A method and system for unmanned hotel management

Abstract

The application relates to the technical field of face recognition and discloses an unmanned hotel management method and system, wherein the unmanned hotel management method comprises the following steps: step S101, obtaining a standardized frame through calculation; step S102, determining a local observation area; step S103, determining an outer-nose-valve boundary curve; step S104, calculating the geometric curvature of the outer-nose-valve boundary curve; step S105, calculating the second-order energy density of the outer-nose-valve collapse curvature; and step S106, generating an access token. The application eliminates geometric distortion caused by small head movements by selecting face reference points for similarity transformation alignment; irrelevant interference is excluded by circumscribing a local observation area according to the face dissection proportion; the boundary curve is extracted by gradient energy line integration, and energy aggregation and normalization of the second-order derivative of the curvature time are combined to avoid the risk of prosthetic attacks; and therefore, the safety and stability of face recognition in the unmanned self-check-in scene are improved.

Description

Technical Field

[0001] This invention relates to the field of facial recognition technology, and more specifically, to an unmanned hotel management method and system. Background Technology

[0002] With the development of intelligent hotels, unmanned self-service check-in is becoming increasingly common. Facial recognition technology, as the core identity verification method in this model, directly impacts the system's performance in terms of security and accuracy. Existing technologies mostly use static facial features for identity comparison, failing to incorporate facial physiological movement characteristics for liveness detection. These methods only extract planar texture or contour information of the face, unable to distinguish between a real human body and spoofed images such as photos or videos. Criminals can bypass verification by creating fake faces, posing a security risk to guest rooms. Furthermore, during facial image acquisition, the user's head is prone to slight translation, rotation, or scaling. These movements cause significant geometric distortion between acquisition frames. Traditional image alignment methods use insufficiently stable reference points, failing to effectively eliminate the effects of distortion. Inter-frame geometric differences reduce the accuracy of subsequent steps such as external nasal valve boundary extraction and curvature calculation. The physiological movement characteristics upon which liveness detection relies cannot be accurately quantified, further affecting the reliability of the detection results.

[0003] Furthermore, existing methods do not effectively combine live physiological movement characteristics with identity features. In unattended scenarios, this fails to guarantee verification security and also makes it difficult to improve the accuracy of identity recognition. These issues collectively hinder the widespread application of facial recognition systems for unattended self-service check-in in hotels. Summary of the Invention

[0004] This invention provides an unmanned hotel management method and system, which solves the technical problems mentioned in the background.

[0005] This invention provides a method for unmanned hotel management, comprising the following steps: Step S101: Collect the video sequence when prompting the user to inhale, detect the tip of the nose, the philtrum and the glabella as the face stabilization reference points, and use the least squares similarity transformation to uniformly align the video sequence to the first frame coordinate system to obtain a standardized frame. Step S102: Based on the facial anatomy proportions, delineate the local observation area containing the external nasal valve structure in the standardized frame to ensure that the anatomical position of the observed object remains constant. Step S103: Calculate the image gradient modulus within the local observation area and search for the path that maximizes the gradient energy line integral to determine the boundary curve of the external nose valve. Step S104: Calculate the geometric curvature of the boundary curve of the external nasal valve, and use the central difference operation to obtain the time second derivative of the curvature in order to quantify the acceleration level change of the boundary shape. Step S105: In the spatiotemporal domain, the second time derivative of curvature is aggregated and normalized to calculate the second energy density of the external nasal valve collapse curvature. The second energy density of the external nasal valve collapse curvature is compared with a preset threshold to output the liveness determination result. Step S106: Only when the liveness determination result is passed, select the frame with the maximum gradient energy line integral to extract the face vector, bind the face vector with the room check-in record and generate an access token.

[0006] This invention provides an unmanned hotel management system, comprising: The standardized frame calculation module collects video sequences when prompting the user to inhale, detects the tip of the nose, the philtrum, and the glabella as face stabilization reference points, and uses least squares similarity transformation to uniformly align the video sequence to the first frame coordinate system to obtain a standardized frame. The local observation area delineation module delineates the local observation area, including the external nasal valve structure, in a standardized frame based on the anatomical proportions of the face, ensuring that the anatomical position of the observed object remains constant. The external nose valve boundary curve determination module calculates the image gradient modulus within the local observation area and searches for the path that maximizes the gradient energy line integral to determine the external nose valve boundary curve. The geometric curvature calculation module calculates the geometric curvature of the boundary curve of the external nasal valve and uses the central difference operation to obtain the time second derivative of the curvature in order to quantify the acceleration level change of the boundary shape. The second-order energy density calculation module performs energy aggregation and normalization on the time second derivative of curvature in the spatiotemporal domain, calculates the second-order energy density of the external nasal valve collapse curvature, and compares the second-order energy density of the external nasal valve collapse curvature with a preset threshold to output the liveness determination result. The access token generation module extracts the face vector from the frame with the maximum gradient energy line integral only when the liveness detection result is passed. It then binds the face vector to the room check-in record and generates an access token.

[0007] The beneficial effects of this invention are as follows: By selecting stable facial reference points for similarity transformation alignment, this invention eliminates geometric distortions caused by small head movements, ensuring consistency in subsequent processing. Local observation areas are delineated based on facial anatomical proportions, focusing on the external nasal valve region and eliminating irrelevant interference. Boundary curves are extracted through gradient energy line integration, combined with energy aggregation and normalization of the second derivative of curvature time, achieving reliable liveness detection and avoiding the risk of spoofing attacks. Only after successful liveness detection is the facial vector of the clearest frame extracted, bound to check-in information to generate an authorization token, forming a complete identity verification and check-in authorization closed loop. This method improves the security and stability of facial recognition in unmanned self-service check-in scenarios, simplifies the system's computational process, reduces invalid calculations, balances identity recognition accuracy and information protection requirements, and is suitable for the practical application scenarios of unmanned hotels. Attached Figure Description

[0008] Figure 1 This is a flowchart of an unmanned hotel management method according to the present invention; Figure 2 This is a schematic diagram of an unmanned hotel management system according to the present invention. Detailed Implementation

