Intelligent camera three-dimensional face recognition method and system based on 3D modeling

Abstract

The invention provides an intelligent camera three-dimensional face recognition method and system based on 3D modeling, and belongs to the technical field of face recognition, multi-angle face image data and synchronous physiological signal data of a target object are collected through an intelligent camera, and a corresponding data set is generated. And based on the dynamic physiological signal set, constraining the multi-angle face image data set, and constructing a three-dimensional face model with physiological feature marks. Extracting a three-dimensional face feature point set fused with physiological features from the three-dimensional face model, wherein the three-dimensional face feature point set comprises geometric coordinates and physiological signal change parameters; and performing hierarchical matching verification on the extracted three-dimensional facial feature point set and a preset standard set, generating a hierarchical matching verification result, and then outputting a face recognition result of the target object, including identity matching and physiological activity judgment conclusions, thereby improving the accuracy and safety of face recognition.

Description

Technical Field

[0001] This invention relates to the field of facial recognition technology, and more specifically, to a method and system for 3D facial recognition using a smart camera based on 3D modeling. Background Technology

[0002] In today's digital age, facial recognition technology is widely used in security, finance, social networking, and many other fields, making its accuracy and security crucial. Traditional facial recognition methods primarily rely on the analysis of two-dimensional images. They extract geometric features from facial images, such as eye spacing and nose shape, and compare them with preset standard features to achieve identity verification. However, two-dimensional facial recognition technology has several limitations. Firstly, two-dimensional images are easily affected by factors such as shooting angle, lighting conditions, and changes in facial expressions, leading to decreased recognition accuracy. For example, under different lighting conditions, the brightness and contrast of facial images can change significantly, making the extracted features unstable and affecting the recognition results. Secondly, two-dimensional facial recognition technology struggles to effectively prevent forgery attacks, such as using photos or videos to forge facial information for deception. This is because such forged information has a certain similarity to real facial images on a two-dimensional plane, making it difficult for traditional methods to accurately distinguish them.

[0003] In addition, some existing technologies attempt to introduce physiological signals to enhance the security of facial recognition, but most of them simply combine physiological signals with two-dimensional images, failing to fully utilize the inherent relationship between physiological signals and the three-dimensional structure of the face, and thus cannot fully and accurately reflect the true characteristics of the target object. The accuracy of recognition and anti-counterfeiting capabilities still need to be improved. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for intelligent camera stereoscopic face recognition based on 3D modeling, the method comprising: The system collects multi-angle facial image data and synchronous physiological signal data of the target object through a smart camera, and generates a multi-angle facial image data set and a dynamic physiological signal set. The multi-angle facial image data set includes facial images from different shooting angles, and the dynamic physiological signal set includes facial micro-movement signals and skin blood flow signals. Based on the dynamic physiological signal set constraining the multi-angle face image data set, a three-dimensional face model with physiological feature markings is constructed. The three-dimensional face model with physiological feature markings simultaneously carries the three-dimensional geometric structure of the face and dynamic physiological feature information. Extract a set of three-dimensional facial feature points that incorporate physiological features from the three-dimensional face model with physiological feature markers. The set of three-dimensional facial feature points that incorporate physiological features includes both geometric coordinate information and parameters related to physiological signal changes. A hierarchical matching verification is performed on the three-dimensional facial feature point set that integrates physiological features and a preset standard three-dimensional facial feature point set to generate a hierarchical matching verification result. The preset standard three-dimensional facial feature point set contains standard physiological feature parameters. The face recognition result of the target object is output based on the hierarchical matching verification result. The face recognition result includes both the identity matching conclusion and the physiological activity determination conclusion.

[0005] Furthermore, embodiments of the present invention also provide a 3D modeling-based intelligent camera stereoscopic face recognition system, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described 3D modeling-based intelligent camera stereo face recognition method by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, the processor of the 3D modeling-based intelligent camera stereo face recognition system reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the 3D modeling-based intelligent camera stereo face recognition system to perform the above-described 3D modeling-based intelligent camera stereo face recognition method.

[0007] Based on the above, a smart camera simultaneously collects multi-angle facial image data and synchronous physiological signal data of the target object, generating a corresponding dataset. This obtains the geometric information and dynamic physiological features of the target object's face from different angles. The multi-angle facial image dataset is constrained based on the dynamic physiological signal dataset to construct a 3D facial model with physiological feature markers. This model integrates the 3D geometric structure of the face with dynamic physiological feature information, providing a more realistic and comprehensive reflection of the target object's facial features. Furthermore, the 3D facial feature point set, extracted from this 3D facial model and incorporating physiological features, contains both geometric coordinate information and parameters related to physiological signal changes. This allows for hierarchical matching verification between the 3D facial feature point set and a pre-defined standard 3D facial feature point set. This enables multi-level comparison of the target object's facial features, improving the accuracy of identity matching, effectively determining the target object's physiological activity, and preventing forgery attacks. This significantly enhances the accuracy, reliability, and security of facial recognition. Attached Figure Description

[0008] Figure 1This is a schematic diagram of the execution flow of the intelligent camera stereo face recognition method based on 3D modeling provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of a 3D modeling-based intelligent camera stereo face recognition system provided in an embodiment of the present invention. Detailed Implementation

[0010] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a 3D modeling-based intelligent camera stereo face recognition method according to an embodiment of the present invention. The following is a detailed description of this 3D modeling-based intelligent camera stereo face recognition method.

[0011] Step S110: Collect multi-angle facial image data and synchronous physiological signal data of the target object through a smart camera, and generate a multi-angle facial image data set and a dynamic physiological signal set. The multi-angle facial image data set includes facial images under different shooting angles, and the dynamic physiological signal set includes facial micro-movement signals and skin blood flow signals.

[0012] In this embodiment, the smart camera employs a multi-lens array design, comprising multiple camera modules facing different directions. These camera modules can simultaneously capture images of the target object's face from different angles. During the acquisition process, the smart camera's control module sends synchronization trigger signals to each camera module, ensuring that all cameras start and stop image acquisition at the same time, thereby guaranteeing the time synchronization of multi-angle facial image data. Each camera module records corresponding shooting angle parameters, including horizontal rotation angle and vertical pitch angle, when acquiring images; these parameters are stored along with the acquired image data.

[0013] Meanwhile, the smart camera integrates a physiological signal acquisition module, which may include a miniature infrared sensor and an accelerometer. The miniature infrared sensor is used to acquire skin blood flow signals from the target object's face, obtaining blood flow signals by emitting infrared light of a specific wavelength and receiving changes in the intensity of reflected light. The accelerometer is used to capture micro-motion signals of the face, detecting acceleration changes caused by minute contractions and relaxations of facial muscles. The physiological signal acquisition module maintains strict time synchronization with the image acquisition module, and its sampling frequency matches the frame rate of image acquisition to ensure a one-to-one correspondence between physiological signal data and image data in the time dimension.

[0014] Before data collection begins, the smart camera guides the target subject, prompting them to keep their face relaxed and avoid deliberate facial expressions or large head movements to ensure the stability and validity of the collected data. During the collection process, the multi-angle facial image data set is stored in the form of an image frame sequence, with each frame containing pixel color information and corresponding shooting angle parameters; the dynamic physiological signal set is stored in the form of time-series data, where facial micro-motion signals contain the amplitude and direction of movement information at different time points, and skin blood flow signals contain blood flow intensity information at different time points.

[0015] Before performing data collection, the smart camera system displays a data collection authorization agreement to the target user through a front-facing interactive interface. The agreement clearly states the type of data to be collected (including multi-angle facial images, facial micro-motion signals, and skin blood flow signals), the purpose of collection (only for 3D facial recognition and physiological activity determination), the data storage period (automatic desensitization within 24 hours after matching), and the scope of data use (limited to local device processing; raw data will not be transmitted to third parties). The target user must confirm authorization by clicking the "Agree to Authorization" button on the touchscreen. The system simultaneously records the authorization process as video evidence (the video is only retained for 72 hours for compliance auditing).

[0016] To protect privacy-sensitive data, a triple-layer protection mechanism can be implemented: First, during image data acquisition, differential blurring of the eye area is automatically applied (preserving iris texture but concealing pupil details). Second, physiological signal data undergoes real-time conversion using a homomorphic encryption algorithm (encrypted data can be directly used for computation but cannot be used to deduce the original signal). Third, all stored data is associated with a unique temporary identity identifier (TID), which is isolated and mapped from the target object's real identity information through an independent key pool. When an unauthorized access attempt is detected, the system will automatically trigger a data destruction mechanism, deleting all temporarily cached original acquired data.

[0017] During data collection, the authorization status of the target object is continuously monitored. If the target object's head leaves the collection area for more than 3 seconds or the interactive interface is actively closed, the collection process is immediately terminated and some cached data is cleared. After collection is completed, a data collection list is displayed to the target object, containing metadata such as collection duration and sampling frequency for each type of data. The target object can trigger the data update process through the "Re-collect" button, and the original data will be overwritten immediately after the new data collection is completed.

[0018] Step S120: Based on the dynamic physiological signal set constraining the multi-angle face image data set, construct a three-dimensional face model with physiological feature markings. The three-dimensional face model with physiological feature markings simultaneously carries the three-dimensional geometric structure of the face and dynamic physiological feature information.

[0019] Step S121: Extract facial region contour information of each image frame from the multi-angle face image data set, and at the same time separate facial micro-motion signals and skin blood flow signals from the dynamic physiological signal set. Associate the temporal variation law of facial micro-motion signals with the spatial morphology of facial region contour information to generate a first association mapping table. The first association mapping table is used to record the correspondence between facial contour morphology and micro-motion amplitude at different time points.

[0020] First, facial region detection is performed on each frame of the multi-angle face image dataset using a deep learning-based face detection algorithm to locate the facial regions in the image. Then, contour extraction is performed on the detected facial regions by processing the image using an edge detection operator to obtain the contour pixel coordinates of the facial regions. These contour pixel coordinates are then concatenated to form facial region contour information, which includes the spatial morphological features of the facial contour.

[0021] Simultaneously, signal separation processing is performed on the dynamic physiological signal set. Since facial micro-motion signals and skin blood flow signals differ in frequency and amplitude characteristics, a bandpass filter is used to separate the signals in different frequency bands. Facial micro-motion signals are mainly concentrated in the lower frequency band, while skin blood flow signals are in the relatively higher frequency band. By setting an appropriate filter cutoff frequency, the two can be effectively separated.

[0022] Next, the temporal variation patterns of facial micro-motion signals are correlated with the spatial morphology of facial contour information. Temporal analysis is performed on the facial micro-motion signals to extract their motion amplitude values ​​at different time points, forming a micro-motion amplitude sequence. Simultaneously, feature extraction is performed on the facial contour information of each image frame, representing the contour morphology with a set of feature parameters, such as the coordinates of key points on the contour and changes in contour curvature. Then, the micro-motion amplitude values ​​at the same time point are mapped to the corresponding facial contour feature parameters, generating a first association mapping table. In the first association mapping table, each time point corresponds to one record, containing the facial contour morphology feature parameters and micro-motion amplitude values ​​at that time point.

[0023] Step S122: Perform spatial distribution inversion processing on the skin blood flow signal, combine the shooting angle parameters of the smart camera to determine the blood flow signal intensity distribution pattern in different areas of the face, and generate a blood flow signal spatial distribution map. The blood flow signal spatial distribution map marks the blood flow signal intensity range of each sub-region of the face, including the forehead, around the eyes, cheeks, and around the lips.

[0024] First, the skin blood flow signal is preprocessed, including noise removal and baseline drift reduction. An adaptive filtering algorithm is used to filter the raw skin blood flow signal to eliminate the influence of environmental interference and physiological noise. Baseline correction is performed on the signal using a polynomial fitting method to remove baseline drift components.

