitive features for each dimension; Comparing and screening the recognition response coefficients of the sensitive features for each dimension according to a feature screening threshold to determine the core recognition features and auxiliary recognition features.

6. The multi-angle face recognition method for an adjustable-focus code scanning camera according to claim 5, wherein Determining the core recognition features and auxiliary recognition features includes: Determining a feature screening threshold according to the recognition features of the recognition scenario, where the feature screening threshold includes a quantity threshold and a recognition response coefficient threshold; Respectively screening and sorting the sensitive features for each dimension according to the recognition response coefficient thresholds of the core recognition features and auxiliary recognition features, and then performing sequential screening based on the quantity threshold according to the sorting to obtain the core recognition features and auxiliary recognition features.

7. The multi-angle face recognition method for the adjustable focus code scanning camera according to claim 5, characterized in that Determining whether the face recognition image reaches a preset recognition target includes: Respectively extracting the recognition requirement parameters of the core recognition feature and the auxiliary recognition feature; Determining the recognition constraint coefficient according to the recognition influence of the core recognition feature and the auxiliary recognition feature; Setting the preset recognition target according to the recognition requirement parameter and its recognition constraint coefficient.

8. The multi-angle face recognition method for the adjustable-focus code scanning camera according to claim 7, characterized in that When it does not reach, adjusting the acquisition parameters of the focus-adjustable code scanning camera according to the deviation amount between the face recognition image and the preset recognition target, including: Obtaining the maximum parameter deviation amount according to the face recognition image and the preset recognition target; Taking the maximum parameter deviation amount as the adjustment target and performing adaptive focusing based on the acquisition parameter adjustment relationship of the focus-adjustable code scanning camera.

9. The multi-angle face recognition method for the adjustable focus code scanning camera according to claim 7, characterized in that, Adjusting the acquisition parameters of the focus-adjustable code scanning camera according to the deviation amount between the face recognition image and the preset recognition target, including: Obtaining the parameter deviation amount of each recognition feature according to the position distribution of the core recognition feature and the auxiliary recognition feature; Performing compensation partitioning according to the position distribution and the parameter deviation amount of the corresponding recognition feature; Based on the parameter deviation amount of each compensation partition, respectively performing acquisition parameter adjustment relationship operations to obtain the acquisition parameters of each compensation partition for partition adjustment.

10. The multi-angle face recognition method for the adjustable-focus code scanning camera according to claim 5, wherein Performing rotation and scaling operations on the face recognition image, projecting it to a standard pose for face comparison, and obtaining a face recognition result, including: Performing three-dimensional feature marking partitioning on the face according to the feature distribution position and distribution angle of the multi-dimensional analysis feature, and establishing a bottom-layer feature partition, where the marking line of the feature partition is the feature line of the analysis feature; Performing re-partitioning in the bottom-layer feature partition according to the feature distribution position and distribution angle of the core recognition feature and the auxiliary recognition feature, and constructing a face feature distribution map, where the face feature distribution map has core feature, auxiliary feature, and basic feature labels; Performing rotation and scaling operations on the face recognition image, aligning and projecting it with the face feature distribution map, identifying the matching feature partition, and sequentially performing feature comparison using the core recognition feature and the auxiliary recognition feature to obtain the face recognition result.

Images