[0009] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0010] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" indicate that the element or object preceding the term encompasses the elements or objects listed following the term and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0011] like Figures 1-2 As shown, an unmanned hotel management method includes the following steps: Step S101: Collect the video sequence when prompting the user to inhale, detect the tip of the nose, the philtrum and the glabella as the face stabilization reference points, and use the least squares similarity transformation to uniformly align the video sequence to the first frame coordinate system to obtain a standardized frame. Step S102: Based on the facial anatomy proportions, delineate the local observation area containing the external nasal valve structure in the standardized frame to ensure that the anatomical position of the observed object remains constant. Step S103: Calculate the image gradient modulus within the local observation area and search for the path that maximizes the gradient energy line integral to determine the boundary curve of the external nose valve. Step S104: Calculate the geometric curvature of the boundary curve of the external nasal valve, and use the central difference operation to obtain the time second derivative of the curvature in order to quantify the acceleration level change of the boundary shape. Step S105: In the spatiotemporal domain, the second time derivative of curvature is aggregated and normalized to calculate the second energy density of the external nasal valve collapse curvature. The second energy density of the external nasal valve collapse curvature is compared with a preset threshold to output the liveness determination result. Step S106: Only when the liveness determination result is passed, select the frame with the maximum gradient energy line integral to extract the face vector, bind the face vector with the room check-in record and generate an access token.

[0012] It should be noted that the video sequence collected when prompting the user to inhale utilizes the physiological movement characteristics of the external nasal valve during inhalation to provide effective data for subsequent liveness detection. During implementation, text or voice prompts are displayed on the acquisition interface, requiring the user to inhale naturally through the nose for a duration consistent with the acquisition window length. For example, prompting the user to inhale for 3 seconds, a video sequence of this time is simultaneously acquired to ensure the complete movement of the external nasal valve is captured, avoiding subsequent failures due to the absence of physiological movement. The tip of the nose, the philtrum, and the glabella are selected as stable facial reference points because these feature points are less affected by facial expressions and have stable positioning. The tip of the nose is located in the center of the face, the philtrum is located at the starting point of the nasolabial groove, and the glabella is located between the eyebrows. These three points are relatively stable during small head movements and are not easily obscured.

[0013] In one embodiment of the present invention, a video sequence is acquired when the user is prompted to inhale, and the tip of the nose, the philtrum, and the glabella are detected as facial stabilization reference points. The video sequence is then uniformly aligned to the coordinate system of the first frame using a least-squares similarity transformation to obtain a standardized frame, including: Establish discrete sampling time ,in It is a count variable, with values ​​ranging from zero to... , The time interval between adjacent frames. , The acquisition window length is used to obtain the original frame set. ,in For the first Image intensity function at each sampling time; In each frame Detection of the tip of the nose philtrum and the point between the eyebrows All three are two-dimensional coordinate vectors; Constructing a similarity transformation model The optimal parameters are then solved using the least squares method. ;

[0014] in For scale parameters, It is a two-dimensional rotation matrix and satisfies and , It is a translation vector. The coordinates of the reference point at the current sampling time. The coordinates of the reference point at the initial sampling time. It is a two-dimensional rotation group; Generate normalized frames using the inverse mapping of the similarity transformation model : ;

[0015] in For similarity transformation model The inverse mapping, For the first The normalized frame image intensity function at each sampling time.

[0016] It should be noted that a time series is a collection of multiple sampling moments arranged in chronological order. The time interval between adjacent frames is the time difference between two adjacent sampling moments, reflecting the frame rate characteristics of the video. It can be obtained by reading the camera's hardware parameters or by calculating the difference in timestamps between two consecutive frames. The total acquisition duration is the time span of the video sequence from the start to the end of acquisition, reflecting the length of the acquisition window. It can be controlled by setting the camera's acquisition time threshold or calculated by setting the total number of acquisition frames combined with the time interval between adjacent frames. The acquisition window length is preferably set to 2 to 5 seconds, covering the complete process of one natural nasal inhalation. The raw frame set is the collection of all raw image frames within the acquisition period, reflecting all acquired image data. It can be obtained by real-time shooting and storage by the camera, and the storage format can be common image formats such as bitmaps. The image intensity function is a function describing the brightness value of each pixel in the image, reflecting the grayscale distribution characteristics of the image. It can be obtained by grayscale processing of the raw image frames; grayscale methods can include weighted averaging. The tip of the nose is a feature point on the human face. The philtrum is a feature point on the human face. The glabella is a feature point on the human face. The tip of the nose, the philtrum, and the glabella can all be obtained through facial landmark detection algorithms, such as facial landmark detection models (MTCNN, RetinaFace, etc.), which will not be elaborated here.

[0017] It should be noted that the two-dimensional planar coordinate vector is a vector describing the position of feature points in the two-dimensional plane of the image, preferably taken as the coordinate value in the pixel coordinate system with the top left corner of the image as the origin. The scale parameter is the scaling ratio parameter in the similarity transformation model, reflecting the degree of scaling between frames, and typically ranges from 0.8 to 1.2 to cover small head scaling. The two-dimensional rotation matrix is ​​the rotation transformation matrix in the similarity transformation model, reflecting the degree of rotation between frames, preferably taken as a two-dimensional orthogonal matrix with a determinant of 1, and the rotation angle typically ranges from -10 degrees to +10 degrees to cover small head rotation. The similarity transformation model is a geometric transformation model that includes scale, rotation, and translation transformations, reflecting the overall geometric transformation relationship between frames, and preferably takes the form of a transformation model that integrates scale, rotation, and translation parameters. The optimal similarity transformation parameters are the combination of similarity transformation parameters that minimizes the sum of squared Euclidean distances, reflecting the optimal inter-frame geometric transformation relationship. The orthogonality constraint is a constraint on the two-dimensional rotation matrix, requiring that the transpose of the matrix equals its inverse, reflecting the orthogonality property of the rotation matrix. The constraint of a determinant of 1 is a constraint condition for a two-dimensional rotation matrix, requiring the determinant value of the matrix to be equal to 1, reflecting the orientation-preserving property of the rotation matrix. Inverse mapping is the inverse transformation relation corresponding to the similarity transformation model, that is, the transformation rule that maps the reference frame coordinate system back to the current frame coordinate system. Inverse coordinate transformation is the transformation operation performed on pixel coordinates using inverse mapping, that is, the process of converting reference frame coordinates to current frame coordinates. The normalized frame sequence is the set of image frames obtained after all original frames have been aligned through similarity transformations, reflecting the image data in a unified coordinate system after alignment.