[0025] Then, spatial distribution inversion processing is performed. Skin blood flow signals are acquired through multiple sampling points, each corresponding to a specific location on the face. Based on the shooting angle parameters of the smart camera and the spatial location information of the sampling points, a mapping relationship between the sampling points and the three-dimensional spatial coordinates of the face is established. Using an interpolation algorithm, the blood flow signal intensity values ​​at other locations on the face are inferred from the known blood flow signal intensity values ​​at the sampling points, thus obtaining the spatial distribution of blood flow signals across the entire face.

[0026] By combining the shooting angle parameters of the smart camera, the spatial distribution of the retrieved blood flow signal is corrected. Different shooting angles may cause deviations in the projection position of sampling points on the face. Through coordinate transformation, the coordinates of sampling points at different angles are unified into a standard facial coordinate system to eliminate the influence of shooting angle on spatial distribution.

[0027] Finally, the distribution patterns of blood flow signal intensity in different facial regions were determined, generating a spatial distribution map of blood flow signals. The face was divided into sub-regions such as the forehead, around the eyes, cheeks, and around the lips. Statistical analysis was performed on the blood flow signal intensity values ​​within each sub-region to determine its intensity range. In the spatial distribution map of blood flow signals, different colors or gray levels were used to represent different blood flow signal intensity ranges, and the location of each sub-region and its corresponding intensity range were marked on the map.

[0028] Step S123: Based on the first association mapping table, the image frames in the multi-angle face image data set are filtered in the time dimension, retaining image frames with continuous micro-motion signals and no sudden changes in contour shape, and removing image frames with interrupted micro-motion signals or abnormal contour shape, generating a set of effective image frames with continuous time. The time interval of each frame in the set of effective image frames with continuous time is consistent with the sampling interval of the micro-motion signal.

[0029] First, the time series of micro-motion signals and the corresponding facial contour feature parameter sequences are extracted from the first association mapping table. Continuity analysis is performed on the time series of micro-motion signals to check for signal interruptions. If, within a certain time interval, the amplitude value of the micro-motion signal suddenly becomes zero or exhibits abnormal jumps, it is determined that the micro-motion signal is interrupted within that time interval.

[0030] Simultaneously, abrupt change detection is performed on the facial contour morphology feature parameter sequence. The difference value of the contour morphology feature parameters at adjacent time points is calculated. If the difference value exceeds a preset threshold, abrupt change in the contour morphology is determined. The preset threshold is obtained based on statistical analysis of a large amount of normal facial image data and represents the maximum allowable range of variation in facial contour morphology under normal circumstances.

[0031] Based on the results of micro-motion signal continuity analysis and contour morphology abrupt change detection, image frames in the multi-angle face image dataset are filtered. Image frames with continuous micro-motion signals and no abrupt changes in contour morphology are retained, while image frames with interrupted micro-motion signals or abnormal contour morphology are discarded. The filtered image frames are arranged in chronological order to generate a set of temporally continuous valid image frames. Since the sampling interval of the micro-motion signal is fixed, to ensure that the time interval between each frame in the set of temporally continuous valid image frames is consistent with the sampling interval of the micro-motion signal, the time interval between adjacent image frames must be equal to the sampling interval of the micro-motion signal during the filtering process. If there is a mismatch in time intervals, interpolation or discarding is required.

[0032] Step S124: Perform depth information reconstruction on each frame of the time-continuous valid image frame set, introduce the blood flow intensity parameter in the blood flow signal spatial distribution map to perform depth calculation, set the region where the blood flow signal intensity value is in the preset high intensity range as the depth calculation priority region, prioritize the acquisition of pixel depth values ​​in the depth calculation priority region, and generate a depth image with blood flow priority marking. In the depth image with blood flow priority marking, the depth value accuracy of the region where the blood flow signal intensity value is in the preset high intensity range is higher than that of other regions.

[0033] First, each frame in the time-continuous set of valid image frames undergoes preprocessing, including image denoising, contrast enhancement, and distortion correction. Gaussian filtering is used to denoise the images, removing Gaussian noise. Histogram equalization is employed to enhance image contrast and improve detail. Finally, distortion correction is performed based on the intrinsic parameters and distortion coefficients of the smart camera to eliminate the impact of lens distortion on the image.

[0034] Then, depth information reconstruction is performed. A stereo matching algorithm is used to calculate the depth value of a pixel using image pairs captured from multiple angles. For each frame, other images with different shooting angles are selected as matching image pairs. By calculating the disparity of corresponding pixels in the image pair, and combining parameters such as the baseline distance and focal length of the smart camera, the depth value of the pixel is calculated based on the principle of triangulation.

[0035] In the depth calculation process, blood flow intensity parameters from the spatial distribution map of blood flow signals are introduced. The spatial distribution map of blood flow signals is registered with the currently processed image, ensuring a one-to-one correspondence between the spatial distribution of blood flow signals and facial regions in the image. Then, regions with blood flow signal intensity values ​​in a preset high-intensity range are designated as depth calculation priority regions. During stereo matching, pixels in the depth calculation priority regions are matched and depth values ​​are calculated first, employing a more refined matching strategy and higher matching accuracy to obtain more accurate depth values. For pixels in other regions, a conventional matching strategy is used for depth calculation.

[0036] Finally, a depth image with blood flow priority markers is generated. This depth image includes not only the depth value of each pixel but also markers for regions where blood flow signal intensity values ​​fall within a preset high-intensity range. Furthermore, because a more precise calculation method is used for the depth priority regions, the depth values ​​in these regions are more accurate than those in other regions.

[0037] Step S125: Based on the shooting angle parameters of the time-continuous effective image frame set, establish a multi-angle coordinate transformation matrix, and uniformly transform the coordinates of the depth images with blood flow priority markings in each frame to a preset three-dimensional coordinate system, so that the depth images at different angles form complementary coverage in three-dimensional space, generating a unified set of multi-angle depth image coordinates, wherein the origin of the coordinates of each depth image in the unified set of multi-angle depth image coordinates is completely consistent with the direction of the coordinate axis.

[0038] Step S1251: Extract the shooting angle parameters of each frame in the set of valid image frames with continuous time from the parameter configuration file of the smart camera, including horizontal rotation angle, vertical pitch angle and lens focal length. Organize the horizontal rotation angle, vertical pitch angle and lens focal length according to the acquisition order of the image frames to generate a frame-angle parameter correspondence table. In the frame-angle parameter correspondence table, each image frame number is associated with a complete set of shooting angle parameters.

[0039] The parameter configuration file for the smart camera is stored in the camera's local storage module. This configuration file contains detailed parameter information for each camera module. When acquiring shooting angle parameters, the configuration file is read, and the corresponding horizontal rotation angle, vertical pitch angle, and lens focal length parameters are extracted based on the camera module number corresponding to each frame in the set of valid images in a continuous time frame. These parameters are then arranged according to the image frame acquisition order, and a corresponding set of shooting angle parameters is assigned to each image frame number, thereby generating a frame-angle parameter mapping table. The frame-angle parameter mapping table can be stored in tabular form, where each row represents an image frame number and each column represents a shooting angle parameter.

[0040] Step S1252: Determine the parameters of the preset three-dimensional coordinate system, including the position of the coordinate origin, the positive direction definition of the horizontal axis, vertical axis, and optical axis. The coordinate origin is set as the center position of the intelligent camera's acquisition area, the horizontal axis is along the horizontal direction, the vertical axis is along the vertical direction, and the optical axis is along the lens optical axis direction.

[0041] The preset 3D coordinate system is a right-handed coordinate system, with the origin set at the center of the smart camera's acquisition area. This center position is determined by measuring the physical structure of the smart camera. The horizontal axis (X-axis) is along the horizontal direction, with the positive direction to the right; the vertical axis (Y-axis) is along the vertical direction, with the positive direction upwards; and the optical axis (Z-axis) is along the lens's optical axis, with the positive direction pointing towards the target object. These definitions allow the 3D coordinate system to accurately describe the position and posture of the face in space.

[0042] Step S1253: For each frame of depth image with blood flow priority marking, construct a coordinate transformation matrix based on the shooting angle parameters corresponding to the frame of depth image with blood flow priority marking. First, construct a rotation matrix based on the horizontal rotation angle and the vertical pitch angle, then construct a scaling matrix based on the lens focal length, and finally construct a translation matrix based on the position of the image frame in the acquisition area. Combine the rotation matrix, scaling matrix, and translation matrix in sequence to obtain the multi-angle coordinate transformation matrix of the frame of depth image with blood flow priority marking.

[0043] For each frame of depth image marked with blood flow priority, the corresponding horizontal rotation angle, vertical pitch angle, and lens focal length parameters are obtained from the frame-angle parameter mapping table. First, a rotation matrix is ​​constructed based on the horizontal rotation angle and the vertical pitch angle. The horizontal rotation angle corresponds to the rotation around the Y-axis, and the vertical pitch angle corresponds to the rotation around the X-axis. The rotation matrices for these two rotations are calculated separately, and then they are multiplied together to obtain the total rotation matrix.

[0044] Next, a scaling matrix is ​​constructed based on the lens focal length. The scaling matrix is ​​used to convert image coordinates into actual three-dimensional spatial coordinates. Its scaling factor is related to the lens focal length; the larger the focal length, the smaller the scaling factor, and vice versa.

[0045] Then, a translation matrix is ​​constructed based on the position of the image frame within the acquisition area. This translation matrix is ​​used to translate the rotated and scaled coordinates so that their origin coincides with the origin of the preset 3D coordinate system. The position of the image frame within the acquisition area can be determined using the physical layout parameters of the smart camera.

[0046] Finally, the rotation, scaling, and translation matrices are multiplied sequentially to obtain the multi-angle coordinate transformation matrix of the depth image with blood flow priority markings in that frame. The matrix combination order is: scaling transformation first, then rotation transformation, and finally translation transformation.

[0047] Step S1254: Combine the two-dimensional coordinates of each pixel in the depth image marked with blood flow priority in each frame with the depth value to form the original three-dimensional coordinates of the pixel.

[0048] In a depth image with blood flow priority marking, each pixel contains two-dimensional image coordinates (u, v) and a corresponding depth value d. The two-dimensional image coordinates are converted to homogeneous coordinates (u, v, 1), and then multiplied by the depth value d to obtain the three-dimensional coordinates (u*d, v*d, d) of the pixel in the camera coordinate system, which are the original three-dimensional coordinates.

[0049] Step S1255: Substitute the original three-dimensional coordinates of each pixel into the multi-angle coordinate transformation matrix corresponding to the depth image of the frame to which the pixel belongs with blood flow priority marking, and convert it into the target three-dimensional coordinates in the preset three-dimensional coordinate system.

[0050] The original 3D coordinates are represented as column vectors, and then multiplied by a multi-angle coordinate transformation matrix to obtain the transformed target 3D coordinates. Matrix multiplication is performed according to the rules of matrix multiplication in linear algebra. Through the above transformation, the coordinates of the pixels are transformed from the camera coordinate system to the preset 3D coordinate system.

[0051] Step S1256: Remove pixel coordinate data that exceeds the preset effective range of the three-dimensional coordinate system after transformation, and retain pixel coordinate data within the effective range.

[0052] The preset effective range of the 3D coordinate system is determined based on the average size of a human face and its possible range of motion. For example, the value ranges of the X, Y, and Z axes are each set to a certain interval. For the transformed target 3D coordinates, it is checked whether the value of each coordinate axis is within the preset effective range. If the value of a coordinate axis exceeds the range, the coordinate data of that pixel is discarded; otherwise, the coordinate data of that pixel is retained.

[0053] Step S1257: Organize the effective pixel coordinate data of each frame of depth image with blood flow priority mark by frame number to form a single frame coordinate transformation result. The single frame coordinate transformation result includes the target three-dimensional coordinates of all effective pixels of the depth image with blood flow priority mark in that frame and the corresponding blood flow priority mark.

[0054] The effective pixel coordinate data after depth image transformation for each frame with blood flow priority markers are organized and stored in order of frame number. Each single-frame coordinate transformation result contains a frame number, the target 3D coordinates of all effective pixels in that frame image, and the corresponding blood flow priority markers. The blood flow priority markers indicate whether the pixel belongs to a region where the blood flow signal intensity value is in a preset high-intensity range.

