Case handling center personnel identity rapid identification system and method

Abstract

The invention relates to the technical field of identity recognition, and discloses a case handling center personnel identity rapid recognition system and method. The method comprises the following steps: firstly, acquiring a facial image sequence, a fingerprint streakline sequence and voiceprint frequency monitoring data of a target person in a case handling scene to form a biological characteristic data set; carrying out multi-modal feature fusion processing on the set to generate an identity feature identifier containing a facial contour matching degree, a fingerprint streakline coincidence rate and voiceprint frequency consistency; calling a pre-trained identity recognition model to perform multi-dimensional matching analysis on the identity feature identifier to obtain an identity confirmation result and an abnormal feature region identifier; performing scene factor dynamic calibration processing on the identity confirmation result based on the association relationship between the illumination intensity sequence and the equipment acquisition parameters; and finally, generating a personnel management strategy set containing a feature acquisition position adjustment scheme and a database updating optimization scheme according to the abnormal feature region identifier.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of identity recognition, in particular to a case-handling center personnel identity rapid recognition system and method. BACKGROUND

[0002] In the daily operation of the case-handling center, personnel identity recognition is an important link to ensure the safety of the place and improve work efficiency. Traditional identity recognition methods rely on a single biological feature, such as face image or fingerprint recognition. However, this single feature recognition mode faces many problems in practical application. The environment of the case-handling center is complex and variable, and the lighting conditions are often unstable. Strong light, backlight or dim environment will interfere with face image acquisition, resulting in blurred image and inaccurate feature extraction. At the same time, the difference in device acquisition parameters will also affect the data quality. Different types of acquisition devices may output biological feature data in different formats, increasing the difficulty of recognition. When relying only on fingerprint recognition, if the person's fingers have stains, damage or dryness, the fingerprint line collection will not be complete, reducing the accuracy of the line overlap rate judgment. Voiceprint recognition is easily affected by background noise. The sound of personnel talking, device running and other sounds in the case-handling center will interfere with the monitoring of voice frequency, causing deviation in frequency consistency analysis. Traditional recognition methods lack dynamic calibration mechanism and cannot adjust the recognition strategy in real time according to the changes of scene factors. When the recognition result is doubtful, it is difficult to quickly locate the abnormal feature area, and it is also difficult to generate a targeted personnel management strategy. The update and maintenance of the database also relies on manual operation, which is low in efficiency and difficult to optimize according to the actual collection situation, resulting in deviation between the feature data in the database and the actual personnel features, further affecting the recognition effect. The matching analysis dimension of the existing identity recognition model is limited, and it can only perform simple feature comparison, cannot comprehensively analyze multiple modal features, and is prone to misjudgment or omission. These problems together result in low accuracy of traditional identity recognition methods, frequent identity confirmation delay or error, and great disturbance to personnel management in the case-handling center. SUMMARY

[0003] The purpose of the present application is to provide a case-handling center personnel identity rapid recognition system and method to solve the problems raised in the background.

[0004] To achieve the above purpose, the present application provides a case-handling center personnel identity rapid recognition method, which comprises: Collecting a set of biological feature data of the target personnel in the case-handling scene, the set of biological feature data including a sequence of face images, a sequence of fingerprint lines and voice frequency monitoring data; perform multi-modal feature fusion processing on the set of biometric data to generate an identity feature identifier of the target person, the identity feature identifier including a facial contour matching degree, a fingerprint line overlap rate, and a voiceprint frequency consistency; invoke a pre-trained identity recognition model to perform multi-dimensional matching analysis processing on the identity feature identifier to generate an identity confirmation result and an abnormal feature region identifier of the target person; perform scene factor dynamic calibration processing on the identity confirmation result to generate a calibrated identity confirmation result, the scene factor dynamic calibration processing being implemented based on an association between the sequence of illumination intensities and device acquisition parameters; generate a set of personnel management strategies according to the abnormal feature region identifier, the set of personnel management strategies including a feature acquisition position adjustment scheme and a database update optimization scheme.

[0005] Preferably, the multi-modal feature fusion processing on the set of biometric data to generate the identity feature identifier of the target person includes: dividing the sequence of facial images into a plurality of feature subsequences according to time windows, each feature subsequence corresponding to an acquisition period; for each feature subsequence, performing the following processing: constructing a multi-modal feature topology of the target person according to the voiceprint frequency monitoring data, the multi-modal feature topology including spatial distribution data of a facial angle field, a fingerprint contact field, and a voiceprint energy field; coupling and analyzing the multi-modal feature topology and the feature subsequence to generate a feature fusion result of a current time window, the feature fusion result including a spatial distribution matrix of facial contour feature quantities, fingerprint line feature quantities, and voiceprint frequency feature quantities; performing weighted superposition processing on the feature fusion results of a plurality of consecutive time windows to calculate the facial contour matching degree, the fingerprint line overlap rate, and the voiceprint frequency consistency; wherein the facial contour matching degree is a maximum similarity value of facial contour feature quantities along a feature extraction direction, the fingerprint line overlap rate is an overlap proportion of fingerprint line feature quantities in a normal direction of a contact area, the voiceprint frequency consistency is a ratio of a variance to a mean value of voiceprint frequency feature quantities within a predetermined time interval.

[0006] Preferably, the coupling and analyzing of the multi-modal feature topology and the feature subsequence to generate the feature fusion result of the current time window includes: Based on the corresponding relationship between the face angle field and the face posture component in the feature subsequence, an angle-feature mapping equation is established, and a first distribution function of the face contour feature quantity is obtained by solving the angle-feature mapping equation; According to the correlation characteristics of the fingerprint contact field and the contact pressure, a fingerprint contact feature calculation model is constructed, which contains the dynamic correction parameters of the skin elasticity coefficient and the sensitivity of the collection device; Combined with the spatial variation rate of the voiceprint energy field and the environmental noise coefficient, a voiceprint feature iterative calculation process is established, which contains a feedback correction mechanism of energy increment and voiceprint feature increment; The output results of the first distribution function, the fingerprint contact feature calculation model and the voiceprint feature iterative calculation process are subjected to spatial interpolation fusion processing to generate multi-modal feature distribution data containing face, fingerprint and voiceprint feature quantities.