[0018] It should be noted that the specific localization method for the nose tip, philtrum, and glabella can employ a facial landmark detection method based on convolutional neural networks. For example, a pre-trained facial landmark model can be used. The original image frame is input, and the model outputs the coordinates of facial feature points including the nose tip, philtrum, and glabella. In practice, the original image is first preprocessed, including grayscale normalization, and then input into the model. The model extracts image features through convolutional layers and outputs the coordinates of feature points through fully connected layers. For example, a detection model containing 68 facial landmarks can be used. This model can stably output the coordinates of the nose tip corresponding to landmark number 30, the philtrum corresponding to landmark number 33, and the glabella corresponding to landmark number 27, thus meeting the localization requirements. The specific steps for solving the optimal similarity transformation parameters in the least squares similarity transformation include: Step 1: Constructing the objective function, which is the sum of squared Euclidean distances between the three reference points after transformation and the reference point of the first frame; Step 2: Substituting the parameters of the similarity transformation into the objective function, where the scale parameter is a positive number, the rotation matrix is ​​represented in cosine and sine form, and the translation vector is a two-dimensional vector; Step 3: Using the Levenberg-Marquardt algorithm to solve for the minimum value of the objective function, which combines the advantages of gradient descent and Gauss-Newton methods, has a fast convergence speed and good stability; Step 4: Outputting the optimal scale, rotation, and translation parameters.

[0019] It should be noted that the forward mapping of the similarity transformation is scaling, then rotation, and finally translation, while the corresponding inverse mapping is inverse translation, then inverse rotation, and finally inverse scaling. Inverse translation involves subtracting the translation vector from the coordinates; inverse rotation involves multiplying the coordinates by the transpose of the rotation matrix (since the inverse of the rotation matrix is ​​equal to its transpose); and inverse scaling involves dividing the coordinates by the scale parameter. The inverse coordinate transformation uses the same inverse mapping implementation, and resampling employs bilinear interpolation, which will not be elaborated upon here. Furthermore, two image quality assurance measures can be adopted during the generation of the standardized frame sequence: first, grayscale value normalization is performed after resampling, adjusting the grayscale values ​​to the range of 0 to 255 to avoid overly bright or dark areas; second, edge smoothing is performed using a Gaussian filtering algorithm with a filter window size of 3×3 and a standard deviation of 1.0 to reduce the jagged effect caused by resampling.

[0020] It should be noted that this invention selects a stable facial reference point, constructs a similar transformation model, and uses the least squares method to solve for the optimal transformation parameters, aligning all acquired frames to the coordinate system of the first frame to generate a standardized frame sequence. This eliminates the interference of small-amplitude rigid body movements of the head on subsequent external nasal valve feature extraction, ensuring that subsequent processing targets the same anatomical location. This makes the acquired video sequence reproducible and comparable, providing stable image data support for subsequent liveness detection, while simplifying the complexity of subsequent processing and improving the robustness of the overall solution.

[0021] In one embodiment of the present invention, a local observation area including the external nasal valve structure is delineated in a standardized frame based on the facial anatomical proportions to ensure that the anatomical position of the observed object remains constant, including: Receive standardized frame sequence With the tip of the nose in the first frame Set a horizontal unit vector in the first frame coordinate system. ; The inner radius is set according to the anatomical proportions of the human face. Outer radius , interior angle and outer corners ,satisfy and ; The local observation area is delineated according to the following formula. : ;

[0022] in These are the pixel coordinates in the first frame's coordinate system. Denotes the Euclidean norm. Indicates the included angle in radians. It is a horizontal unit vector. and These represent the positive angle range and the negative angle range, respectively. Output local observation area .

[0023] It should be noted that the horizontal unit vector is a unit-length vector along the horizontal direction in the first frame's coordinate system, reflecting the horizontal reference of the coordinate system. It is preferably a unit vector along the horizontal rightward direction of the image. The inner radius is a distance threshold defining the radial inner boundary of the local observation area, reflecting the inner radial range of the observation area. It is preferably 10 to 20 pixels, matching the normal distance from the tip of the nose to the inner edge of the external nasal valve. The outer radius is a distance threshold defining the radial outer boundary of the local observation area, reflecting the outer radial range of the observation area. It is preferably 30 to 40 pixels, matching the normal distance from the tip of the nose to the outer edge of the external nasal valve, forming a reasonable bandwidth with the inner radius. The inner angle is an angle threshold defining the circumferential inner boundary of the local observation area, reflecting the inner angle range of the observation area. It is preferably 15 to 25 degrees, matching the inner angle range of a single external nasal valve. The outer angle is the angular threshold that defines the circumferential outer boundary of the local observation area, reflecting the outer angle range of the observation area. The preferred value is 40 to 50 degrees. This angle matches the outer angle range of the unilateral external nasal valve and forms a reasonable angular bandwidth with the inner angle.

[0024] It should be noted that the local observation area is the set of pixels in the normalized frame that satisfy the distance and angle constraints, reflecting the target area for extracting the external nasal valve boundary. The distance constraint is a rule limiting the distance range between pixels in the local observation area and the nose tip, reflecting the radial filtering condition of the observation area. The preferred value is that the Euclidean distance between the pixel and the nose tip lies between the inner and outer radii. The angle constraint is a rule limiting the angle range of pixels in the local observation area relative to the nose tip, reflecting the circumferential filtering condition of the observation area. The preferred value is that the angle between the vector connecting the pixel to the nose tip and the horizontal unit vector lies within a positive or negative angle range. The positive angle range limits the positive angle range of the observation area for one side of the external nasal valve, reflecting the circumferential position of the one-sided external nasal valve. The preferred value is the angle range from the inner angle to the outer angle, which accurately covers the angle range of the one-sided external nasal valve. The negative angle range limits the negative angle range of the observation area for the other side of the external nasal valve, reflecting the circumferential position of the other side of the external nasal valve. The preferred value is the angle range from the negative outer angle to the negative inner angle, which accurately covers the angle range of the other side of the external nasal valve.