[0055] Step S1258: Combine the single-frame coordinate transformation results of all frames, correct the image frame coordinate transformation results with coverage gaps, so that all facial areas are covered by coordinate data, and generate a unified set of multi-angle depth image coordinates. In the unified set of multi-angle depth image coordinates, the coordinate data of the depth images of each frame marked with blood flow priority are sorted according to the coordinate values ​​of the preset three-dimensional coordinate system.

[0056] The single-frame coordinate transformation results of all frames are combined to obtain a set containing the coordinate data of all valid pixels. This set is then analyzed to check for any gaps in facial area coverage. If gaps exist, they are corrected by interpolating the coordinate data of adjacent frames to fill in the missing coordinates. Finally, all coordinate data are sorted according to the coordinate values ​​in a preset 3D coordinate system to generate a unified set of multi-angle depth image coordinates. The sorting can be performed sequentially along the X-axis, Y-axis, and Z-axis, first sorting by X-axis coordinate values ​​from smallest to largest, then for points with the same X-axis coordinate values, sorting by Y-axis coordinate values ​​from smallest to largest, and so on.

[0057] Step S126: Based on the unified set of multi-angle depth image coordinates, construct an initial three-dimensional face geometric framework. Refer to the first association mapping table to perform framework construction, and convert the contour change corresponding to the micro-motion signal into dynamic adjustment parameters of the geometric framework, so that the initial three-dimensional face geometric framework can reflect the geometric shape changes caused by facial micro-motion, and generate a three-dimensional geometric framework with micro-motion dynamic parameters. In the three-dimensional geometric framework with micro-motion dynamic parameters, each geometric vertex is associated with a corresponding micro-motion amplitude parameter.

[0058] First, the 3D coordinate data of all pixels are extracted from a unified set of multi-angle depth image coordinates. Then, a 3D reconstruction algorithm is used to construct an initial 3D face geometric framework. The 3D reconstruction algorithm can employ a point cloud-based approach, using the 3D coordinates of all pixels as point cloud data, and then constructing the 3D geometric surface of the face through surface fitting.

[0059] Referring to the first association mapping table, the contour changes corresponding to facial micro-motion signals are extracted. Based on the facial contour morphological feature parameters and micro-motion amplitude values ​​at different time points in the first association mapping table, a relationship model between contour changes and micro-motion amplitude is calculated. This relationship model is applied to the initial 3D facial geometric framework, transforming the contour changes corresponding to micro-motion signals into dynamic adjustment parameters for the geometric framework. These dynamic adjustment parameters can be the displacement and rotation angle of geometric vertices, etc. By adjusting these parameters, the initial 3D facial geometric framework can undergo corresponding morphological changes with facial micro-motions.

[0060] When generating a 3D geometric frame with micro-motion dynamic parameters, a corresponding micro-motion amplitude parameter is associated with each geometric vertex. Based on the distribution of micro-motion amplitude in the first association mapping table, the micro-motion amplitude parameter is assigned to each vertex of the 3D geometric frame, so that the micro-motion amplitude parameter of each vertex corresponds to the facial micro-motion at its location.

[0061] Step S127: Extract facial texture information from each frame of the time-continuous valid image frame set. Classify the texture information according to the blood flow signal spatial distribution map based on the blood flow intensity region. Using the average blood flow signal intensity value of the facial region as a benchmark, the region with a blood flow signal intensity value higher than the average value is classified as the first blood flow intensity region, and the region with a blood flow signal intensity value lower than the average value is classified as the second blood flow intensity region. Process the texture information of the first blood flow intensity region using a first compression algorithm, and process the texture information of the second blood flow intensity region using a second compression algorithm. The compression ratio of the first compression algorithm is lower than that of the second compression algorithm. Generate a classified facial texture set. In the classified facial texture set, the texture detail resolution of different blood flow intensity regions varies.

[0062] First, facial texture information is extracted from each frame of a time-continuous set of valid image frames. Facial texture information includes features such as pixel color, brightness, and texture pattern, which can be obtained by performing operations such as color space conversion and texture feature extraction on the image. For example, the image is converted from the RGB color space to the HSV color space, and the hue, saturation, and brightness components are extracted as part of the texture information; at the same time, methods such as gray-level co-occurrence matrix are used to extract the texture feature parameters of the image.

[0063] Based on the spatial distribution map of blood flow signals, the average blood flow signal intensity value of the facial region is calculated. The blood flow signal intensity values ​​of each sub-region of the face are then averaged to obtain the average blood flow signal intensity value for the entire facial region. Regions with blood flow signal intensity values ​​higher than the average are designated as the first blood flow intensity region, and regions with values ​​lower than the average are designated as the second blood flow intensity region.

[0064] The texture information of the first and second blood flow intensity regions is processed using different compression algorithms. The first compression algorithm uses a lower compression ratio to retain more texture details; for example, the JPEG2000 compression algorithm can be used, with higher image quality parameters set. The second compression algorithm uses a higher compression ratio to reduce the amount of data while preserving basic texture features; for example, the JPEG compression algorithm can be used, with lower image quality parameters set.

[0065] After compression, the texture information from different blood flow intensity regions is integrated to generate a categorized facial texture set. In this set, the texture detail resolution of the first blood flow intensity region is higher, while the texture detail resolution of the second blood flow intensity region is lower, thus achieving differentiated processing of texture information based on blood flow intensity.

[0066] Step S128: Map the classified facial texture set onto the surface of the three-dimensional geometric frame with micro-motion dynamic parameters, so that the resolution of the texture details and the depth value accuracy corresponding to the blood flow priority marker are matched. The texture detail resolution of the first blood flow intensity region is matched with the depth value accuracy of the corresponding region. At the same time, the micro-motion signals and blood flow signals in the dynamic physiological signal set are embedded in the corresponding regions of the three-dimensional geometric frame in the form of markers to generate the final three-dimensional face model with physiological feature markers. In the three-dimensional face model with physiological feature markers, each facial region contains not only three-dimensional geometric shape and texture details, but also the corresponding micro-motion parameters and blood flow signal parameters.

[0067] The texture information from the categorized facial texture set is mapped onto the surface of a 3D geometric framework with micro-motion dynamic parameters according to the location of the facial regions. During the mapping process, it is ensured that the resolution of the texture details is adapted to the depth value accuracy corresponding to the blood flow priority markers; that is, the high-resolution texture of the first blood flow intensity region corresponds to the region with higher depth value accuracy, and the low-resolution texture of the second blood flow intensity region corresponds to the region with lower depth value accuracy.

[0068] Simultaneously, micro-motion signals and blood flow signals from the dynamic physiological signal set are embedded in the corresponding regions of the three-dimensional geometric framework in a labeled form. For micro-motion signals, the micro-motion amplitude parameters at each time point are associated with the corresponding geometric vertices in the three-dimensional geometric framework; for blood flow signals, the blood flow intensity interval information in the spatial distribution map of blood flow signals is associated with the facial sub-regions in the three-dimensional geometric framework. Through this embedding method, each facial region in the three-dimensional face model with physiological feature labels not only contains three-dimensional geometric shape and texture details, but also is associated with the corresponding micro-motion parameters and blood flow signal parameters.

[0069] Step S130: Extract a set of three-dimensional facial feature points that incorporate physiological features from the three-dimensional face model with physiological feature markers. The set of three-dimensional facial feature points that incorporate physiological features includes both geometric coordinate information and parameters related to physiological signal changes.

[0070] Step S131: Separate the three-dimensional geometric structure data and dynamic physiological feature data of the face from the three-dimensional face model with physiological feature markers. Divide the facial micro-motion signals and skin blood flow signals in the dynamic physiological feature data into facial sub-regions to generate a second association mapping table. The second association mapping table records the range of micro-motion amplitude and blood flow intensity corresponding to each sub-region of the forehead, around the eyes, cheeks, around the lips, and bridge of the nose.

[0071] Step S1311: Read the complete data file of the three-dimensional face model with physiological feature markings, and parse the three-dimensional geometric data segment and physiological feature data segment stored in the complete data file of the three-dimensional face model with physiological feature markings. The three-dimensional geometric data segment includes model vertex coordinates, face index, and texture coordinates. The physiological feature data segment includes facial micro-motion signal temporal data and skin blood flow signal spatial distribution data.

[0072] The complete data file of the 3D face model with physiological feature markers is stored in a specific format, which includes a file header, 3D geometric data segments, and physiological feature data segments. When reading the file, the file header is first parsed to obtain information such as the starting address and length of the data segments. Then, the 3D geometric data segments and physiological feature data segments are read separately based on this information. In the 3D geometric data segment, the model vertex coordinates are stored as floating-point numbers, with each vertex containing X, Y, and Z coordinate values; the facet index describes the triangular faces of the model surface, with each facet consisting of the indices of three vertices; and the texture coordinates map the texture image onto the model surface, with each vertex corresponding to a texture coordinate (u, v). The facial micro-motion signal temporal data in the physiological feature data segment includes time nodes and corresponding micro-motion amplitude values; the skin blood flow signal spatial distribution data includes sampling point coordinates and corresponding blood flow intensity values.

[0073] Step S1312: Group the model vertex coordinates in the three-dimensional geometric data segment according to the facial anatomy structure, determine the vertex coordinate range corresponding to each sub-region of forehead, around the eyes, cheeks, around the lips, and bridge of the nose, and generate a coordinate mapping table. The coordinate mapping table records the coordinate range of all vertices contained in each sub-region.

[0074] Based on knowledge of facial anatomy, the vertex coordinates of the model in the 3D geometric data segment are grouped into sub-regions such as forehead, around the eyes, cheeks, around the lips, and bridge of the nose. Through analysis of a large number of standard face models, the approximate coordinate range of each sub-region is determined. For example, the forehead sub-region is usually located in the upper part of the face, and its Y-axis coordinate value is relatively large; the area around the eyes sub-region surrounds the eyes, and its coordinate range is determined based on the position of the eyes. For each sub-region, a clustering algorithm is used to cluster the vertex coordinates belonging to that region, thereby determining the coordinate range of all vertices contained in that sub-region. The name of each sub-region and its corresponding coordinate range are recorded in a coordinate mapping table.

[0075] Step S1313: Extract facial micro-motion signal time-series data from the physiological feature data segment. Based on the spatial coordinates of the monitoring points, refer to the coordinate mapping table to allocate the micro-motion amplitude value of each monitoring point to the corresponding facial sub-region, and generate a time-series data set. In the time-series data set, each sub-region corresponds to a set of time-series data containing time nodes and micro-motion amplitude.

[0076] The facial micro-motion signal time-series data is read from the physiological feature data segment. This data includes the spatial coordinates of multiple monitoring points and the corresponding micro-motion amplitude time-series sequences. For each monitoring point, the facial sub-region to which the monitoring point belongs is determined based on its spatial coordinates and a coordinate mapping table. Then, the micro-motion amplitude value of the monitoring point is assigned to the corresponding sub-region according to time nodes, generating a time-series data set. Each sub-region in the time-series data set corresponds to a set of time-series data, and each set of time-series data includes a time node and the corresponding micro-motion amplitude value.

[0077] Step S1314: Extract the spatial distribution data of skin blood flow signals from the physiological feature data segment, and assign the blood flow intensity value of each sampling point to the corresponding facial sub-region by referring to the coordinate mapping table, thereby generating a spatial data set. In the spatial data set, each sub-region corresponds to a set of spatial data containing the coordinates of the sampling point and the blood flow intensity.

[0078] Similarly, spatial distribution data of skin blood flow signals is extracted from physiological feature data segments. This spatial distribution data includes the coordinates of multiple sampling points and their corresponding blood flow intensity values. Referring to a coordinate mapping table, the blood flow intensity value of each sampling point is assigned to a corresponding facial sub-region, generating a spatial dataset. Each sub-region in the spatial dataset corresponds to a set of spatial data, and each set of spatial data contains the coordinates of the sampling points and their corresponding blood flow intensity values.