[0007] Preferably, the pre-trained identity recognition model is called to perform multi-dimensional matching analysis processing on the identity feature identification to generate an identity confirmation result and an abnormal feature area identification of the target personnel, including: The face contour matching degree is input into the first feature analysis layer of the identity recognition model, and the distribution coordinates and the similarity change curve of the face feature difference area are determined through a feature difference factor calculation module; The fingerprint line coincidence rate is input into the second feature analysis layer of the identity recognition model, and a contact feature matching calculation is performed to generate a feature matching probability and a matching deviation prediction value of the fingerprint contact area; The voiceprint frequency consistency is input into the third feature analysis layer of the identity recognition model, and the feature offset amount and the feature stability evolution data of the voiceprint frequency are calculated based on a voiceprint feature degradation model; The similarity change curve, the feature matching probability and the feature offset amount are fused to generate a comprehensive matching index of the target personnel, and the identity confirmation result is determined according to the comparison result of the comprehensive matching index and a preset recognition threshold; Based on the spatial superposition result of the distribution coordinates, the matching deviation prediction value and the feature stability evolution data, the geometric positions of the feature difference area, the matching deviation path and the feature unstable area are identified.

[0008] Preferably, the identity confirmation result is subjected to scene factor dynamic calibration processing to generate a calibrated identity confirmation result, including: Extreme illumination values and illumination change frequencies in the illumination intensity sequence are extracted, and a dynamic adjustment amount of the device collection parameters with illumination change is calculated; According to the dynamic adjustment amount, the face contour matching degree is calculated for illumination feature compensation, to generate a calibrated face contour matching degree; Based on the correlation between the illumination change frequency and the feature collection stability, the fingerprint line coincidence rate is collected for deviation correction processing, to generate a calibrated fingerprint line coincidence rate; According to the device noise change data under extreme illumination, the voiceprint frequency consistency is processed for feature anti-noise adaptation adjustment, to generate a calibrated voiceprint frequency consistency; The calibrated face contour matching degree, fingerprint line coincidence rate and voiceprint frequency consistency are input into the identity recognition model for recalculation, to generate an identity confirmation result after calibration of scene factors.

[0009] Preferably, the face contour matching degree is calculated for illumination feature compensation according to the dynamic adjustment amount, to generate a calibrated face contour matching degree, comprising: Obtain the initial collection parameters of the target personnel under the reference illumination and the dynamic adjustment amount, and establish a collection parameter-illumination correlation function; Calculate the face feature compensation amount according to the collection parameter-illumination correlation function, wherein the face feature compensation amount is the product of the illumination change amount and the collection parameter change amount; Superimpose the face feature compensation amount into the calculation process of the face contour matching degree, to generate a face contour matching degree calibration value containing illumination influence; The face contour matching degree calibration value is processed for noise suppression effect compensation, which is realized based on the product factor of the device noise curve and the illumination holding time.

[0010] Preferably, the personnel management strategy set is generated according to the abnormal feature area identification, comprising: For the identification of the feature difference area, calculate the optimal collection position, which is realized by adjusting the coverage range proportion of adjacent collection devices; According to the identification of the matching deviation path, construct a database update optimization scheme, which contains the selection of feature sample supplement area and the optimization configuration of sample feature parameters; Based on the identification of the feature unstable area, generate a collection frequency adjustment strategy, which dynamically adjusts the collection interval and collection time length according to the feature stability prediction value; The optimal collection position, the database update optimization scheme and the collection frequency adjustment strategy are processed for priority sorting, to generate a management strategy set containing execution timing and implementation parameters.

[0011] Preferably, the database update optimization scheme is constructed, comprising: extracting geometric features of the matching deviation path, calculating a path bending radius and a deviation direction angle; selecting a coverage density of the feature sample supplement according to the bending radius, the coverage density being inversely proportional to the bending radius; adjusting a collection direction of a sample feature parameter based on the deviation direction angle, so that the collection direction forms a predetermined included angle with the deviation direction; dynamically adjusting a sample supplement frequency according to feature collection accuracy test data, to ensure that the supplement sample is below a feature library capacity threshold value; generating an optimized parameter configuration table including the coverage density, the collection direction and the supplement frequency.

[0012] Preferably, the method further comprises: collecting actual recognition results and feature matching deviations of the target person within a preset calibration period; performing deviation analysis processing on the actual recognition results and the predicted recognition results to generate a first error correction coefficient; performing time domain comparison processing on the feature matching deviations and the predicted deviation values to generate a second error correction coefficient; adjusting weight parameters of the identity recognition model according to the first error correction coefficient and the second error correction coefficient to generate an optimized identity recognition model; applying the optimized identity recognition model to subsequent batches of personnel identity recognition tasks.

[0013] Preferably, the present application further comprises a case handling center personnel identity rapid identification system, the system comprising a processor and a memory, the memory and the processor being connected, the memory being used for storing programs, instructions or codes, and the processor being used for executing the programs, instructions or codes in the memory to realize the case handling center personnel identity rapid identification method as described above.

[0014] Compared with the prior art, the present application has the following advantages: By collecting the face image sequence, fingerprint line sequence and voiceprint frequency monitoring data and other multi-modal biological characteristics of the target person, the personnel identity is comprehensively captured. The identity feature identifier generated by the multi-modal feature fusion processing covers the face contour matching degree, the fingerprint line coincidence rate and the voiceprint frequency consistency, and constructs a feature system of the personnel identity from multiple angles, reducing the limitations of single feature recognition.