[0025] It should be noted that the specific numerical standards for facial anatomical proportions are as follows: Anatomical statistics of adult faces can be referenced, with the radial distance between the external nasal valve and the tip of the nose ranging from 10 to 40 pixels, and the circumferential angle ranging from 15 to 50 degrees. For children's faces, these parameters can be scaled down proportionally, with the radial distance ranging from 5 to 20 pixels and the circumferential angle ranging from 10 to 35 degrees. This standard is based on statistical analysis of a large number of facial images and is adaptable to the facial features of different age groups. The coordinate system of the first frame has the upper left corner of the image as the origin, with the positive x-axis pointing to the right and the positive y-axis pointing downwards. The horizontal unit vector is the unit vector in the positive x-axis direction, with coordinates of 1 and 0. This definition conforms to the general coordinate system settings in the field of image processing, facilitating subsequent calculations of vectors and angles. The specific calculation and judgment methods for distance constraints include: first, obtaining the coordinates of the pixel and the nose tip; second, calculating the difference between the two coordinates to obtain the horizontal and vertical differences; third, calculating the sum of the squares of the horizontal and vertical differences; and fourth, taking the square root of the sum of the squares to obtain the Euclidean distance. The judgment method is to determine whether the Euclidean distance is greater than or equal to the inner radius and less than or equal to the outer radius. If it meets the condition, the distance constraint is satisfied. The specific angle calculation and interval judgment process for angle constraints includes: first, obtaining the coordinate difference between the pixel and the nose tip to obtain the difference vector; second, calculating the angle between the difference vector and the horizontal unit vector, which can be calculated using the arctangent function; and third, converting the angle to an angle value. The judgment method is to determine whether the angle belongs to a positive or negative angle interval. If it meets the condition, the angle constraint is satisfied. The specific implementation steps for filtering and determining pixels in the local observation area include: Step 1: traversing all pixels in the standardized frame; Step 2: calculating the Euclidean distance and included angle for each pixel; Step 3: determining whether the pixel satisfies both the distance constraint and the angle constraint; Step 4: storing the coordinates of pixels that meet the conditions; and Step 5: outputting the stored coordinate set as the local observation area.

[0026] It should be noted that this invention, based on facial anatomical proportions, uses both distance and angle constraints to delineate a local observation area containing both external nasal valves within a standardized frame. This ensures that subsequent external nasal valve boundary extraction always targets the same anatomical region, eliminating interference from other irrelevant facial areas. Specifically, this invention sets parameters based on facial anatomical proportions, eliminating the need for manual parameter tuning and improving the robustness of the solution. The dual-constraint screening method accurately delineates the target area, avoiding omissions or redundancy. The uniformity of the observation area ensures consistency in subsequent boundary extraction and curvature calculation, providing a stable regional basis for liveness detection. Simultaneously, the symmetrical angle interval design can simultaneously cover both external nasal valves, simplifying the calculation process and making the implementation of the solution more efficient.

[0027] In one embodiment of the present invention, calculating the image gradient modulus within a local observation area and searching for the path that maximizes the gradient energy line integral to determine the external nasal valve boundary curve includes: Receive standardized frame sequence With local observation area ; In the local observation area Internally calculate the spatial gradient of each frame of the normalized image: ;

[0028] in and These are the partial derivatives with respect to the horizontal and vertical coordinates, respectively. Calculate image gradient magnitude : ;

[0029] Define candidate curve family : ;

[0030] in For arc length parameterized curves, For arc length parameters, The total arc length of the curve. This indicates a constraint on the magnitude of the unit tangent vector; Construct the objective functional Target functional The line integral of the image gradient magnitude along the candidate curve: ;

[0031] Determine the boundary curve of the external nasal valve : ;

[0032] Output external nose valve boundary curve With image gradient mode .

[0033] It should be noted that the partial derivative of the horizontal coordinate is the rate of change of the standardized image gray value with respect to the horizontal coordinate, reflecting the degree of gray value change in the horizontal direction. The partial derivative of the vertical coordinate is the rate of change of the standardized image gray value with respect to the vertical coordinate, reflecting the degree of gray value change in the vertical direction. The image gradient magnitude is the arithmetic square root of the sum of the squares of the partial derivatives of the horizontal and vertical coordinates, reflecting the strength of local gray value changes in the image. The candidate curve family is the set of all curves within the local observation area that satisfy the continuous differentiability condition and the unit tangent vector magnitude constraint, reflecting the candidate range for the extraction of the external nasal valve boundary. The continuous differentiability condition is the constraint that the candidate curve has a derivative at any point in the domain and that the derivative is continuous, reflecting the smoothness of the curve. The unit tangent vector magnitude constraint is the constraint that the length of the tangent vector of the candidate curve is always equal to 1, reflecting that the curve is parameterized by arc length. The target functional is the path integral of the image gradient magnitude along the candidate curve, reflecting the overall edge strength of the candidate curve. The gradient energy line integral is the specific calculated value of the target functional, reflecting the edge energy magnitude of the candidate curve. The boundary curve of the external nasal valve is the curve with the largest gradient energy line integral among the candidate curve family, reflecting the true geometric shape of the outer edge of the external nasal valve.

[0034] It should be noted that the partial derivatives of the horizontal and vertical coordinates are calculated using the Sobel operator, including: Step 1, converting the normalized frame into an 8-bit grayscale image; Step 2, defining the horizontal and vertical Sobel operator templates; Step 3, for each pixel in the local observation area, taking a 3×3 pixel region centered on that point, multiplying the pixel value within the region by the corresponding position in the operator template, and summing the results to obtain the horizontal or vertical partial derivative at that point; Step 4, normalizing the calculated partial derivative values ​​to the range of 0 to 255. The specific generation method of the candidate curve family within the local observation area includes: Step 1, uniformly selecting 50 to 200 starting points at the radius boundary within the observation area, and selecting the same number of ending points at the outer radius boundary; Step 2, using cubic spline interpolation to generate curves connecting the starting and ending points, ensuring the continuity of the first derivative of the curves; Step 3, parameterizing the arc length of the generated curves so that the magnitude of the tangent vector is always equal to 1. The preferred number of curves is 50 to 200. Fewer than 50 curves may miss the true boundary, while more than 200 curves will increase computation time and reduce the efficiency of the scheme.