[0079] Step S1315: Statistically analyze the micro-motion time series data of each sub-region in the time series data set to determine the range of micro-motion amplitude of the sub-region.

[0080] Statistical analysis is performed on the micro-motion time series data of each sub-region in the time series dataset to calculate the maximum and minimum values ​​of the micro-motion amplitude within that sub-region, thereby determining the range of micro-motion amplitude. For example, for the forehead sub-region, the micro-motion amplitude values ​​at all corresponding time points are traversed, and the maximum and minimum values ​​are found to obtain the range of micro-motion amplitude for the forehead sub-region.

[0081] Step S1316: Statistically analyze the blood flow intensity data of each sub-region in the spatial data set to determine the blood flow intensity range of the sub-region.

[0082] Similarly, statistical analysis is performed on the blood flow intensity data of each sub-region in the spatial dataset to calculate the maximum and minimum values ​​of blood flow intensity within that sub-region, thus determining the range of blood flow intensity. For example, for the cheek region, the blood flow intensity values ​​of all its sampling points are traversed to find the maximum and minimum values, thereby obtaining the range of blood flow intensity for the cheek region.

[0083] Step S1317: Associate the name, micro-motion amplitude range, and blood flow intensity range of each sub-region to form a second mapping association table. In the second mapping association table, the time node or number of sampling points of the data source is marked for the micro-motion amplitude range and blood flow intensity range of each sub-region.

[0084] A second association mapping table is formed by mapping the name, micromotion amplitude range, and blood flow intensity range of each sub-region one-to-one. In this table, the number of time points or sampling points from which the data for the micromotion amplitude range and blood flow intensity range of each sub-region are indicated to demonstrate the reliability and representativeness of the data. For example, the micromotion amplitude range of the forehead sub-region is indicated based on a certain number of time points, and the blood flow intensity range is indicated based on a certain number of sampling points.

[0085] Step S132: Based on the second association mapping table, mark the boundary coordinates of each sub-region in the three-dimensional geometric structure data of the three-dimensional face model with physiological feature markings, determine the geometric vertex distribution density in each sub-region, and for each sub-region, calculate the product of its micro-motion amplitude range and blood flow intensity range as the physiological signal change activity of the sub-region; set the region with physiological signal change activity higher than the average activity of all sub-regions as the first feature point extraction region, and set the region with physiological signal change activity lower than the average activity of all sub-regions as the second feature point extraction region, and generate a feature point extraction density distribution map, wherein the geometric vertex sampling interval of the first feature point extraction region is smaller than that of the second feature point extraction region.

[0086] Based on the names and coordinate ranges of each sub-region in the second association mapping table, the boundary coordinates of each sub-region are marked in the 3D geometric structure data of the 3D face model with physiological feature markers. The geometric vertex distribution density within each sub-region is calculated by counting the geometric vertices within the boundary coordinates.

[0087] For each subregion, the product of its micromotion amplitude range and blood flow intensity range is calculated, and this product is taken as the physiological signal activity level of that subregion. For example, if the micromotion amplitude range of a certain subregion is A to B and the blood flow intensity range is C to D, then the physiological signal activity level of that subregion is (BA) × (DC).

[0088] Calculate the average physiological signal activity level across all sub-regions. Sub-regions with physiological signal activity levels above this average are designated as the first feature point extraction region, and those below are designated as the second feature point extraction region. Based on the division of the first and second feature point extraction regions, generate a feature point extraction density distribution map. Different extraction regions are represented by different colors or markers in the map, and the sampling interval of the geometric vertices of each region is marked. The sampling interval of the first feature point extraction region is smaller than that of the second feature point extraction region.

[0089] Step S133: Based on the feature point extraction density distribution map, geometric vertices are extracted in the first feature point extraction region using a first sampling interval, and geometric vertices are extracted in the second feature point extraction region using a second sampling interval. The first sampling interval is smaller than the second sampling interval. Geometric vertices are extracted in the second feature point extraction region using sparse sampling. At the same time, the three-dimensional coordinate values ​​of each extracted vertex in the three-dimensional face model with physiological feature markers are recorded to generate an initial three-dimensional feature point coordinate set. In the initial three-dimensional feature point coordinate set, the number of feature points in the first feature point extraction region is greater than that in the second feature point extraction region.

[0090] Based on the sampling intervals marked in the feature point extraction density distribution map, geometric vertices are extracted in the first feature point extraction region according to the first sampling interval, and in the second feature point extraction region according to the second sampling interval. The sampling interval can refer to the distance between two adjacent extraction points in three-dimensional space. The first sampling interval is smaller, so more geometric vertices can be extracted in the first feature point extraction region; the second sampling interval is larger, so relatively fewer geometric vertices are extracted in the second feature point extraction region.

[0091] During the extraction process, all geometric vertices in the 3D geometric structure data of the 3D face model with physiological feature markers are traversed. For each vertex, its extraction region is determined, and then the sampling interval of that region determines whether to extract the vertex. If a vertex belongs to the first feature point extraction region and meets the sampling interval requirement, its 3D coordinates are extracted; similarly, vertices in the second feature point extraction region are processed in a similar way. The 3D coordinates of all extracted vertices are recorded to generate an initial set of 3D feature point coordinates. Since the first sampling interval is smaller than the second sampling interval, the number of feature points in the first feature point extraction region is greater than that in the second feature point extraction region in the initial set of 3D feature point coordinates.

[0092] Step S134: Extract physiological signal change parameters for each sub-region from the dynamic physiological feature data, including the temporal fluctuation frequency of the micro-motion signal and the intensity change period of the skin blood flow signal. Bind the temporal fluctuation frequency of the micro-motion signal and the intensity change period of the skin blood flow signal to the feature points of the corresponding sub-region in the initial three-dimensional feature point coordinate set, so that each feature point contains both three-dimensional coordinate values ​​and is associated with the physiological signal change parameters of the sub-region to which it belongs, and generate a feature point set with physiological parameter binding.

[0093] Physiological signal change parameters for each sub-region are extracted from dynamic physiological characteristic data. For micromotion signals, time-domain analysis methods such as Fourier transform are used to extract their temporal fluctuation frequency; for skin blood flow signals, their intensity change period is extracted by analyzing their time-series data.

[0094] The extracted micro-motion signal's temporal fluctuation frequency and the skin blood flow signal's intensity variation period are bound to feature points in corresponding sub-regions within the initial three-dimensional feature point coordinate set. Based on the sub-region to which a feature point belongs, physiological signal variation parameters of that sub-region are assigned to that feature point. This ensures that each feature point, in addition to its three-dimensional coordinate values, is also associated with physiological signal variation parameters such as the micro-motion signal's temporal fluctuation frequency and the skin blood flow signal's intensity variation period of its respective sub-region, thereby generating a feature point set bound with physiological parameters.

[0095] Step S135: Redundancy screening is performed on the feature point set bound to physiological parameters. The three-dimensional coordinate distance between adjacent feature points is compared with the similarity of physiological parameters. Feature points with more representative coordinates and physiological parameters are retained, and redundant feature points are removed to generate a deredundant feature point set.

[0096] Step S1351: Group all feature points in the feature point set bound with physiological parameters into facial sub-regions, and sort the feature points in each sub-region by horizontal axis coordinate values ​​from smallest to largest to generate a sorted feature point list.

[0097] All feature points in the set of feature points bound to physiological parameters are grouped according to their respective facial sub-regions, such as the forehead group, the peri-eye group, and the cheek group. For each sub-region, the feature points are sorted in ascending order of their horizontal axis (X-axis) coordinate values, generating a sorted feature point list. This ensures that adjacent feature points are spatially close, facilitating subsequent redundancy analysis.

[0098] Step S1352: For each feature point in the sorted feature point list, select two adjacent feature points in the sorted order as a feature point to be analyzed, starting from the first feature point, until all adjacent feature point pairs in the sub-region are selected.

[0099] For each sorted list of feature points, starting from the first feature point, select two adjacent feature points in sequence as a pair of feature points to be analyzed. For example, if the sorted feature points are P1, P2, P3, and P4, then the pairs of feature points to be analyzed are (P1, P2), (P2, P3), and (P3, P4). This process is repeated to iterate through all adjacent pairs of feature points within the sub-region.

[0100] Step S1353: Compare the three-dimensional coordinate distance of each feature point to be analyzed, and at the same time analyze the similarity of the physiological parameters of each feature point to be analyzed.

[0101] For each feature point to be analyzed, calculate the three-dimensional coordinate distance between them. The three-dimensional coordinate distance can be calculated using the Euclidean distance formula, which is the square root of the sum of the squares of the subtraction of the X, Y, and Z coordinate values ​​of the two feature points.

[0102] Simultaneously, the similarity of physiological parameters between the feature point pairs to be analyzed is performed. Physiological parameters include the temporal fluctuation frequency of micromotion signals and the intensity variation period of skin blood flow signals. The degree of difference between these two parameters for each feature point is calculated. For example, the absolute or relative difference between the two parameters can be calculated, and the magnitude of the difference is used as an indicator of similarity; the smaller the difference, the higher the similarity.

[0103] Step S1354: Retrieve the distance standard and physiological parameter similarity standard of adjacent feature points from the preset screening parameter library.

[0104] The pre-defined screening parameter library stores distance standards and physiological parameter similarity standards for adjacent feature points. These standards were determined based on extensive experimental data and experience. The distance standard specifies the maximum allowable distance between adjacent feature points; if the three-dimensional coordinate distance between two feature points is less than this standard, they are considered to be relatively close in spatial location. The physiological parameter similarity standard specifies the maximum allowable difference in physiological parameters between adjacent feature points; if the difference in physiological parameters between two feature points is less than this standard, they are considered to have a high degree of physiological parameter similarity.

[0105] Step S1355: If the three-dimensional coordinate distance of the feature point pair to be analyzed meets the distance standard and the physiological parameter similarity of the feature point pair to be analyzed meets the similarity standard, then it is determined that there is redundancy in the feature point to be analyzed.

[0106] The 3D coordinate distance of the feature point pair to be analyzed is compared with a distance standard. If the 3D coordinate distance is less than or equal to the distance standard, and the physiological parameter difference of the feature point pair is less than or equal to the physiological parameter similarity standard, then the feature point to be analyzed is considered redundant. This is because the two feature points are close in spatial location and have similar physiological parameters, and the information they carry overlaps; only one of them needs to be retained.

[0107] Step S1356: For the redundant feature point pairs to be analyzed, analyze the coordinate representativeness and parameter representativeness of the two feature points to be analyzed. Retain the feature points whose coordinate representativeness and parameter representativeness are higher than the other feature point. The coordinate representativeness is judged by the degree of correlation between the feature point and the geometric center of the sub-region, and the parameter representativeness is judged by the degree of correlation between the physiological parameters of the feature point and the average physiological parameters of the sub-region.

[0108] For feature point pairs deemed redundant, further analysis of the coordinate representativeness and parameter representativeness of the two feature points is required. Coordinate representativeness is determined by the degree of correlation between the feature point and the geometric center of the sub-region; the smaller the distance from the feature point to the geometric center of the sub-region, the higher the coordinate representativeness. Parameter representativeness is determined by the degree of correlation between the physiological parameters of the feature point and the average physiological parameters of the sub-region; the smaller the difference between the physiological parameters of the feature point and the average physiological parameters of the sub-region, the higher the parameter representativeness.

[0109] Taking into account both coordinate representativeness and parameter representativeness, feature points whose combined score of coordinate representativeness and parameter representativeness is higher than that of the other feature point are retained. For example, weights can be assigned to coordinate representativeness and parameter representativeness respectively, and a combined score can be calculated; feature points with higher scores are retained.