[0015] The pre-trained identity recognition model can perform multi-dimensional matching analysis on the identity feature identification, not only generating an identity confirmation result, but also identifying abnormal feature areas, making the recognition process more accurate and traceable. In complex case handling scenarios, changes in light intensity and device acquisition parameters will affect the recognition result, and the scene factor dynamic calibration process is based on the correlation between light intensity sequence and device acquisition parameters to dynamically adjust the identity confirmation result, making the recognition result more consistent with the actual scene, avoiding misjudgment caused by environmental factors. The personnel management strategy set generated based on the abnormal feature area identification contains feature collection position adjustment schemes and database update optimization schemes, which can solve the problems that occur in the collection process. When an abnormal feature area is identified, the feature collection position can be adjusted in time to ensure that the biometric data collected subsequently is more accurate; at the same time, the database update optimization scheme can keep the feature information in the database synchronized with the actual situation, improving the adaptability and stability of the overall recognition system. This multi-link cooperative working mode makes the identity recognition process more intelligent and efficient, better coping with the complex and variable environment of the case handling center, reducing identity misjudgment, optimizing personnel management processes, and making the daily operation of the case handling center more smooth. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A working principle diagram of the case handling center personnel identity rapid identification method described in the present application; Figure 2 A flowchart of multi-modal feature fusion processing; Figure 3 A flowchart of multi-dimensional matching analysis; Figure 4 A flowchart of scene factor dynamic calibration; Figure 5 A flowchart of database update optimization scheme construction. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0018] Please refer to Figures 1-5 The present application provides a case handling center personnel identity rapid identification method, which comprises.

[0019] The biological feature data set of the target person in the case-handling scene includes a face image sequence, a fingerprint line sequence, and voiceprint frequency monitoring data. The biological feature data set is subjected to multi-modal feature fusion processing to generate an identity feature identifier of the target person, which includes a face contour matching degree, a fingerprint line coincidence rate, and a voiceprint frequency consistency. A pre-trained identity recognition model is called to perform multi-dimensional matching analysis processing on the identity feature identifier to generate an identity confirmation result and an abnormal feature region identifier of the target person. The identity confirmation result is subjected to scene factor dynamic calibration processing to generate a calibrated identity confirmation result, and the scene factor dynamic calibration processing is realized based on the correlation between the light intensity sequence and the device acquisition parameters. A set of personnel management strategies is generated according to the abnormal feature region identifier, which includes a feature collection location adjustment scheme and a database update optimization scheme.

[0020] Embodiment 1: In the case-handling center personnel identity recognition scene, when the biological feature data set of the target person is subjected to multi-modal feature fusion processing to generate an identity feature identifier, the following implementation can be performed: The face image sequence is divided into multiple feature subsequences according to a time window, and each feature subsequence corresponds to an acquisition period. For example, if the acquisition period is set to 1 second and the time window is set to 5 seconds, the face image sequence of every 5 seconds will be divided into 5 feature subsequences, and each subsequence corresponds to the face image data collected within 1 second.

[0021] For each feature subsequence, corresponding processing is performed. For example, a multi-modal feature topology structure of the target person is constructed according to the voiceprint frequency monitoring data. The multi-modal feature topology structure includes spatial distribution data of a face angle field, a fingerprint contact field, and a voiceprint energy field. The face angle field reflects the distribution of the target person's face at different angles, such as lowering the head, raising the head, or turning the head during the collection process. The face angle field records the distribution of these different angles in space; the fingerprint contact field reflects the relevant data distribution when the fingerprint contacts the collection device, including the position of the contact, the pressure distribution, etc.; and the voiceprint energy field shows the distribution of the voiceprint energy in space. Different sound positions and intensities will affect the distribution of the voiceprint energy field.

[0022] The multi-modal feature topology is coupled with the feature sub-sequence for analysis and processing to generate a feature fusion result of the current time window, which includes a spatial distribution matrix of the face contour feature quantity, the fingerprint line feature quantity and the voiceprint frequency feature quantity. Specifically, based on a corresponding relationship between the face angle field and the face posture component in the feature sub-sequence, an angle-feature mapping equation is established, and a first distribution function of the face contour feature quantity is obtained by solving the equation. For example, when the angle data in the face angle field and the posture component of the face in the feature sub-sequence have a corresponding relationship, the distribution function of the face contour feature quantity in space can be obtained by solving the equation.

[0023] According to the correlation characteristics of the fingerprint contact field and the contact pressure, a fingerprint contact feature calculation model is constructed, which includes dynamic correction parameters of the skin elasticity coefficient and the sensitivity of the collection device. Because the elasticity of the skin of different people is different, the sensitivity of the collection device may also differ, which will affect the collection of the fingerprint line feature quantity. The dynamic correction parameters can adjust these factors to make the calculated fingerprint line feature quantity more accurate.

[0024] Combined with the spatial variation rate of the voiceprint energy field and the environmental noise coefficient, a voiceprint feature iterative calculation process is established, which includes a feedback correction mechanism of the energy increment and the voiceprint feature increment. Environmental noise will interfere with the collection of the voiceprint, and the spatial variation rate of the voiceprint energy field reflects the change of the voiceprint energy. Through iterative calculation and combined with the feedback correction mechanism, the calculation result of the voiceprint frequency feature quantity can be continuously optimized.

[0025] The output results of the first distribution function, the fingerprint contact feature calculation model and the voiceprint feature iterative calculation process are subjected to spatial interpolation fusion processing to generate multi-modal feature distribution data containing face, fingerprint and voiceprint feature quantities, i.e. the feature fusion result of the current time window.

[0026] The feature fusion results of a plurality of continuous time windows are subjected to weighted superposition processing to calculate the face contour matching degree, the fingerprint line overlap rate and the voiceprint frequency consistency. The face contour matching degree is the maximum similarity value of the face contour feature quantity along the feature extraction direction. For example, in the feature extraction direction, the currently calculated face contour feature quantity is compared with the feature quantity stored in the database to find the maximum similarity value as the face contour matching degree.