[0035] It should be noted that this invention highlights edge regions within the local observation area by calculating the gradient modulus of the image, constructs a family of candidate curves that satisfy constraints, and uses the maximum value of the gradient energy line integral to filter the true boundary curve of the external nasal valve. This provides a continuous and stable geometric basis for subsequent curvature calculation, eliminating interference from irrelevant regions and noise. Specifically, this invention uses the Sobel operator to calculate the gradient modulus, which effectively enhances edge features and suppresses noise. The dual constraints of the candidate curve family ensure the smoothness and parameter uniformity of the curves, avoiding errors introduced by irregular curves. By traversing the candidate curves and filtering the curve corresponding to the maximum integral value, the overall edge information of the curves can be integrated, improving the robustness of boundary extraction. This ensures that the extracted boundary curve is consistent with the true anatomical structure of the external nasal valve.

[0036] In one embodiment of the present invention, the geometric curvature of the boundary curve of the external nasal valve is calculated, and the second time derivative of the curvature is obtained using central difference operation to quantify the acceleration level change of the boundary shape, including: Receive external nasal valve boundary curve sequence The arc length parameterization is expressed as ,in This is the arc length parameter, with values ​​ranging from zero to the total arc length. Discrete time satisfies ; Calculate geometric curvature : ;

[0037] in These are the first derivatives of the horizontal and vertical coordinates with respect to the arc length parameter, respectively. These are the second derivatives of the horizontal and vertical coordinates with respect to the arc length parameter, respectively. The second time derivative of curvature is obtained using central difference operations. : ;

[0038] in The time interval between adjacent frames. Let be the geometric curvature at the next sampling time. The geometric curvature at the previous sampling time; Output geometric curvature The time second derivative with curvature The spatiotemporal sequence.

[0039] It should be noted that the total arc length is the sum of the lengths of the external nasal valve boundary curves, reflecting the spatial scale characteristics of the curve. A preferred value is 50 to 150 pixels; this range matches the typical pixel length of the external nasal valve boundary in adults, covering the size differences of different face shapes. The arc length parameter range is the numerical interval of the arc length parameter, reflecting the parameter's coverage of the boundary curve. The first derivative of the lateral coordinate is the rate of change of the lateral coordinate of the boundary curve with respect to the arc length parameter, reflecting the lateral tangent direction of the curve at that location. The first derivative of the longitudinal coordinate is the rate of change of the longitudinal coordinate of the boundary curve with respect to the arc length parameter, reflecting the longitudinal tangent direction of the curve at that location. The second derivative of the lateral coordinate is the rate of change of the first derivative of the lateral coordinate with respect to the arc length parameter, reflecting the change in the lateral curvature trend of the curve. The second derivative of the longitudinal coordinate is the rate of change of the first derivative of the longitudinal coordinate with respect to the arc length parameter, reflecting the change in the longitudinal curvature trend of the curve. Geometric curvature is an intrinsic geometric quantity describing the local curvature of the boundary curve, reflecting the morphological characteristics of the external nasal valve boundary. Central difference operation is a numerical method for calculating the second derivative using three adjacent frames of data. The second derivative of curvature over time is the quadratic rate of change of geometric curvature with respect to time, reflecting the acceleration characteristics of the morphological changes of the external nasal valve boundary. The spatiotemporal sequence is a data set of geometric curvature and its second derivative over time along the spatial arc length and time dimension, reflecting the dynamic change process of the external nasal valve boundary morphology.

[0040] It should be noted that the first and second derivatives of the boundary curve coordinates with respect to the arc length parameter are obtained using a cubic spline interpolation algorithm. This includes: Step 1, sorting the discrete coordinate points of the boundary curve according to the arc length parameter; Step 2, constructing a system of interpolation equations for the cubic spline function to ensure the continuity of the function at the sampling points and the continuity of its first and second derivatives; Step 3, solving the system of equations to obtain the coefficients of the spline function; and Step 4, calculating the first and second derivatives of the spline function to obtain the derivative values ​​corresponding to any arc length parameter. Furthermore, the core of the parameterization method for the boundary curve using arc length parameterization is to eliminate the influence of coordinate system scaling and rotation, using the curve's own length as the parameter. In practice, the boundary curve is first discretized into 100 to 200 uniform sampling points, the Euclidean distance between adjacent sampling points is calculated, and the cumulative arc length is obtained by summing these distances. Then, using the cumulative arc length as the independent variable, cubic spline interpolation is used to obtain the coordinate values ​​corresponding to any arc length parameter.

[0041] It should be noted that this invention performs arc-length parameterization on the boundary curve, calculates the geometric curvature of the curve, and then solves for the second time derivative of the curvature through central difference operation to form a spatiotemporal sequence. This quantifies the acceleration level change of the external nasal valve boundary morphology, capturing the rapid deformation characteristics during inhalation. Specifically, this invention uses arc-length parameterization to eliminate the influence of coordinate system scaling and rotation, ensuring the intrinsic nature of the curvature. Cubic spline interpolation is used to calculate the derivative, improving the stability of the curvature solution. Central difference operation balances computational accuracy and noise suppression, accurately extracting the acceleration characteristics of morphological changes. This provides reliable kinetic data for subsequent energy aggregation, ensuring that the characteristics for liveness detection have physiological significance, while simplifying the data processing flow and improving the feasibility of the solution.

[0042] In one embodiment of the present invention, energy aggregation and normalization are performed on the second time derivative of curvature in the spatiotemporal domain to calculate the second energy density of the external nasal valve collapse curvature. The second energy density of the external nasal valve collapse curvature is compared with a preset threshold to output a liveness determination result, including: Time second derivative of the received curvature ,in This is the arc length parameter, and its value ranges from zero to the total arc length of the outer nose valve boundary. Discrete time satisfies , The time interval between adjacent frames. This is a time index, with values ​​ranging from zero to... ; Calculate the effective time window length : ;

[0043] in This indicates the number of time steps that can be used for central difference. Calculate the second-order energy density of the collapsed curvature of the external nasal valve. : ;

[0044] in The energy is the square of the time second derivative of the curvature. This represents the integral over the arc length. This represents the discrete integral with respect to time. For spatial and temporal normalization factors; Set preset threshold The following comparison logic is executed: ;

[0045] Output the liveness determination result.