[0110] Step S1357: After processing each sub-region, all retained feature points are summarized to form a set of redundant feature points. In the set of redundant feature points, the number of feature points in each sub-region is dynamically adjusted according to the size of the sub-region and the amplitude of physiological signal changes. The number of feature points retained in the first feature point extraction region is greater than that in the second feature point extraction region.

[0111] After redundancy filtering of the sorted feature point list for each sub-region, all retained feature points are aggregated to form a deredundant feature point set. In this set, the number of feature points in each sub-region is dynamically adjusted based on the size of the sub-region and the amplitude of physiological signal changes. Larger sub-regions with greater physiological signal changes retain a relatively larger number of feature points; conversely, smaller sub-regions retain a relatively smaller number of feature points. Since the physiological signal activity in the first feature point extraction region is higher than that in the second feature point extraction region, after redundancy removal, the first feature point extraction region still retains more feature points than the second feature point extraction region.

[0112] Step S136: Standardize the three-dimensional coordinate values ​​of each feature point in the set of redundant feature points. Using the geometric center of the three-dimensional face model with physiological feature markings as the origin, convert the coordinates of all feature points into offset coordinates relative to the geometric center to generate a set of standardized coordinate feature points.

[0113] First, the geometric center of the 3D face model with physiological feature markers is calculated. The coordinates of the geometric center (X0, Y0, Z0) are obtained by averaging the 3D coordinates of all vertices of the model.

[0114] Then, the 3D coordinates of each feature point in the redundant feature point set are standardized. The coordinates (X0, Y0, Z0) of the geometric center are subtracted from the coordinates (X, Y, Z) of each feature point to obtain the offset coordinates (X-X0, Y-Y0, Z-Z0) relative to the geometric center. These offset coordinates are used as the standardized coordinates of the feature points to generate a standardized coordinate feature point set. Standardization eliminates spatial differences in face positions, making feature points from different faces comparable.

[0115] Step S137: Supplement and improve the physiological parameters associated with each feature point in the standardized coordinate feature point set, extract the physiological signal change parameters of the corresponding sub-region from the dynamic physiological feature data, so that each feature point is completely associated with the micro-motion frequency and blood flow cycle parameters, and generate a feature point set with complete parameters.

[0116] Check whether the physiological parameters associated with each feature point in the standardized coordinate feature point set are complete. If the physiological parameters of some feature points are missing or incomplete, extract the physiological signal change parameters of the corresponding sub-region from the dynamic physiological feature data to supplement and improve them. For example, if the micromotor frequency parameter of a certain feature point is missing, extract the micromotor frequency parameter from the dynamic physiological feature data of the sub-region to which the feature point belongs and supplement it to the feature point. Ensure that each feature point is completely associated with physiological parameters such as micromotor frequency and blood flow cycle, generating a feature point set with complete parameters.

[0117] Step S138: The complete set of feature points is classified and organized according to facial sub-regions to form a forehead feature point subset, an eye periphery feature point subset, a cheek feature point subset, a lip periphery feature point subset, and a nose bridge feature point subset. All subsets are combined to form the final three-dimensional facial feature point set that integrates physiological features. In the three-dimensional facial feature point set that integrates physiological features, the feature points in each subset are sorted in ascending order of coordinate offset.

[0118] The complete set of feature points is categorized and organized according to facial sub-regions. Feature points belonging to the same sub-region are grouped into subsets, such as a forehead feature point subset, a feature point subset around the eyes, etc. Within each subset, feature points are sorted in ascending order of their coordinate offsets (i.e., standardized coordinates). The sorting can be performed sequentially according to the offsets along the X, Y, and Z axes, starting with sorting by X-axis offset from smallest to largest. For points with the same X-axis offset, then sorting by Y-axis offset from smallest to largest, and so on. All subsets are combined to form the final 3D facial feature point set incorporating physiological features.

[0119] Step S140: Perform hierarchical matching verification on the three-dimensional facial feature point set with fused physiological features and the preset standard three-dimensional facial feature point set to generate hierarchical matching verification results. The preset standard three-dimensional facial feature point set contains standard physiological feature parameters.

[0120] Step S141: Extract the standardized coordinate values ​​of all feature points from the three-dimensional facial feature point set that integrates physiological features to form a set of geometric coordinates to be matched; extract the standard coordinate values ​​of all feature points from the preset standard three-dimensional facial feature point set to form a set of standard geometric coordinates.

[0121] From the set of 3D facial feature points that incorporate physiological features, the standardized coordinate values ​​(i.e., offset coordinates relative to the geometric center) of each feature point are extracted. These coordinate values ​​are then combined to form a set of geometric coordinates to be matched. The set of geometric coordinates to be matched is a collection of multiple 3D coordinate points, each corresponding to the standardized coordinates of a feature point.

[0122] The pre-stored standard 3D facial feature point set contains 3D facial feature point information of known individuals. Standard coordinate values ​​of all feature points are extracted from this standard 3D facial feature point set to form a standard geometric coordinate set. The structure of the standard geometric coordinate set is similar to that of the geometric coordinate set to be matched; it is also a set containing multiple 3D coordinate points.

[0123] Step S142: Compare the coordinate differences of the feature points in the corresponding sub-regions of the set of geometric coordinates to be matched with the set of standard geometric coordinates, including the coordinate differences of each feature point in each sub-region of the forehead, around the eyes, cheeks, around the lips, and bridge of the nose, and generate a sub-region coordinate difference record, in which each element corresponds to the coordinate difference of a single feature point.

[0124] The set of geometric coordinates to be matched and the set of standard geometric coordinates are divided into facial sub-regions, resulting in a subset of coordinates to be matched and a subset of standard coordinates for each sub-region. For each sub-region, a one-to-one correspondence is established between feature points in the subset of coordinates to be matched and feature points in the subset of standard coordinates. This correspondence can be achieved through methods such as feature point indexing or spatial location matching.

[0125] For each corresponding pair of feature points, calculate their coordinate differences. Coordinate differences can be represented by calculating the Euclidean distance, which is the distance between the coordinates of the feature point to be matched and the coordinates of the standard feature point. Record the coordinate differences of all feature points within each sub-region to generate a sub-region coordinate difference record. The sub-region coordinate difference record can be stored in the form of a list or array, where each element corresponds to the coordinate difference value of a single feature point.

[0126] Step S143: Based on the coordinate difference record of the sub-region, evaluate the geometric matching degree of each sub-region. Combined with the preset geometric matching standard of the sub-region, if the geometric matching degree of all sub-regions meets the corresponding standard, it is determined that the preliminary matching of the geometric structure features is successful and a preliminary matching result is generated; if the geometric matching degree of at least one sub-region does not meet the corresponding standard, it is determined that the preliminary matching of the geometric structure features is unsuccessful and a preliminary matching failure result is generated.

[0127] Step S1431: For each sub-region in the sub-region coordinate difference record, extract the coordinate differences of all feature points in the sub-region to form a three-dimensional difference record of the sub-region.

[0128] From the sub-region coordinate difference record, extract the coordinate difference values ​​of all feature points within each sub-region to form a three-dimensional difference record for that sub-region. The three-dimensional difference record is a collection containing the coordinate difference values ​​of all feature points within that sub-region.

[0129] Step S1432: Based on the three-dimensional difference record of the sub-region, comprehensively evaluate the overall coordinate matching of feature points within the sub-region and determine the geometric matching degree of the sub-region.

[0130] Statistical analysis is performed on the 3D difference records of this sub-region to calculate statistical measures such as the mean and variance of the coordinate difference values, in order to comprehensively evaluate the overall coordinate matching of feature points within the sub-region. For example, the smaller the mean and variance of the coordinate difference values, the higher the geometric matching degree of the sub-region. These statistical measures can be mapped to a geometric matching degree index, which can range from 0 to 1, with values ​​closer to 1 indicating a higher matching degree.

[0131] Step S1433: Retrieve the geometric matching standard for each sub-region from the preset geometric matching standard library. In the preset geometric matching standard library, the geometric matching standard for different sub-regions is set according to the importance of the sub-region in face recognition. The difference in the geometric matching standard for the sub-regions around the eyes and lips is allowed to be smaller than that for the geometric matching standard for the forehead and cheeks.

[0132] A pre-defined geometric matching standard library stores the geometric matching standards for each sub-region. The geometric matching standards for different sub-regions are set according to their importance in face recognition. For example, the sub-regions around the eyes and lips are of high importance in face recognition; subtle changes in their geometric shape have a significant impact on the recognition results. Therefore, the allowable range of differences in the geometric matching standards for these sub-regions is relatively small. Conversely, sub-regions such as the forehead and cheeks are of relatively low importance, and their allowable range of differences is relatively large. The geometric matching standard for each sub-region is retrieved from the geometric matching standard library, such as the maximum allowable range of differences or the minimum value of the geometric matching degree index.

[0133] Step S1434: Compare the geometric matching degree of each sub-region with the corresponding sub-region geometric matching standard. If the geometric matching degree of a sub-region meets the corresponding sub-region geometric matching standard, then mark the sub-region geometric matching as passed; otherwise, mark the sub-region geometric matching as failed.

[0134] The geometric matching degree index of each sub-region is compared with the corresponding sub-region geometric matching standard. If the geometric matching degree index is greater than or equal to the minimum value specified in the standard, or the statistical value of the coordinate difference is less than or equal to the maximum value specified in the standard, the sub-region is marked as having passed geometric matching; otherwise, the sub-region is marked as having failed geometric matching.

[0135] Step S1435: Statistically analyze the labeling results of all sub-regions. If all sub-regions are labeled as having passed geometric matching, then the entire geometric structure feature is preliminarily matched.

[0136] The geometric matching results of all sub-regions are statistically analyzed. If all sub-regions are marked as having passed geometric matching, the entire geometric structure feature is determined to have passed preliminary matching. If at least one sub-region is marked as having failed geometric matching, the geometric structure feature is determined to have failed preliminary matching.

[0137] Step S1436: Generate a preliminary geometric matching result file, which includes the geometric matching degree of each sub-region, the corresponding geometric matching standard of the sub-region, and the matching mark result of each sub-region.

[0138] When the geometric structure features are determined to be initially matched, a preliminary geometric matching result file is generated. This result file contains the geometric matching degree index for each sub-region, the corresponding geometric matching standard for the sub-region (such as the allowable difference range or the minimum value of the geometric matching degree index), and the matching mark result (pass or fail) for each sub-region. The above information can be stored in the result file in the form of text or structured data for subsequent analysis and processing.

[0139] Step S1437: Organize the preliminary geometric matching result file so that all relevant data of the sub-regions have been recorded, and use the preliminary geometric matching result file as the final preliminary geometric matching result.

[0140] The preliminary geometric matching is processed through the result file to check whether the relevant data for all sub-regions has been correctly recorded, ensuring the completeness and accuracy of the data. If any data is missing or incorrect, it is supplemented and corrected. After processing, this result file is used as the final preliminary geometric matching result.

[0141] Step S144: When the geometric structure features are initially matched, extract the physiological signal change parameters associated with each feature point in the three-dimensional facial feature point set that integrates physiological features to form a set of physiological parameters to be matched; extract the standard physiological parameters associated with each feature point from the preset standard three-dimensional facial feature point set to form a set of standard physiological parameters.

[0142] Once the geometric structural features have passed the initial matching, the physiological signal change parameters associated with each feature point are extracted from the three-dimensional facial feature point set that integrates physiological features. These parameters include the temporal fluctuation frequency of micro-motion signals and the intensity change period of skin blood flow signals. The above parameters are then combined to form a set of physiological parameters to be matched.

[0143] Standard physiological parameters associated with each feature point are extracted from a pre-defined set of standard three-dimensional facial feature points. These standard physiological parameters are the physiological signal change parameters of known individuals. They are then combined to form a set of standard physiological parameters.

[0144] Step S145: Compare the physiological parameter differences of the corresponding sub-region feature points in the set of physiological parameters to be matched with those in the set of standard physiological parameters, including differences in micromotor frequency and blood flow cycle, and generate a sub-region physiological parameter difference record. In the sub-region physiological parameter difference record, each element corresponds to the physiological parameter difference of a single feature point.