[0027] The fingerprint line overlap rate is the overlap proportion of the fingerprint line feature quantity in the normal direction of the contact area. When the fingerprint contacts the collection device, in the normal direction of the contact area, the currently collected fingerprint line feature quantity and the line feature quantity of the corresponding fingerprint in the database will have an overlapping part, and the proportion of this overlapping part in the total is the fingerprint line overlap rate.

[0028] The voiceprint frequency consistency is a ratio of a variance to a mean of the voiceprint frequency feature quantity in a predetermined time interval. In a predetermined time interval, such as 10 seconds, the variance and the mean of the voiceprint frequency feature quantity in the interval are calculated, and the ratio of the two is the voiceprint frequency consistency, which can reflect the stability of the voiceprint frequency in this time period.

[0029] Through such a process, the multi-modal feature fusion processing of the biometric data set is completed, and the identity feature identifier of the target person is generated, providing basic data for subsequent identity recognition. In actual operation, each step requires precise algorithms and device support to ensure that the generated identity feature identifier is accurate and reliable, meeting the needs of the case handling center for rapid identification of personnel identity. Due to the differences in biometric characteristics of different target personnel, the calculation results of various parameters in the processing process will also be different, and the system will conduct targeted analysis and processing according to these different results. At the same time, in the setting of collection cycle, time window and other parameters, adjustments can be made according to the actual case handling scene and needs to achieve the best processing effect. For example, in a scene with high personnel flow, the collection cycle and time window can be appropriately shortened to improve the efficiency of identity recognition; while in a scene with extremely high recognition accuracy requirements, the relevant parameters can be extended to ensure the sufficiency and accuracy of feature extraction.

[0030] In embodiment 2, when the multi-modal feature topology structure is coupled with the feature sub-sequence for analysis and processing to generate the feature fusion result of the current time window, the following implementation can be performed: Based on the correspondence between the face angle field and the face posture component in the feature sub-sequence, an angle-feature mapping equation is established, and the first distribution function of the face contour feature quantity is obtained by solving the angle-feature mapping equation. The face angle field contains spatial angle data such as deflection angle and pitch angle of the face of the target person at different collection times, and the face posture component in the feature sub-sequence corresponds to the dynamic posture change parameters of the face in the collection cycle, such as lateral rotation amplitude and upward inclination degree. By extracting the correspondence of the two on the time axis, a mapping equation with angle parameters as independent variables and face feature components as dependent variables is constructed. For example, when the deflection angle of the face angle field at a certain time is 30 degrees, and the horizontal offset of the face posture component in the feature sub-sequence is 5 pixels, these corresponding data are substituted into the preset binary linear equation model, and the equation coefficients are solved to determine the complete angle-feature mapping equation. Using this equation, the distribution of the face contour feature quantity under different face angles can be calculated to form the first distribution function, which takes spatial coordinates as parameters and outputs the face contour feature quantity value at the corresponding position.

[0031] According to the correlation characteristics of the fingerprint contact field and the contact pressure, a fingerprint contact feature calculation model is constructed, which contains the dynamic correction parameters of the skin elasticity coefficient and the sensitivity of the collection device. The fingerprint contact field records the pressure distribution data of the fingerprint and the contact surface of the collection device, including the pressure value and the pressure change rate of each point in the contact area. The size of the contact pressure will affect the imaging clarity of the fingerprint lines, while the skin elasticity coefficient determines the deformation degree of the fingerprint under different pressures, and the sensitivity of the collection device affects the conversion accuracy of the pressure signal. In the model construction process, the corresponding relationship between the contact pressure and the fingerprint line feature quantity under different skin elasticity coefficients (such as multiple gradient values in the range of 2.5MPa to 4.5MPa) is collected through experiments, and the signal output difference under different device sensitivities (such as 0.1mV / Pa to 0.5mV / Pa) is recorded. These data are used as training samples to determine the correction coefficients of the skin elasticity coefficient and the device sensitivity through multiple regression analysis, forming dynamic correction parameters. In actual calculation, the model will adjust the fingerprint line feature quantity according to the real-time collected contact pressure data combined with the dynamic correction parameters to obtain more realistic fingerprint feature data.

[0032] Combined with the spatial variation rate of the voiceprint energy field and the environmental noise coefficient, a voiceprint feature iterative calculation process is established, which contains the feedback correction mechanism of energy increment and voiceprint feature increment. The spatial variation rate of the voiceprint energy field reflects the energy attenuation or enhancement speed of the voiceprint signal in the propagation process, and the environmental noise coefficient is determined by the frequency distribution, intensity and other parameters of the background noise. In the iterative calculation, first, the voiceprint frequency feature quantity in the first stage is calculated according to the initial voiceprint energy field data, then the correlation degree of the feature quantity and the energy increment (the difference between the current energy and the energy at the previous moment) is calculated, and the feature adjustment quantity is obtained. At the same time, the feature adjustment quantity is corrected according to the environmental noise coefficient, for example, when the environmental noise coefficient is higher than the preset threshold, the weight of the feature adjustment quantity is reduced. The corrected feature adjustment quantity is superimposed on the current voiceprint frequency feature quantity to complete one iteration. Repeat the process until the feature quantity change value of two consecutive iterations is less than the preset threshold (such as 0.02Hz), stop iteration and output the final voiceprint frequency feature quantity.