[0046] It should be noted that the effective time window length is the length of the time interval for central difference calculation, reflecting the effective analysis duration of the second derivative of curvature in time. A preferred value is 2 to 4 seconds, meaning this range covers the dynamic curvature changes during a single natural nasal inhalation, excluding interference from invalid time frames at the beginning and end. The energy term is the square of the second derivative of curvature in time, reflecting the energy characteristics of the acceleration of the morphological changes at the external nasal valve boundary. The total energy value is the sum of the energy term after spatial arc-length integration and time discrete integration, reflecting the total energy intensity in the spatiotemporal domain. The second-order energy density of the external nasal valve collapse curvature is the result of spatiotemporal normalization of the total energy value, reflecting the energy intensity per unit spacetime. The preset threshold is the critical value used to determine whether the individual is alive. The result of the liveness determination is the output conclusion after comparing the energy density with the preset threshold.

[0047] It should be noted that the core of energy aggregation and normalization of the temporal second derivative of curvature in the spatiotemporal domain is to integrate energy information in both spatial and temporal dimensions step by step, and then eliminate individual differences through normalization. In practice, the energy term is first integrated along the total arc length of the outer nasal valve boundary to integrate the energy distribution in the spatial dimension. Then, the integral result is discretely integrated over time within the effective time window to integrate the energy changes in the temporal dimension. Finally, normalization is performed using the product of the total arc length and the effective time window length. For example, with a total arc length of 80 pixels, an effective time window of 3 seconds, a spatial integration result of 50, a temporal integration result of 30, and a total energy value of 1500, the normalized energy density is 1500 divided by 240, resulting in 6.25. Setting the square of the temporal second derivative of curvature as the energy term is key to converting the acceleration characteristics of morphological changes into non-negative energy characteristics. This preserves the amplitude of change while avoiding the loss of energy characteristics due to the cancellation of positive and negative values.

[0048] It should be noted that the number of time steps available for center difference within the effective time window is equal to the total number of acquired frames minus 2. This is because center difference requires data from the previous and next frames of the current frame; center difference cannot be calculated between the first and last frames. For example, if the total number of acquired frames is 101 and the number of time steps is 99, combined with the 0.03-second time interval between adjacent frames, the effective time window length is 99 multiplied by 0.03, resulting in 2.97 seconds. The energy term integral along the total arc length of the external nose valve boundary can be calculated using the trapezoidal integration method. For example, the external nose valve boundary can be discretized into 100 to 200 sampling points, the arc length increment of adjacent sampling points can be calculated, the energy term of adjacent sampling points can be averaged and multiplied by the arc length increment, and finally, all segmented integral values ​​can be summed to obtain the total spatial integral value. The accuracy control method is to set the number of sampling points to 150, which balances integration accuracy and computational efficiency. The method for calibrating the preset threshold includes: Step 1: Collect 500 to 1000 live samples and an equal number of non-live samples; Step 2: Calculate the second-order energy density of the external nasal valve collapse curvature of all samples; Step 3: Calculate the mean and standard deviation of the energy density of the live samples. The preset threshold is the mean minus 1 standard deviation. For example, if the mean of the live samples is 6 and the standard deviation is 1, the preset threshold is 5, which can cover more than 95% of the live samples.

[0049] It should be noted that this invention performs energy aggregation and normalization on the second-order time derivative of curvature in the spatiotemporal domain to calculate the second-order energy density of the external nasal valve collapse curvature, and then compares it with a preset threshold to achieve liveness detection. This transforms the dynamic change characteristics of the external nasal valve boundary morphology into a unified and comparable energy density index, eliminating the influence of different individuals and acquisition conditions. Specifically, this invention uses square operations to construct energy terms, highlighting significant change characteristics; integrates spatiotemporal energy step by step to ensure information integrity; normalization processing ensures the uniformity of the judgment index; and the preset threshold is based on sample statistical calibration to ensure the rationality of the judgment logic. This makes the liveness detection results unaffected by the size of the external nasal valve and the acquisition time, with clear judgment criteria and easy engineering implementation, providing stable and reliable liveness detection support for unmanned self-check-in in hotels.

[0050] In one embodiment of the present invention, only when the liveness determination result is passed, the frame with the maximum gradient energy line integral is selected to extract the face vector, the face vector is bound to the room check-in record and an access token is generated, including: Receive the liveness detection result; if the liveness detection result is passed, proceed to the calculation process. Calculate the gradient energy line integral along the boundary curve of the external nasal valve for each frame. : ;

[0051] in For image gradient magnitude, The arc length parameterization of the boundary curve of the external nasal valve is used. This is the arc length parameter, with values ​​ranging from zero to the total arc length at the boundary. ; Select the frame with the largest gradient energy line integral. Frame number satisfy: ;

[0052] in The last frame number within the acquisition window; Calculate face vectors : ;

[0053] in For feature mapping function, For the real number field dimensional vector; Build registration records :

[0054] Generate permission token : ;

[0055] in This is an encoding mapping function; it outputs a face vector. Registration records and permission tokens .

[0056] It should be noted that the clearest frame is the single frame image with the largest gradient energy line integral value in the normalized frame sequence. The feature mapping function is an algorithmic model used to extract key facial features. The face vector is the real-domain multi-dimensional feature data output by the feature mapping function, preferably a 512-dimensional real-number vector. The registration record is a structured data set integrating check-in identity and check-in information, reflecting the user's check-in authorization information, preferably structured data containing the check-in identifier, face vector, room number, and valid time period. The check-in identifier is unique information used to identify the check-in user, which can be obtained by reading the ID card number collected by an ID card reader or the mobile phone number entered by the user. The room number is the unique identifier code of the hotel room, reflecting the location of the room booked by the user, which can be obtained by reading the user reservation data from the hotel reservation management system. The valid time period is the time interval of the user's room stay, reflecting the valid scope of the check-in authorization, which can be obtained by reading the reservation check-in time and check-out time from the hotel reservation management system. The encoding mapping function is an algorithm that encrypts and transforms the registration record. The authorization token is an encrypted string output by the encoding mapping function, reflecting the user's check-in authorization certificate, preferably a hexadecimal string encrypted with the Advanced Encryption Standard (AES) algorithm.