[0145] The set of physiological parameters to be matched and the set of standard physiological parameters are divided into facial sub-regions, resulting in a subset of physiological parameters to be matched and a subset of standard physiological parameters for each sub-region. For each sub-region, a one-to-one correspondence is established between feature points in the subset of physiological parameters to be matched and feature points in the subset of standard physiological parameters.

[0146] For each corresponding pair of feature points, calculate their physiological parameter differences. Micromotor frequency differences can be represented by calculating the absolute or relative difference between the micromotor frequency of the feature point to be matched and the micromotor frequency of the standard feature point; blood flow cycle differences can be represented by calculating the absolute or relative difference between the blood flow cycle of the feature point to be matched and the blood flow cycle of the standard feature point. Record the physiological parameter differences of all feature points within each sub-region to generate a sub-region physiological parameter difference record. In the sub-region physiological parameter difference record, each element corresponds to the physiological parameter difference of a single feature point.

[0147] Step S146: Based on the physiological parameter difference records of the sub-regions, evaluate the consistency of physiological characteristics of each sub-region. Combined with the preset sub-region physiological matching standard, if the consistency of physiological characteristics of all sub-regions meets the corresponding standard, the physiological characteristic consistency verification is determined to be passed and a physiological verification pass result is generated; if the consistency of physiological characteristics of at least one sub-region does not meet the corresponding standard, the physiological characteristic consistency verification is determined to be failed and a physiological verification failure result is generated.

[0148] Statistical analysis was performed on the physiological parameter differences recorded for each sub-region, calculating the mean, variance, and other statistical measures of the physiological parameter differences to comprehensively assess the consistency of physiological characteristics of feature points within the sub-region. For example, the smaller the mean and variance of the physiological parameter differences, the higher the consistency of physiological characteristics of the sub-region.

[0149] The physiological matching criteria for each sub-region are retrieved from a pre-defined physiological matching criteria library. The physiological matching criteria for different sub-regions can be set according to the stability and importance of their physiological characteristics. The consistency assessment result of the physiological characteristics of each sub-region is compared with the corresponding physiological matching criteria. If the criteria are met, the consistency verification of the physiological characteristics of that sub-region is marked as passed; otherwise, it is marked as failed.

[0150] The labeling results of all sub-regions are statistically analyzed. If all sub-regions pass the verification, the physiological characteristic consistency verification is deemed successful, and a physiological verification success result is generated; otherwise, the verification is deemed unsuccessful, and a physiological verification failure result is generated.

[0151] Step S147: When the geometric structural features pass the initial matching and the physiological feature consistency verification passes, the hierarchical matching verification result with overall matching pass is generated by combining the geometric matching degree and the physiological feature consistency evaluation result. The hierarchical matching verification result with overall matching pass includes the matching details of each sub-region.

[0152] Step S1471: Extract the geometric matching degree and corresponding geometric matching standard of each sub-region from the preliminary geometric matching results, and determine the geometric matching contribution of each sub-region by combining the importance of the sub-region in face recognition.

[0153] From the preliminary geometric matching results, obtain the geometric matching degree index and corresponding geometric matching standard for each sub-region. Based on the importance of each sub-region in face recognition, assign an importance weight to each sub-region. The importance weight can be set based on experience or experimental data; for example, sub-regions around the eyes and lips have higher weights, while those around the forehead and cheeks have lower weights. Multiply the geometric matching degree index of each sub-region by its importance weight to obtain the geometric matching contribution of that sub-region.

[0154] Step S1472: Extract the physiological characteristic consistency assessment results and corresponding physiological matching criteria of each sub-region from the physiological verification pass results, and determine the physiological matching contribution of each sub-region by combining the physiological characteristic stability of the sub-region.

[0155] Similarly, the physiological characteristic consistency assessment results and corresponding physiological matching criteria for each sub-region are extracted from the physiological verification pass results. Based on the physiological characteristic stability of the sub-region, a stability weight is assigned to each sub-region. Sub-regions with higher physiological characteristic stability have higher weights, while those with lower stability have lower weights. The physiological characteristic consistency assessment result of each sub-region is multiplied by its stability weight to obtain the physiological matching contribution of that sub-region.

[0156] Step S1473: Retrieve the weight ratio of geometric matching contribution and the weight ratio of physiological matching contribution from the preset weight configuration file. The sum of the weight ratio of geometric matching contribution and the weight ratio of physiological matching contribution is 1.

[0157] The preset weight configuration file stores the weight ratios of geometric matching contribution and physiological matching contribution in the overall matching evaluation. For example, the weight ratio of geometric matching contribution is 0.6, the weight ratio of physiological matching contribution is 0.4, and the sum of the two is 1. These two weight ratios are retrieved from this file.

[0158] Step S1474: Combine the geometric matching contribution of each sub-region, the physiological matching contribution of each sub-region, the weighted proportion of the geometric matching contribution, and the weighted proportion of the physiological matching contribution to comprehensively evaluate the overall matching level.

[0159] Calculate the sum of the geometric matching contributions of all sub-regions and multiply it by the weighted proportion of the geometric matching contribution; calculate the sum of the physiological matching contributions of all sub-regions and multiply it by the weighted proportion of the physiological matching contribution; add these two products together to obtain the overall matching level evaluation value. The higher the overall matching level evaluation value, the higher the overall matching degree between the set of feature points to be matched and the standard set of feature points.

[0160] Step S1475: If the overall matching level is higher than the preset overall matching standard, then generate a hierarchical matching verification result indicating that the overall matching has passed.

[0161] The overall matching level assessment value is compared with a preset overall matching standard. If the assessment value is higher than the standard, a stratified matching verification result indicating that the overall matching has passed is generated. The overall matching standard can be a preset threshold.

[0162] Step S1476: In the hierarchical matching verification results where the overall matching has passed, record the geometric matching contribution of each sub-region, the physiological matching contribution of each sub-region, and the evaluation basis for the overall matching level.

[0163] In the hierarchical matching verification results where the overall matching is successful, the geometric matching contribution, physiological matching contribution, and the calculation process and basis of the overall matching level evaluation value of each sub-region are recorded in detail, including the weight ratios used, the importance weight of the sub-region, and the stability weight.

[0164] Step S1477: Organize the hierarchical matching verification results that have passed the overall matching to form the final hierarchical matching verification results that have passed the overall matching.

[0165] The hierarchical matching verification results that pass the overall matching are organized to ensure the accuracy and completeness of all data, forming the final result file.

[0166] Step S148: When the initial matching of geometric structural features fails or the consistency verification of physiological features fails, record the specific information of the failed sub-region, including the name of the failed sub-region, the corresponding difference and matching criteria, and generate a hierarchical matching verification result containing the reason for failure. In the hierarchical matching verification result containing the reason for failure, it is marked whether the matching of geometric structural features failed or the consistency verification of physiological features failed.

[0167] When preliminary matching of geometric structural features fails or consistency verification of physiological features fails, record the name of the failed sub-region, the differences in the sub-region (such as the statistical values ​​of coordinate difference values ​​or physiological parameter difference values), and the corresponding matching criteria. In the hierarchical matching verification results, indicate the type of failure, i.e., whether it was a failure in geometric structural feature matching or a failure in physiological feature consistency verification, and explain the specific reasons for the failure, generating a hierarchical matching verification result that includes the reasons for the failure.

[0168] Step S150: Output the face recognition result of the target object based on the hierarchical matching verification result. The face recognition result includes both the identity matching conclusion and the physiological activity determination conclusion.

[0169] Step S151: Parse the hierarchical matching verification result and extract the overall matching marker, overall matching details, or failed sub-region information from the hierarchical matching verification result.

[0170] The hierarchical matching verification results are analyzed to extract the overall matching marker (pass or fail), overall matching details (such as the matching contribution of each sub-region, the overall matching level evaluation value, etc.), or information on failed sub-regions (such as the name of the failed sub-region, the difference, the matching criteria, etc.).

[0171] Step S152: If the layered matching verification result is that the overall matching is successful, then retrieve the target object identity information corresponding to the standard three-dimensional facial feature point set from the identity information database associated with the standard three-dimensional facial feature point set. The target object identity information includes name, identity identifier, and facial feature registration time.

[0172] The pre-defined standard 3D facial feature point set is associated with an identity information database that stores the identity information of known individuals, with each point corresponding to a specific standard 3D facial feature point set. When the hierarchical matching verification result is a successful overall match, the corresponding target object's identity information, including name, identity identifier (such as ID card number, employee ID, etc.), and facial feature registration time, is retrieved from the identity information database based on the index or identifier of the standard 3D facial feature point set.

[0173] Step S153: The retrieved target object identity information is formatted and filled with the target object identity information according to the preset identity information template to generate structured identity information. In the structured identity information, each information item is marked with data type and filing source.

[0174] The retrieved target identity information is formatted according to a preset identity information template. The identity information template defines the name, order, and format requirements of the information items. Each information item is filled into the corresponding position in the template, and the data type (such as string, date, etc.) and filing source (such as public security system, enterprise internal database, etc.) of each information item are marked to generate structured identity information.

[0175] Step S154: Combine the physiological characteristic consistency assessment results in the overall matching details with the preset activity judgment criteria to generate a physiological activity judgment conclusion. If the physiological characteristic consistency meets the activity judgment criteria, the target object is judged to have physiological activity; if it does not meet the criteria, the physiological activity of the target object is judged to be questionable.

[0176] The preset activity determination criteria are used to determine whether a target object possesses physiological activity. Examples include the maximum allowable range of physiological parameter differences or the minimum value of a physiological characteristic consistency assessment index. The physiological characteristic consistency assessment results in the overall matching details are compared with the activity determination criteria. If the criteria are met, the target object is determined to possess physiological activity; otherwise, the target object's physiological activity is questionable.

[0177] Step S155: Combine the structured identity information with the physiological activity determination conclusion to form an identity matching conclusion. The identity matching conclusion indicates that the identity matching was successful and the corresponding target object's identity information, and also indicates the physiological activity determination result.

[0178] The structured identity information and the physiological activity determination result are combined to form an identity matching conclusion. The identity matching conclusion is marked as a successful match, and the corresponding target object's identity information is listed, along with the physiological activity determination result (physiological activity present or questionable).

[0179] Step S156: If the hierarchical matching verification result is a matching failure, analyze the reason for the failure based on the information of the failed sub-region. If the failure is due to geometric structure feature matching, form a conclusion of identity matching failure and explain the differences in geometric structure features; if the failure is due to physiological feature consistency verification, form a conclusion of identity matching failure and explain the differences in physiological features.

[0180] When the hierarchical matching verification result is a matching failure, the reason for the failure is analyzed based on the information of the failed sub-region. If the failure is due to geometric structural feature matching, the differences in geometric structural features of the failed sub-region are analyzed, such as regions with large coordinate differences, to form a conclusion of identity matching failure and explain the differences. If the failure is due to physiological feature consistency verification, the differences in physiological features of the failed sub-region are analyzed, such as large differences in micromotor frequency or blood flow cycle, to form a conclusion of identity matching failure and explain the differences.

[0181] Step S157: Perform language standardization on the generated identity matching conclusion, and supplement with suggested operations, such as suggesting re-collecting face data when the match fails, and suggesting secondary verification when physiological activity is questionable.

[0182] The generated identity matching results are processed to standardize the language, using concise, accurate, and easy-to-understand expressions. Simultaneously, supplementary operational suggestions are provided based on the matching results. For example, when a match fails, it is recommended to re-collect facial data; when physiological activity is questionable, it is recommended to conduct secondary verification or adopt other verification methods.

[0183] Step S158: Integrate the standardized identity matching conclusion with the physiological activity determination conclusion, and form the final face recognition result using a preset output format. The preset output format includes the face recognition result generation time, verification process number, and key parameter details.