[0033] The output results of the first distribution function, the fingerprint contact feature calculation model and the voiceprint feature iterative calculation process are subjected to spatial interpolation fusion processing to generate multi-modal feature distribution data containing face, fingerprint and voiceprint feature quantities. The first distribution function of the face contour feature quantity takes two-dimensional plane coordinates as parameters, the output result of the fingerprint line feature quantity is a three-dimensional matrix data of the contact area, and the voiceprint frequency feature quantity is in the form of a time-frequency two-dimensional graph. During spatial interpolation fusion, first, the coordinate systems of the three types of feature data are converted into a three-dimensional coordinate system with the case center spatial coordinate system as the reference, and then the Kriging interpolation method is used to fill data in the sparse areas of the feature quantities, for example, in the spatial overlap area of the face features and the fingerprint features, the fusion feature values of the intermediate points are estimated according to the feature quantity values of the surrounding known points. For the data difference in the time dimension, the time sequence of the voiceprint features is aligned with the collection time points of the face and fingerprint features through linear interpolation, and finally a four-dimensional multi-modal feature distribution matrix containing spatial coordinates (x, y, z) and time coordinate t is formed, and each element in the matrix corresponds to the fusion feature quantity at a certain time and a certain spatial position.

[0034] Through the above steps, the coupling analysis and processing of the multi-modal feature topology structure and the feature subsequence are completed, and the feature fusion result of the current time window is generated, which integrates the feature information of the face, fingerprint and voiceprint in a complete manner and can be directly used for subsequent weighted superposition processing. In actual application, the biological features of different target persons are different, for example, some people have dramatic changes in face contour, some people have uniform fingerprint pressure distribution, and some people have stable voiceprint energy field, and the system will automatically adjust the mapping equation parameters, correction coefficients and iteration times according to these differences to ensure the accuracy of the feature fusion result. At the same time, in view of the possible equipment jitter, personnel movement and other situations in the case center, abnormal data points exceeding the preset threshold will be automatically ignored during the feature fusion process to reduce the influence of interference factors.

[0035] In the implementation of embodiment 3, when the pre-trained identity recognition model is called to perform multi-dimensional matching analysis and processing on the identity feature label to generate the identity confirmation result of the target person and the abnormal feature area label, the following implementation is adopted: The face contour matching degree is input to a first feature analysis layer of the identity recognition model, and the distribution coordinates and similarity change curve of the face feature difference region are determined by a feature difference factor calculation module. The face contour matching degree includes the contour feature similarity values of each region of the face, and the feature difference factor calculation module performs gradient analysis on these similarity values. When the similarity value of a certain region is lower than a preset threshold, the region is marked as a feature difference region, and its two-dimensional distribution coordinates in the face coordinate system (such as the x-axis and y-axis coordinates with the left eye pupil as the origin) are recorded. The similarity change curve takes time as the horizontal axis and similarity as the vertical axis, and reflects the dynamic change trend of the face contour matching degree in the collection period. For example, during the head rotation of the target person, the curve will present the corresponding fluctuation form.

[0036] The fingerprint line coincidence rate is input to a second feature analysis layer of the identity recognition model, and contact feature matching calculation is performed to generate feature matching probability and matching deviation prediction value of the fingerprint contact region. The fingerprint line coincidence rate includes overlapping ratio data of each fingerprint line node, and the contact feature matching calculation first normalizes these data into matching probability values between 0 and 1, and then predicts the matching probability change trend at subsequent time through a Markov chain model. The matching deviation prediction value is calculated based on the difference between the current coincidence rate and the historical average coincidence rate. For example, when the current coincidence rate of a certain fingerprint region is 0.7 and the historical average coincidence rate is 0.85, the matching deviation prediction value is -0.15, and an initial direction vector of the matching deviation path is generated accordingly.

[0037] The voiceprint frequency consistency is input to a third feature analysis layer of the identity recognition model, and the feature offset and feature stability evolution data of the voiceprint frequency are calculated based on a voiceprint feature degradation model. The voiceprint frequency consistency reflects the stability of the voiceprint frequency in the time interval, and the voiceprint feature degradation model predicts the offset trend of the voiceprint frequency over time according to the consistency value. The feature offset is the difference between the predicted offset value and the actual collection value. The feature stability evolution data is presented in the form of time series, recording the change of the voiceprint frequency stability index (such as variance value) in the consecutive collection period. When the stability index decreases for 3 consecutive periods, the corresponding time interval is marked as a potential feature instability region.

[0038] The similarity change curve, feature matching probability and feature offset are fused to generate a comprehensive matching index of the target person, and the identity confirmation result is determined according to the comparison result of the comprehensive matching index and the preset recognition threshold. The calculation formula of the comprehensive matching index is:

[0039] wherein, is the comprehensive matching index, the average amplitude of the face similarity change curve (value range 0 to 1), the weighted average of the fingerprint feature matching probability (value range 0 to 1), the voiceprint frequency feature offset (unit: Hz), and a, b, and g are weight coefficients (and a+b+g=1, such as a=0.4, b=0.3, and g=0.3). When S is greater than a preset recognition threshold (such as 0.75), the identity confirmation result is a matching success; and when S is less than or equal to the preset recognition threshold, the identity confirmation result is a matching failure.

[0040] Based on the spatial superposition result of the distribution coordinates, the matching deviation prediction value, and the feature stability evolution data, the geometric positions of the feature difference area, the matching deviation path, and the feature unstable area are identified. The geometric position of the feature difference area is presented in the form of polygon boundary coordinates, for example, the face feature difference area can be marked as a rectangular area with (5, 3), (8, 3), (8, 6), and (5, 6) as the vertices. The matching deviation path is connected by consecutive deviation prediction value coordinate points, for example, the deviation path of the fingerprint area can present a curve trajectory from (20, 15) to (22, 18) to (25, 20). The feature unstable area is marked with a rectangular frame in a time-frequency two-dimensional coordinate system, where the horizontal axis is time (unit: seconds) and the vertical axis is frequency (unit: Hz), and the frame selection range covers the entire time interval and the corresponding frequency range in which the feature stability index decreases.