[0057] It should be noted that the core of the feature extraction method using a feature mapping function to extract real-domain multi-dimensional face vectors from the clearest frame is to convert the face image into a high-dimensional feature vector through a deep learning model, thereby achieving a digital representation of the face. In practice, the image size of the clearest frame is adjusted to 112×112, and input into a pre-trained deep residual convolutional neural network model. The model extracts features through convolutional and pooling layers, ultimately outputting a 512-dimensional real-valued face vector. The preferred algorithm for the feature mapping function is a 50-layer deep residual convolutional neural network model with parameters including an input image size of 112x112, 3 input channels, a 3x3 kernel size, a linear rectified activation function, batch normalization, and an output layer dimension of 512. In practice, this model is used to perform forward propagation calculations on the clearest frame, and the output 512-dimensional vector is the face vector, which will not be elaborated further here. The specific encoding rules of the encoding mapping function adopt the Advanced Encryption Standard (AES-128) algorithm, with a key length of 128 bits and a block cipher mode of chained cipher blocks. The encryption logic is as follows: first, serialize the registration record into a string in Extensible Markup Language (XML) format; second, convert the string into a UTF-8 encoded byte stream; third, encrypt the byte stream in blocks using a preset 128-bit key; and fourth, convert the encrypted byte stream into a hexadecimal string to form an access token.

[0058] It should be noted that the standardized storage format for registration records adopts Extensible Markup Language (XML), with guest identification, facial vector, room number, and valid time period set as different label fields. Data verification rules use a hash algorithm; the hash value of the registration record is calculated and stored along with the record. During subsequent retrievals, the hash value is recalculated and compared; if they do not match, the record is considered tampered with. Furthermore, the validity verification process for the access token is as follows: first, the access token is read; second, the token is decrypted using a preset key to obtain the registration record; third, the valid time period in the registration record is verified to include the current time; and fourth, the facial vector is verified to match the user's current facial features. If all four steps pass, the record is considered valid. Invalidity is triggered by the expiration of the valid time period, the user completing check-out procedures, or the access token failing verification more than three times.

[0059] It should be noted that this invention only extracts facial vectors from frames with the clearest edge features after a successful liveness detection, binds them to core check-in information, and encrypts and generates an access token. This ensures the accuracy of facial feature extraction, avoids invalid processing of non-liveness data, and protects the privacy and security of check-in information. Specifically, this invention improves the robustness of facial vector extraction by selecting the clearest frames through gradient energy line integration. The feature mapping function uses a deep learning model to ensure the discriminative power of facial features. Registration records integrate key check-in information, realizing the association between identity and permissions. The encryption processing of the encoding mapping function ensures the security of data transmission and storage. This forms a complete closed-loop access control system for unmanned self-service check-in, simplifies the operation process, and improves system efficiency and user experience.

[0060] In one embodiment of the present invention, an unmanned hotel management system includes: The standardized frame calculation module collects video sequences when prompting the user to inhale, detects the tip of the nose, the philtrum, and the glabella as face stabilization reference points, and uses least squares similarity transformation to uniformly align the video sequence to the first frame coordinate system to obtain a standardized frame. The local observation area delineation module delineates the local observation area, including the external nasal valve structure, in a standardized frame based on the anatomical proportions of the face, ensuring that the anatomical position of the observed object remains constant. The external nose valve boundary curve determination module calculates the image gradient modulus within the local observation area and searches for the path that maximizes the gradient energy line integral to determine the external nose valve boundary curve. The geometric curvature calculation module calculates the geometric curvature of the boundary curve of the external nasal valve and uses the central difference operation to obtain the time second derivative of the curvature in order to quantify the acceleration level change of the boundary shape. The second-order energy density calculation module performs energy aggregation and normalization on the time second derivative of curvature in the spatiotemporal domain, calculates the second-order energy density of the external nasal valve collapse curvature, and compares the second-order energy density of the external nasal valve collapse curvature with a preset threshold to output the liveness determination result. The access token generation module extracts the face vector from the frame with the maximum gradient energy line integral only when the liveness detection result is passed. It then binds the face vector to the room check-in record and generates an access token.

[0061] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0062] This invention ensures that the user's consent has been obtained on the spot when collecting facial data through interface text, voice prompts, or user-cooperative actions, and that the data is used only for identity verification and liveness detection of the personnel entering the site, which complies with the principle of legitimacy and necessity under the Personal Information Protection Law.

[0063] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A method for unmanned hotel management, characterized in that, Includes the following steps: Step S101: Collect the video sequence when prompting the user to inhale, detect the tip of the nose, the philtrum and the glabella as the face stabilization reference points, and use the least squares similarity transformation to uniformly align the video sequence to the first frame coordinate system to obtain a standardized frame. Step S102: Based on the facial anatomy proportions, delineate the local observation area containing the external nasal valve structure in the standardized frame to ensure that the anatomical position of the observed object remains constant. Step S103: Calculate the image gradient modulus within the local observation area and search for the path that maximizes the gradient energy line integral to determine the boundary curve of the external nose valve. Step S104: Calculate the geometric curvature of the boundary curve of the external nasal valve, and use the central difference operation to obtain the time second derivative of the curvature in order to quantify the acceleration level change of the boundary shape. Step S105: In the spatiotemporal domain, the second time derivative of curvature is aggregated and normalized to calculate the second energy density of the external nasal valve collapse curvature. The second energy density of the external nasal valve collapse curvature is compared with a preset threshold to output the liveness determination result. Step S106: Only when the liveness determination result is passed, select the frame with the maximum gradient energy line integral to extract the face vector, bind the face vector with the room check-in record and generate an access token. Receive the standardized frame sequence and the first frame nose tip, and set the horizontal unit vector in the first frame coordinate system; The inner radius, outer radius, inner angle, and outer angle are set according to the anatomical proportions of the human face. The inner radius is limited to be smaller than the outer radius, and the inner angle is limited to be smaller than the outer angle and less than 90 degrees. Define a local observation region in the first frame coordinate system, and stipulate that the pixels in the local observation region satisfy distance constraints and angle constraints. The distance constraint stipulates that the Euclidean distance between the pixel and the nose tip in the first frame is between the inner radius and the outer radius; the angle constraint stipulates that the angle between the line vector connecting the pixel to the nose tip in the first frame and the horizontal unit vector is within the positive angle range or the negative angle range determined by the inner angle and the outer angle. Output the local observation area for use in calculating the boundary curve of the external nose valve; The time second derivative sequence of the received curvature is used to determine the total arc length of the external nose valve boundary and the time interval between adjacent frames. The effective time window length is defined as the product of the number of time steps for center difference and the time interval between adjacent frames; the square of the time second derivative of curvature is calculated as the energy term. First, integrate the energy term along the total arc length of the outer nose valve boundary, and then perform discrete integration of the integration result with respect to time within the effective time window to obtain the total energy value. Divide the total energy value by the product of the total arc length of the outer nose valve boundary and the effective time window length to complete the spatial and temporal normalization process and obtain the second-order energy density of the collapse curvature of the outer nose valve. The second-order energy density of the collapsed curvature of the external nasal valve is compared with a preset threshold. If the second-order energy density of the collapsed curvature of the external nasal valve is greater than or equal to the preset threshold, a pass result is output. If the second-order energy density of the collapsed curvature of the external nasal valve is less than the preset threshold, a rejection result is output.