[0184] The standardized identity matching results and physiological activity determination results are integrated and formatted according to a preset output format. The preset output format includes the face recognition result generation time, verification process number (for tracing the verification process), and detailed key parameters (such as the overall matching level assessment value, matching status of each sub-region, etc.). The final face recognition result can be output in text, JSON, or other formats for use by users or related systems.

[0185] Throughout the data collection process, data potentially containing sensitive information, such as facial images and physiological signals, is involved. To protect user privacy, the smart camera encrypts the data in real time during collection, employing symmetric encryption algorithms to encrypt image and physiological signal data before storage and transmission. During data transmission, secure communication protocols (such as HTTPS) are used to ensure data is not leaked or tampered with during transmission. For data stored locally or on a server, access control mechanisms are implemented, allowing only authorized personnel to access and process this data, thereby achieving privacy protection and preventing the leakage of sensitive data.

[0186] In one exemplary embodiment, a 3D modeling-based intelligent camera stereo face recognition system is provided. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2 As shown, this 3D modeling-based intelligent camera stereo face recognition system includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by a processor, this computer program implements a 3D modeling-based intelligent camera stereoscopic face recognition method. The display unit of this 3D modeling-based intelligent camera stereoscopic face recognition system is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of this 3D modeling-based intelligent camera stereoscopic face recognition system can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad set on the casing of the 3D modeling-based intelligent camera stereoscopic face recognition system, or an external keyboard, touchpad, or mouse, etc.