[0041] For the identification of the feature difference area, the optimal collection position is calculated, and the optimal collection position is achieved by adjusting the coverage range proportion of adjacent collection devices. The distribution coordinates of the feature difference area reflect the insufficient feature capture of the current collection device in this area, and at this time, the overlap degree of the coverage ranges of the adjacent two collection devices needs to be analyzed. For example, when the face feature difference area is located at the edge of the coverage range of device A (coverage proportion is 0.3) and the middle of the coverage range of device B (coverage proportion is 0.7), by deflecting the lens angle of device A by 5 degrees towards the difference area, the coverage proportion of device A in this area is increased to 0.5, and the coverage proportion of device B remains unchanged at 0.7, at this time, the coverage range proportion of the two devices is 5:7, and the corresponding area center point is the optimal collection position.

[0042] According to the identification of the matching deviation path, a database update optimization scheme is constructed, which includes the selection of the feature sample supplement area and the optimization configuration of the sample feature parameters. The geometric features of the matching deviation path are extracted, and the path bending radius and the deviation direction angle are calculated. The path bending radius is obtained by fitting the curve equation of the deviation path, for example, when the deviation path is approximately a circular arc with a radius of 10 mm, its bending radius is 10 mm; the deviation direction angle is the angle between the tangent direction of the path and the x-axis of the fingerprint coordinate system, for example, if the tangent direction of a certain path segment points to the northeast direction of the coordinate system, its deviation direction angle is 45 degrees. According to the bending radius, the coverage density of the feature sample supplement is selected, and the coverage density is inversely proportional to the bending radius, that is, the smaller the bending radius (the more tortuous the path), the higher the coverage density, for example, when the bending radius is 5 mm, the coverage density is set to 3 sample points per square millimeter, and when the bending radius is 20 mm, the coverage density is set to 1 sample point per square millimeter. Based on the deviation direction angle, the collection direction of the sample feature parameters is adjusted, so that the collection direction and the deviation direction form a predetermined angle (such as 30 degrees), for example, when the deviation direction angle is 60 degrees, the collection direction is adjusted to 90 degrees, so as to more comprehensively capture the feature details of the deviation area. According to the feature collection accuracy test data, the sample supplement frequency is dynamically adjusted, for example, when the collection accuracy of a certain area is 0.8, the sample supplement frequency is set to once a week, and when the accuracy decreases to 0.7 or below, the supplement frequency is adjusted to 3 times a week, while ensuring that the supplemented samples are below the feature library capacity threshold. An optimization parameter configuration table containing coverage density, collection direction and supplement frequency is generated as the basis for database update.

[0043] Based on the identification of the feature unstable area, a collection frequency adjustment strategy is generated, which dynamically adjusts the collection interval and collection time according to the feature stability prediction value. The time-frequency coordinate range of the feature unstable area reflects the unstable state of the voiceprint feature in a specific period, and the feature stability prediction value is obtained by linear regression analysis of the feature stability evolution data, for example, when the prediction value is 0.3 (the value range is 0 to 1, and the lower the value, the more unstable), the collection interval is shortened from 5 seconds to 3 seconds, and the collection time is extended from 1 second to 2 seconds, so as to obtain more feature data. When the feature stability prediction value rises to 0.6 or above, the collection interval and time are restored to the initial settings.

[0044] The optimal collection position, database update optimization scheme and collection frequency adjustment strategy are prioritized to generate a management strategy set containing execution timing and implementation parameters. The prioritization is determined based on the severity of feature differences, bias and instability, for example, when the area ratio of the feature difference region exceeds 20% of the total face area, the corresponding optimal collection position adjustment is listed as the highest priority and needs to be executed within 1 hour; if the database update optimization scheme involves fingerprint sample supplementation and the matching bias path length exceeds 15 mm, it is listed as the second highest priority and is scheduled to be implemented within 24 hours; the collection frequency adjustment strategy is sorted according to the duration of the feature instability region, and the strategy with a duration longer than 10 minutes is executed in priority to the strategy with a shorter duration. Implementation parameters include specific angles for device adjustment, specific coordinate ranges for sample supplementation, specific numerical values for collection frequency, etc., and execution timing specifies the start time and completion deadline of each strategy, forming a complete set of personnel management strategies.

[0045] In the implementation of the scene factor dynamic calibration processing of the identity confirmation result to generate the calibrated identity confirmation result, the following methods can be implemented: Extreme illumination values and illumination change frequency in the illumination intensity sequence are extracted, and the dynamic adjustment amount of device collection parameters with illumination change is calculated. The illumination intensity sequence contains continuous collection of illumination data in the case scene, such as in the collection period from 9:00 to 10:00 in the morning, the illumination intensity may gradually rise from 500 lux to 1200 lux, among which the extreme illumination value is 1200 lux (higher than the preset threshold of 800 lux). The illumination change frequency is determined by counting the number of illumination intensity fluctuations within a unit time, for example, within 1 minute, the illumination intensity appears 6 times of fluctuation exceeding ±100 lux, then the illumination change frequency is 6 times / minute. The device collection parameters include exposure time, ISO sensitivity, etc., and the dynamic adjustment amount is calculated according to the corresponding relationship between illumination intensity and parameters, such as when the illumination intensity rises from 500 lux to 1200 lux, the dynamic adjustment amount of exposure time is -30 ms (shortened from 100 ms to 70 ms), and the dynamic adjustment amount of ISO sensitivity is -100 (reduced from 400 to 300).