2. The unmanned hotel management method according to claim 1, characterized in that, Establish a time series containing multiple sampling times, set the time interval between adjacent frames to a fixed value, set the total acquisition time to the acquisition window length, and obtain the original frame set composed of the image intensity function; In the original frame image corresponding to each sampling time, locate the tip of the nose, the philtrum, and the glabella, and define the tip of the nose, the philtrum, and the glabella as two-dimensional plane coordinate vectors; Construct a similarity transformation model that includes scale parameters, a two-dimensional rotation matrix, and a translation vector; The goal is to minimize the sum of squared Euclidean distances between the transformed tip of the nose, philtrum, and glabella at the current sampling time and the reference point corresponding to the initial sampling time. The optimal similarity transformation parameters are then solved. The two-dimensional rotation matrix must satisfy the orthogonality constraint and the constraint that the determinant is one. By using the inverse mapping of the similarity transformation model, the original frame image at the current sampling time is subjected to coordinate inverse transformation and resampling to generate a standardized frame sequence aligned with the coordinate system at the initial sampling time.

3. The unmanned hotel management method according to claim 1, characterized in that, Receive standardized frame sequences and local observation areas; For each frame of standardized image within the local observation area, calculate the partial derivatives of the horizontal and vertical coordinates, and then calculate the arithmetic square root of the sum of the squares of the partial derivatives of the horizontal and vertical coordinates to obtain the image gradient magnitude. Define a family of candidate curves located within a local observation region, and constrain the curves in the family to satisfy the condition of continuous differentiability and the unit tangent vector magnitude constraint. Construct a target functional, which is defined as the path integral of the image gradient magnitude along the candidate curve; Traverse the candidate curve family to search for the curve that maximizes the target functional value, and determine the curve that maximizes the target functional value as the external nasal valve boundary curve; output the external nasal valve boundary curve and the image gradient modulus.

4. The unmanned hotel management method according to claim 1, characterized in that, Receive the boundary curve sequence of the external nasal valve, use arc length parameterization to describe the boundary curve, determine the total arc length and the range of arc length parameter values, and use time discretization settings. For each sampling time, calculate the first and second derivatives of the boundary curve coordinates with respect to the arc length parameter; construct a geometric curvature calculation formula, with the numerator set as the product of the first derivative of the horizontal coordinate and the second derivative of the vertical coordinate minus the product of the first derivative of the vertical coordinate and the second derivative of the horizontal coordinate, and the denominator set as the cube of the sum of the squares of the first derivatives of the horizontal coordinate and the squares of the first derivatives of the vertical coordinate; obtain the geometric curvature according to the geometric curvature calculation formula.

5. The unmanned hotel management method according to claim 4, characterized in that, Calculate the time second derivative of geometric curvature using central difference operations; Calculate the sum of the geometric curvature at the next sampling time and the geometric curvature at the previous sampling time, subtract twice the geometric curvature at the current sampling time, and divide the result by the square of the time interval between adjacent frames; define the calculation result as the time second derivative of curvature; output the spatiotemporal sequence of geometric curvature and the time second derivative of curvature.

6. The unmanned hotel management method according to claim 1, characterized in that, Receive the liveness detection result. If the liveness detection result is passed, proceed with the following steps. For each frame in the standardized frame sequence, the path integral of the image gradient magnitude along the boundary curve of the external nose valve is calculated using the arc length parameterization of the external nose valve boundary curve. The path integral is defined as the gradient energy line integral. Traverse all frames within the acquisition window and select the frame with the largest gradient energy line integral value. The frame with the largest gradient energy line integral value is determined as the clearest frame. The feature mapping function is used to extract features from the clearest frame to generate a multi-dimensional face vector in the real number domain. A registration record is constructed, which includes the occupant's identifier, facial vector, room number, and valid time period; the registration record is encoded using an encoding mapping function to generate an access token; Output face vectors, registration records, and permission tokens.

7. An unmanned hotel management system, characterized in that, Implementing an unmanned hotel management method as described in any one of claims 1 to 6, comprising: The standardized frame calculation module collects video sequences when prompting the user to inhale, detects the tip of the nose, the philtrum, and the glabella as face stabilization reference points, and uses least squares similarity transformation to uniformly align the video sequence to the first frame coordinate system to obtain a standardized frame. The local observation area delineation module delineates the local observation area, including the external nasal valve structure, in a standardized frame based on the anatomical proportions of the face, ensuring that the anatomical position of the observed object remains constant. The external nose valve boundary curve determination module calculates the image gradient modulus within the local observation area and searches for the path that maximizes the gradient energy line integral to determine the external nose valve boundary curve. The geometric curvature calculation module calculates the geometric curvature of the boundary curve of the external nasal valve and uses the central difference operation to obtain the time second derivative of the curvature in order to quantify the acceleration level change of the boundary shape. The second-order energy density calculation module performs energy aggregation and normalization on the time second derivative of curvature in the spatiotemporal domain, calculates the second-order energy density of the external nasal valve collapse curvature, and compares the second-order energy density of the external nasal valve collapse curvature with a preset threshold to output the liveness determination result. The access token generation module extracts the face vector from the frame with the maximum gradient energy line integral only when the liveness detection result is passed. It then binds the face vector to the room check-in record and generates an access token.

Images