[0187] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A 3D modeling-based intelligent camera stereo face recognition method, characterized in that, The method comprises: Collecting multi-angle face image data and synchronous physiological signal data of a target object through an intelligent camera, generating a multi-angle face image data set and a dynamic physiological signal set, the multi-angle face image data set containing face images under different shooting angles, and the dynamic physiological signal set containing face micro-motion signals and skin blood flow signals; Constraining the multi-angle face image data set based on the dynamic physiological signal set, and constructing a three-dimensional face model with physiological feature markers, the three-dimensional face model with physiological feature markers carrying face three-dimensional geometric structure and dynamic physiological feature information at the same time; Extracting a three-dimensional face feature point set with fused physiological features from the three-dimensional face model with physiological feature markers, the three-dimensional face feature point set with fused physiological features containing both geometric coordinate information and physiological signal change parameters; Performing hierarchical matching verification on the three-dimensional face feature point set with fused physiological features and a preset standard three-dimensional face feature point set, generating a hierarchical matching verification result, and the preset standard three-dimensional face feature point set containing standard physiological feature parameters; Outputting a face recognition result of the target object according to the hierarchical matching verification result, the face recognition result containing both an identity matching conclusion and a physiological activity determination conclusion. 2.The 3D modeling based intelligent camera stereo face recognition method according to claim 1, characterized in that, The method comprises: Extracting face region contour information of each image frame from the multi-angle face image data set, separating face micro-motion signals and skin blood flow signals from the dynamic physiological signal set at the same time, correlating time-domain variation rules of the face micro-motion signals with spatial forms of the face region contour information, and generating a first correlation mapping table, the first correlation mapping table being used to record corresponding relationships between face contour forms and micro-motion amplitudes at different time nodes; Performing spatial distribution inversion processing on the skin blood flow signals, combining shooting angle parameters of the intelligent camera, determining blood flow signal intensity distribution rules of different regions of the face, and generating a blood flow signal spatial distribution map, the blood flow signal spatial distribution map marking blood flow signal intensity intervals of each sub-region of the face, including a forehead, an eye surrounding, a cheek, and a lip surrounding; Based on the first correlation mapping table, performing time dimension screening on image frames in the multi-angle face image data set, retaining image frames with continuous micro-motion signals and non-mutated contour forms, and eliminating image frames with interrupted micro-motion signals or abnormal contour forms, thereby generating a time-continuous effective image frame set, time intervals of each frame of image in the time-continuous effective image frame set being consistent with sampling intervals of the micro-motion signals; The depth information reconstruction is performed on each image in the set of time-continuous effective image frames, the blood flow intensity parameter in the blood flow signal space distribution map is introduced to perform depth calculation, the region with blood flow signal intensity value in a preset high intensity interval is set as a depth calculation priority region, the pixel depth value of the depth calculation priority region is preferentially acquired, a depth image with blood flow priority marking is generated, and the depth value accuracy of the depth image with blood flow priority marking in the region with blood flow signal intensity value in a preset high intensity interval is higher than that in other regions; According to the shooting angle parameters of the set of time-continuous effective image frames, a multi-angle coordinate conversion matrix is established, the coordinates of each frame of the depth image with blood flow priority marking are uniformly converted into a preset three-dimensional coordinate system, the depth images of different angles are complementarily overlapped in the three-dimensional space, a multi-angle depth image coordinate uniform set is generated, and the coordinate origin and coordinate axis direction of each depth image in the multi-angle depth image coordinate uniform set are completely consistent; Based on the multi-angle depth image coordinate uniform set, an initial three-dimensional face geometric framework is constructed, the framework construction is performed with reference to the first correlation mapping table, the contour change amount corresponding to the micro-motion signal is converted into a dynamic adjustment parameter of the geometric framework, the initial three-dimensional face geometric framework can reflect the geometric shape change caused by the facial micro-motion, a three-dimensional geometric framework with micro-motion dynamic parameters is generated, and each geometric vertex in the three-dimensional geometric framework with micro-motion dynamic parameters is associated with a corresponding micro-motion amplitude parameter. The facial texture information of each frame of image in the set of time-continuous effective image frames is extracted, the texture information is classified according to the blood flow intensity region according to the blood flow signal space distribution map, the region with blood flow signal intensity value higher than the average value is divided into a first blood flow intensity region, and the region with blood flow signal intensity value lower than the average value is divided into a second blood flow intensity region; the texture information of the first blood flow intensity region is processed by using a first compression algorithm, and the texture information of the second blood flow intensity region is processed by using a second compression algorithm, wherein the compression rate of the first compression algorithm is lower than that of the second compression algorithm, a classified facial texture set is generated, and the texture detail resolution of different blood flow intensity regions in the classified facial texture set is different; The classified facial texture set is mapped to the surface of the three-dimensional geometric framework with micro-motion dynamic parameters, the resolution of the texture detail and the depth value accuracy corresponding to the blood flow priority marking are kept in a corresponding adaptive relationship, the texture detail resolution of the first blood flow intensity region is adapted to the depth value accuracy of the corresponding region, the micro-motion signal and the blood flow signal in the dynamic physiological signal set are embedded in the corresponding region of the three-dimensional geometric framework in a marked form, and a final three-dimensional face model with physiological feature marking is generated, and each facial region in the three-dimensional face model with physiological feature marking contains a three-dimensional geometric shape, texture detail, micro-motion parameter and blood flow signal parameter. 3.The 3D modeling based intelligent camera stereo face recognition method of claim 2, wherein, The shooting angle parameters of the set of time-continuous effective image frames are used to establish a multi-angle coordinate conversion matrix, and the coordinates of each frame of blood flow priority marked depth image are uniformly converted into a preset three-dimensional coordinate system, so that the depth images of different angles form complementary coverage in the three-dimensional space to generate a multi-angle depth image coordinate uniform set, including: The shooting angle parameters of each image in the set of time-continuous effective image frames are extracted from the parameter configuration file of the intelligent camera, including the horizontal rotation angle, the vertical pitch angle, and the lens focal length. The horizontal rotation angle, the vertical pitch angle, and the lens focal length are arranged in the order of image frame collection to generate a frame-angle parameter correspondence table. In the frame-angle parameter correspondence table, each image frame number is associated with a complete set of shooting angle parameters. The parameters of the preset three-dimensional coordinate system are determined, including the coordinate origin position, the positive direction definition of the horizontal axis, the vertical axis, and the optical axis. The coordinate origin is set as the center position of the intelligent camera collection area, the horizontal axis is along the horizontal direction, the vertical axis is along the vertical direction, and the optical axis is along the lens optical axis direction. For each frame of blood flow priority marked depth image, a coordinate conversion matrix is constructed based on the shooting angle parameters corresponding to the frame of blood flow priority marked depth image. First, a rotation matrix is constructed according to the horizontal rotation angle and the vertical pitch angle. Then, a scaling matrix is constructed according to the lens focal length. Finally, a translation matrix is constructed in combination with the position of the image frame in the collection area. The rotation matrix, the scaling matrix, and the translation matrix are combined in order to obtain the multi-angle coordinate conversion matrix of the frame of blood flow priority marked depth image. The two-dimensional coordinates and depth values of each pixel point in each frame of blood flow priority marked depth image are combined to form the original three-dimensional coordinates of the pixel point. The original three-dimensional coordinates of each pixel point are substituted into the multi-angle coordinate conversion matrix corresponding to the frame of blood flow priority marked depth image to which the pixel point belongs, and the target three-dimensional coordinates in the preset three-dimensional coordinate system are converted. The pixel point coordinate data that exceeds the effective range of the preset three-dimensional coordinate system after conversion is removed, and the pixel point coordinate data within the effective range is retained. The effective pixel point coordinate data of each frame of blood flow priority marked depth image after conversion is arranged according to the frame number to form a single-frame coordinate conversion result. The single-frame coordinate conversion result includes the target three-dimensional coordinates of all effective pixel points of the frame of blood flow priority marked depth image and the corresponding blood flow priority mark. All single-frame coordinate conversion results are combined to correct the image frame coordinate conversion results with coverage gaps, so that all facial regions are covered with coordinate data to generate a multi-angle depth image coordinate uniform set. In the multi-angle depth image coordinate uniform set, the coordinate data of each frame of blood flow priority marked depth image is sorted according to the coordinate values of the preset three-dimensional coordinate system. 4.The 3D modeling based intelligent camera stereo face recognition method of claim 1, wherein, The three-dimensional facial feature point set fused with physiological features is extracted from the three-dimensional face model with physiological feature marks, including: Separating the face three-dimensional geometry data and dynamic physiological characteristic data from the three-dimensional face model with physiological characteristic markers, dividing the face micro-motion signal and skin blood flow signal in the dynamic physiological characteristic data according to the face sub-regions to generate a second correlation mapping table, which records the micro-motion amplitude range and blood flow intensity range corresponding to each sub-region of the forehead, the eye periphery, the cheek, the lip periphery and the nose bridge; Based on the second correlation mapping table, marking the boundary coordinates of each sub-region in the three-dimensional geometry data of the three-dimensional face model with physiological characteristic markers, determining the geometry vertex distribution density in each sub-region, calculating the product of the micro-motion amplitude range and the blood flow intensity range of each sub-region as the physiological signal change activity of the sub-region; setting the region with the physiological signal change activity higher than the average activity of all sub-regions as the first feature point extraction region, and setting the region with the physiological signal change activity lower than the average activity of all sub-regions as the second feature point extraction region to generate a feature point extraction density distribution map, and the geometry vertex sampling interval of the first feature point extraction region is smaller than that of the second feature point extraction region; According to the feature point extraction density distribution map, the geometry vertices are extracted in the first feature point extraction region with a first sampling interval and in the second feature point extraction region with a second sampling interval, and the first sampling interval is smaller than the second sampling interval, and the geometry vertices are extracted in the second feature point extraction region with sparse sampling, and the three-dimensional coordinate values of each extracted vertex in the three-dimensional face model with physiological characteristic markers are recorded to generate an initial three-dimensional feature point coordinate set, and the number of feature points in the first feature point extraction region is more than that in the second feature point extraction region in the initial three-dimensional feature point coordinate set; Extracting the physiological signal change parameters of each sub-region from the dynamic physiological characteristic data, including the time domain fluctuation frequency of the micro-motion signal and the intensity change period of the skin blood flow signal, and binding the time domain fluctuation frequency of the micro-motion signal and the intensity change period of the skin blood flow signal with the feature points of the corresponding sub-region in the initial three-dimensional feature point coordinate set, so that each feature point contains a three-dimensional coordinate value and is associated with the physiological signal change parameters of the sub-region, to generate a feature point set with physiological parameter binding; Performing redundancy screening on the feature point set with physiological parameter binding, comparing the three-dimensional coordinate distance and physiological parameter similarity of adjacent feature points, retaining the feature points with more representative coordinates and physiological parameters, and eliminating redundant feature points to generate a de-redundant feature point set; Performing standardization processing on the three-dimensional coordinate values of each feature point in the de-redundant feature point set, taking the geometric center of the three-dimensional face model with physiological characteristic markers as the origin, and converting the coordinates of all feature points into offset coordinates relative to the geometric center to generate a standardized coordinate feature point set. The physiological parameters associated with each feature point in the standardized coordinate feature point set are supplemented and improved, the physiological signal change parameters of the corresponding sub-regions are extracted from the dynamic physiological signal data, and the micro-motion frequency and blood flow cycle parameters are associated with each feature point to generate a parameter-complete feature point set; The parameter-complete feature point set is classified and arranged according to the facial sub-regions to form a forehead feature point subset, an eye surrounding feature point subset, a cheek feature point subset, a lip surrounding feature point subset, and a nose bridge feature point subset, and all the subsets are combined to form a final three-dimensional facial feature point set with fused physiological features, wherein the feature points in each subset are sorted in ascending order according to the coordinate offset. 5.The 3D modeling based intelligent camera stereo face recognition method of claim 4, wherein, The facial three-dimensional geometric structure data and the dynamic physiological feature data are separated from the three-dimensional face model with physiological feature markers, the facial micro-motion signals and the skin blood flow signals in the dynamic physiological feature data are divided according to the facial sub-regions, and a second association mapping table is generated, including: The complete data file of the three-dimensional face model with physiological feature markers is read, the three-dimensional geometric data segment and the physiological feature data segment stored in the complete data file of the three-dimensional face model with physiological feature markers are parsed, the three-dimensional geometric data segment includes model vertex coordinates, face index, and texture coordinates, and the physiological feature data segment includes facial micro-motion signal time series data and skin blood flow signal spatial distribution data; The model vertex coordinates in the three-dimensional geometric data segment are grouped according to the facial anatomical structure, the vertex coordinate ranges corresponding to the forehead, eye surrounding, cheek, lip surrounding, and nose bridge sub-regions are determined, and a coordinate mapping table is generated, which records the coordinate intervals of all vertices included in each sub-region; The facial micro-motion signal time series data is extracted from the physiological feature data segment, the micro-motion amplitude value of each monitoring point is assigned to the corresponding facial sub-region according to the spatial position coordinates of the monitoring point and the coordinate mapping table, and a time series data set is generated, wherein each sub-region corresponds to a group of time series data including time nodes and micro-motion amplitudes; The skin blood flow signal spatial distribution data is extracted from the physiological feature data segment, the blood flow intensity value of each sampling point is assigned to the corresponding facial sub-region according to the coordinate mapping table, and a spatial data set is generated, wherein each sub-region corresponds to a group of spatial data including sampling point coordinates and blood flow intensity; The micro-motion time series data of each sub-region in the time series data set is statistically analyzed to determine the micro-motion amplitude range of the sub-region; The blood flow intensity data of each sub-region in the spatial data set is statistically analyzed to determine the blood flow intensity range of the sub-region; The name, micro-motion amplitude range, and blood flow intensity range of each sub-region are associated to form a second mapping association table, wherein the micro-motion amplitude range and the blood flow intensity range of each sub-region in the second association mapping table are marked with the time nodes or the number of sampling points of the data source. 6.The 3D modeling based intelligent camera stereo face recognition method of claim 4, wherein, The redundant feature point set bound with the physiological parameter is screened, the three-dimensional coordinate distance and the physiological parameter similarity of adjacent feature points are compared, the feature point with higher representative of the coordinate and the physiological parameter is reserved, the redundant feature point is eliminated, and a de-redundant feature point set is generated, including: Grouping all feature points in the feature point set bound with the physiological parameter according to the facial sub-region, and sorting the feature points in each sub-region according to the horizontal axis coordinate value from small to large to generate a sorted feature point list; For each feature point in the sorted feature point list, two adjacent feature points are selected as a pair of analyzed feature points in turn according to the sorting order, starting from the first feature point until all adjacent feature point pairs in the sub-region are selected; The three-dimensional coordinate distance of each pair of analyzed feature points is compared, and the physiological parameter similarity of each pair of analyzed feature points is analyzed; The distance standard and the physiological parameter similarity standard of adjacent feature points are called from the preset screening parameter library; If the three-dimensional coordinate distance of the analyzed feature point pair meets the distance standard and the physiological parameter similarity of the analyzed feature point pair meets the similarity standard, it is determined that the analyzed feature point pair is redundant; For the analyzed feature point pair determined to be redundant, the coordinate representativeness and the parameter representativeness of the two feature points in the analyzed feature point pair are analyzed, the feature point with higher coordinate representativeness and parameter representativeness than the other feature point is reserved, the coordinate representativeness is judged by the correlation degree of the feature point and the geometric center of the sub-region, and the parameter representativeness is judged by the correlation degree of the physiological parameter of the feature point and the average physiological parameter of the sub-region; After the processing of each sub-region is completed, all reserved feature points are collected to form a de-redundant feature point set, and the number of feature points in each sub-region in the de-redundant feature point set is dynamically adjusted according to the size of the sub-region and the physiological signal change amplitude, and the number of reserved feature points in the first feature point extraction region is more than that in the second feature point extraction region. 7.The 3D modeling based intelligent camera stereo face recognition method according to claim 1, characterized in that, The three-dimensional facial feature point set fused with the physiological feature and the preset standard three-dimensional facial feature point set are subjected to hierarchical matching verification to generate a hierarchical matching verification result, including: The standardized coordinate values of all feature points are extracted from the three-dimensional facial feature point set fused with the physiological feature to form a to-be-matched geometric coordinate set, and the standard coordinate values of all feature points are extracted from the preset standard three-dimensional facial feature point set to form a standard geometric coordinate set; The coordinate differences of the corresponding sub-region feature points in the to-be-matched geometric coordinate set and the standard geometric coordinate set are compared, including the coordinate differences of each feature point in the forehead, eye, cheek, lip, and nose bridge sub-regions, to generate a sub-region coordinate difference record, and each element in the sub-region coordinate difference record corresponds to the coordinate difference of a single feature point. Based on the sub-region coordinate difference record, the geometric matching degree of each sub-region is evaluated, and the preset sub-region geometric matching standard is combined. If the geometric matching degree of all sub-regions meets the corresponding standard, it is determined that the geometric structure feature preliminary matching passes, and a geometric preliminary matching pass result is generated. If the geometric matching degree of at least one sub-region does not meet the corresponding standard, it is determined that the geometric structure feature preliminary matching fails, and a geometric preliminary matching failure result is generated. When the geometric structure feature preliminary matching passes, the physiological signal change parameters associated with each feature point in the fusion physiological feature three-dimensional face feature point set are extracted to form a to-be-matched physiological parameter set. The standard physiological parameters associated with each feature point in the preset standard three-dimensional face feature point set are extracted to form a standard physiological parameter set. The physiological parameter differences of the corresponding sub-region feature points in the to-be-matched physiological parameter set and the standard physiological parameter set are compared, including micro-motion frequency difference and blood flow cycle difference, to generate a sub-region physiological parameter difference record. In the sub-region physiological parameter difference record, each element corresponds to the physiological parameter difference of a single feature point. Based on the sub-region physiological parameter difference record, the physiological feature consistency of each sub-region is evaluated, and the preset sub-region physiological matching standard is combined. If the physiological feature consistency of all sub-regions meets the corresponding standard, it is determined that the physiological feature consistency verification passes, and a physiological verification pass result is generated. If the physiological feature consistency of at least one sub-region does not meet the corresponding standard, it is determined that the physiological feature consistency verification fails, and a physiological verification failure result is generated. When the geometric structure feature preliminary matching passes and the physiological feature consistency verification passes, the geometric matching degree and the physiological feature consistency evaluation results are combined to generate a hierarchical matching verification result of overall matching pass. The hierarchical matching verification result of overall matching pass includes the matching details of each sub-region. When the geometric structure feature preliminary matching fails or the physiological feature consistency verification fails, the specific information of the failed sub-region is recorded, including the failed sub-region name, the corresponding difference and the matching standard, and a hierarchical matching verification result containing the failure reason is generated. In the hierarchical matching verification result containing the failure reason, it is marked whether the geometric structure feature matching fails or the physiological feature consistency verification fails. 8.The 3D modeling based intelligent camera stereo face recognition method of claim 7, wherein, The sub-region coordinate difference record is based on the evaluation of the geometric matching degree of each sub-region, and the preset sub-region geometric matching standard is combined. If the geometric matching degree of all sub-regions meets the corresponding standard, it is determined that the geometric structure feature preliminary matching passes, and a geometric preliminary matching pass result is generated, including: For each sub-region in the sub-region coordinate difference record, the coordinate difference of all feature points in the sub-region is extracted to form a three-dimensional difference record of the sub-region; Based on the three-dimensional difference record of the sub-region, the overall coordinate matching of the feature points in the sub-region is comprehensively evaluated to determine the geometric matching degree of the sub-region; retrieve the corresponding sub-region geometric matching standard of each sub-region from a preset geometric matching standard library, wherein the geometric matching standards of different sub-regions are set according to the importance of the sub-region in face recognition, and the difference allowed range of the geometric matching standards of the sub-regions around the eyes and lips is smaller than that of the sub-regions on the forehead and cheeks; compare the geometric matching degree of each sub-region with the corresponding sub-region geometric matching standard, if the geometric matching degree of a certain sub-region meets the corresponding sub-region geometric matching standard, mark the sub-region geometric matching as passing; if not, mark the sub-region geometric matching as failing; statistically analyze the marking results of all sub-regions, if all sub-regions are marked as geometric matching passing, determine that the entire geometric structure feature preliminary matching passes; generate a geometric preliminary matching passing result file, which contains the geometric matching degree of each sub-region, the corresponding sub-region geometric matching standard, and the matching marking result of each sub-region; organize the geometric preliminary matching passing result file, so that the relevant data of all sub-regions are recorded, and the geometric preliminary matching passing result file is used as the final geometric preliminary matching passing result. 9.The 3D modeling based intelligent camera stereo face recognition method of claim 7, wherein, When the geometric structure feature preliminary matching passes and the physiological feature consistency verification passes, combine the geometric matching degree and the physiological feature consistency evaluation result to generate a hierarchical matching verification result of overall matching passing, including: extract the geometric matching degree of each sub-region and the corresponding sub-region geometric matching standard from the geometric preliminary matching passing result, and determine the geometric matching contribution degree of each sub-region in combination with the importance of the sub-region in face recognition; extract the physiological feature consistency evaluation result of each sub-region and the corresponding sub-region physiological matching standard from the physiological verification passing result, and determine the physiological matching contribution degree of each sub-region in combination with the physiological feature stability of the sub-region; retrieve the weight proportion of the geometric matching contribution degree and the weight proportion of the physiological matching contribution degree from a preset weight configuration file, and the sum of the weight proportion of the geometric matching contribution degree and the weight proportion of the physiological matching contribution degree is 1; combine the geometric matching contribution degree of each sub-region, the physiological matching contribution degree of each sub-region, and the weight proportion of the geometric matching contribution degree and the weight proportion of the physiological matching contribution degree to comprehensively evaluate the overall matching level; if the overall matching level is higher than a preset overall matching standard, generate a hierarchical matching verification result of overall matching passing; in the hierarchical matching verification result of overall matching passing, record the geometric matching contribution degree of each sub-region, the physiological matching contribution degree of each sub-region, and the evaluation basis of the overall matching level; organize the hierarchical matching verification result of overall matching passing to form the final hierarchical matching verification result of overall matching passing.

10. A 3D modeling-based intelligent camera stereo face recognition system, characterized in that, including: a processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the machine-executable instructions to perform the method of claim 1 to 9.

Images