[0046] The face contour matching degree is compensated according to the dynamic adjustment amount, and a calibrated face contour matching degree is generated. The initial acquisition parameters and the dynamic adjustment amount of the target personnel under the reference illumination are obtained, and an acquisition parameter-illumination correlation function is established. The reference illumination is set to 500 lux, and the corresponding initial acquisition parameters are exposure time 100 ms and ISO 400. The acquisition parameter-illumination correlation function takes the illumination intensity as the independent variable, and outputs the corresponding exposure time and ISO adjustment value, for example, the function form is exposure time = 100-0.05×(illumination intensity-500). When the illumination intensity is 1200 lux, the calculated exposure time is 70 ms, which is consistent with the dynamic adjustment amount. The face feature compensation amount is calculated according to the function, and the face feature compensation amount is the product of the illumination change amount and the acquisition parameter change amount, such as the illumination change amount is 700 lux (1200-500), and the acquisition parameter change amount is-30 ms (exposure time adjustment amount), then the face feature compensation amount is 700×(-30)=-21000 (lux·ms). The compensation amount is superimposed in the calculation process of the face contour matching degree, and a face contour matching degree calibration value containing the influence of illumination is generated, for example, the original matching degree is 0.82, and the calibration value after compensation is 0.79. The calibration value is subjected to noise suppression effect compensation processing, the noise coefficient of the device noise curve under 70 ms exposure time is 0.02, the illumination retention time is 15 minutes, and the product factor is 0.02×15=0.3. According to this, the calibration value 0.79 is adjusted to 0.79×(1+0.3)=1.027, and then constrained in the range of 0 to 1 through normalization processing, and finally the calibrated face contour matching degree of 0.98 is obtained.

[0047] Based on the correlation between the illumination change frequency and the feature acquisition stability, the fingerprint line coincidence rate is subjected to acquisition deviation correction processing, and a calibrated fingerprint line coincidence rate is generated. The illumination change frequency will affect the stability of the optical sensor of the fingerprint acquisition device, and when the frequency is 6 times per minute, the acquisition stability coefficient is 0.92 (value range 0 to 1). The original value of the fingerprint line coincidence rate is 0.78, and the acquisition deviation correction processing is realized by multiplying the original value by the stability coefficient, i.e. 0.78×0.92=0.7176, to obtain the preliminary correction value. At the same time, according to the periodic characteristics of the illumination change, the preliminary correction values of the continuous three acquisition periods are subjected to sliding average processing, such as the correction values of the subsequent two periods are 0.73 and 0.75 respectively, and the sliding average value is (0.7176+0.73+0.75) / 3≈0.7325, which is used as the calibrated fingerprint line coincidence rate.

[0048] According to the device noise variation data under extreme light, the voiceprint frequency consistency is processed by feature anti-noise adaptive adjustment to generate calibrated voiceprint frequency consistency. The device noise variation data corresponding to the extreme light value 1200 lux shows that the noise power spectrum of the voiceprint collection device is increased by 15% in the frequency band of 1 kHz to 3 kHz. The original calculation value of the voiceprint frequency consistency is 0.85 (the ratio of variance to mean is 0.15, 1-0.15=0.85). The feature anti-noise adaptive adjustment is realized by weighted filtering of voiceprint frequency feature quantity in the frequency band, and the weight coefficient is set according to the noise increase ratio, such as reducing the weight of 1 kHz to 3 kHz frequency band by 15%. The adjusted voiceprint frequency consistency calculation value is 0.82. Combined with the ratio of noise duration (such as 20 minutes of extreme light) to device anti-noise threshold (30 minutes) (20 / 30≈0.67), the 0.82 is adjusted again to obtain 0.82×0.67≈0.55 as the calibrated voiceprint frequency consistency.

[0049] The calibrated face contour matching degree 0.98, the fingerprint line coincidence rate 0.7325 and the voiceprint frequency consistency 0.55 are input into the identity recognition model to recalculate and generate the identity confirmation result after the calibration of the scene factors. The identity recognition model calculates the comprehensive matching index according to the weight distribution of the three (such as face 0.4, fingerprint 0.3, voiceprint 0.3), that is, 0.98×0.4+0.7325×0.3+0.55×0.3≈0.392+0.2198+0.165≈0.7768. The preset recognition threshold is 0.75. Since 0.7768>0.75, the final identity confirmation result is matching success.

[0050] In another example, if the extreme light value in the light intensity sequence is 300 lux (lower than the reference light), the light change frequency is 2 times / minute, and the dynamic adjustment amount of the device acquisition parameters is the exposure time +20 ms (from 100 ms to 120 ms) and ISO +50 (from 400 to 450). The face feature compensation amount is (300-500) x 20 = (-200) x 20 = -4000 (lux ms), and the original face contour matching degree 0.86 is compensated to a calibrated value of 0.88. After compensation by the noise suppression effect (noise coefficient 0.03, light retention time 10 minutes, product factor 0.3), the value is 0.88 x 1.3 ≈ 1.144, and after normalization, the value is 0.96. The original value of the fingerprint line coincidence rate is 0.81, combined with the stability coefficient 0.97 corresponding to the light change frequency of 2 times / minute, the preliminary correction value is 0.81 x 0.97 ≈ 0.7857, and after sliding average, the value is 0.79. The device noise under extreme light is increased by 5% in the frequency band of 1 kHz to 3 kHz, and the original value of the voiceprint frequency consistency is 0.88, which is adjusted to 0.86. Combined with the ratio of noise duration 8 minutes to threshold 30 minutes 0.27, the final calibrated value is 0.86 x 0.27 ≈ 0.232. The comprehensive matching index is 0.96 x 0.4 + 0.79 x 0.3 + 0.232 x 0.3 ≈ 0.384 + 0.237 + 0.0696 ≈ 0.6906, which is lower than the preset recognition threshold 0.75, and the identity confirmation result is a matching failure.

[0051] Through the above process, the scene factor dynamic calibration process is completed, and the calibrated identity confirmation result is generated. The result considers the influence of light intensity change on multi-modal features, making the identity recognition result more suitable for the environmental conditions in the actual case handling scene. The calibration parameters and calculation results under different light conditions are different, and the system will automatically select the corresponding calibration logic according to the real-time acquisition of light data to ensure the adaptability and accuracy of the processing process.

[0052] In the preset calibration period, the system continuously records the actual result of each identity recognition (such as matching success or failure), and at the same time, stores the deviation data of the face contour matching degree, the fingerprint line coincidence rate, the voiceprint frequency consistency and the database reference value in each identification process, i.e. the feature matching deviation. For example, in a certain identification, the actual recognition result is matching success, and the deviation of the actual value of the face contour matching degree from the reference value is -0.05, the deviation of the fingerprint line coincidence rate is 0.03, and the deviation of the voiceprint frequency consistency is -0.02. These data are included in the acquisition range.

[0053] The actual recognition result and the predicted confirmation result are subjected to deviation analysis processing to generate a first error correction coefficient. The predicted confirmation result is a matching probability value (range 0 to 1) output by the identity recognition model in the recognition process, and the actual recognition result is represented by a binary variable (1 represents success and 0 represents failure). The deviation analysis processing is realized by calculating the mean square error of the two, for example, in a 24-hour calibration period, a total of 100 recognitions are performed, of which 80 actual results are successful, corresponding to an average predicted probability of 0.85; 20 actual results are failures, corresponding to an average predicted probability of 0.6. After the mean square error is calculated by the formula, the first error correction coefficient is determined according to the ratio of the mean square error to the maximum allowed error. If the mean square error is 0.12 and the maximum allowed error is 0.2, then the first error correction coefficient is 0.12 / 0.2 = 0.6.

[0054] The feature matching deviation and the predicted deviation value are subjected to time domain comparison processing to generate a second error correction coefficient. The predicted deviation value is the feature matching deviation predicted by the identity recognition model based on historical data. The time domain comparison processing needs to calculate the difference value of the feature matching deviation and the predicted deviation value at the same time, and integrate the difference values of multiple consecutive time points. For example, the sequence of feature matching deviation of face contour matching degree within 1 hour is [-0.05, -0.03, 0.01, 0.02], and the corresponding predicted deviation value sequence is [-0.04, -0.05, 0, 0.03]. The difference value sequence of the two is [-0.01, 0.02, 0.01, -0.01], and the integral (accumulation) of the sequence is 0.01. Similarly, the deviation difference integrals of fingerprint and voiceprint are 0.03 and -0.02 respectively, and the absolute value average of the three is taken as the comprehensive deviation integral, i.e. (0.01 + 0.03 + 0.02) / 3 = 0.02. The second error correction coefficient is the ratio of the comprehensive deviation integral to the deviation threshold value. If the deviation threshold value is 0.05, then the second error correction coefficient is 0.02 / 0.05 = 0.4.

[0055] The weight parameters of the identity recognition model are adjusted according to the first error correction coefficient and the second error correction coefficient to generate an optimized identity recognition model. The weight parameters of the identity recognition model include the proportions of facial, fingerprint, and voiceprint features in the calculation of the comprehensive matching index (a, b, g), and the initial values are 0.4, 0.3, and 0.3, respectively. The adjustment process uses a weighted correction method, for example, the first error correction coefficient is 0.6, and the second error correction coefficient is 0.4, and the weights of the two are set to 0.6 and 0.4, respectively. For the facial feature weight a, the correction amount is (the first error correction coefficient x the facial prediction deviation influence degree + the second error correction coefficient x the facial feature deviation integral proportion) x the initial weight, where the facial prediction deviation influence degree is the proportion of facial features in the prediction error sample (such as 0.3), and the facial feature deviation integral proportion is the ratio of the facial deviation integral to the total integral (such as 0.01 / 0.06 ≈ 0.17). The calculated correction amount is (0.6 x 0.3 + 0.4 x 0.17) x 0.4 ≈ 0.036, and the corrected a value is 0.4 + 0.036 = 0.436. Similarly, b and g are adjusted, and the optimized weight parameters are obtained.

[0056] The optimized identity recognition model is applied to subsequent batches of personnel identity recognition tasks. After the weight parameter adjustment is completed, the system automatically replaces the original model with the optimized model for new recognition processes. For example, in subsequent recognition of the same target personnel, the weight of the facial contour matching degree is increased to 0.436, and the weights of the fingerprint and voiceprint are adjusted accordingly. When calculating the comprehensive matching index, the model will focus more on the influence of facial features, thereby adapting to the feature weight imbalance problem found in the early deviation analysis. In the new recognition task, the system continuously collects actual results and deviation data to provide input for model optimization in the next calibration cycle, forming a cyclic iterative optimization mechanism.

[0057] In another scenario, the preset calibration period is 12 hours, during which 50 recognitions are performed, and the deviation between the actual recognition result and the predicted confirmation result is large, and the first error correction coefficient is calculated as 0.8. The deviation of the fingerprint line coincidence rate in the feature matching deviation is more significant, resulting in a second error correction coefficient of 0.5. When adjusting the model weight, the weight b of the fingerprint feature will be greatly corrected from the initial 0.3 to 0.35 to reduce the influence of the fingerprint feature deviation on the recognition result. In subsequent recognition, the optimized model will change the recognition accuracy for target personnel with slightly worn fingerprint lines, and the system will still enter the next calibration cycle according to the established process to continuously optimize the model parameters.

[0058] It is to be understood that the terminology used herein such as first and second, and the like, is only used to distinguish one entity or action from another entity or action, and does not necessarily require or imply any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0059] While embodiments of the present application have been shown and described with reference to particular embodiments thereof, it will be understood by those skilled in the art that various changes in form and details can be made therein without departing from the spirit and scope of the application. The scope of the application is defined by the appended claims and their equivalents.

Claims

1. A method for rapid identification of personnel in a case-handling center, characterized in that, The method includes: Collect a set of biometric data of the target personnel in the case handling scenario. The set of biometric data includes facial image sequences, fingerprint ridge sequences and voiceprint frequency monitoring data. The biometric data set is subjected to multimodal feature fusion processing to generate the identity feature identifier of the target person. The identity feature identifier includes facial contour matching degree, fingerprint line overlap rate and voiceprint frequency consistency. A pre-trained identity recognition model is invoked to perform multi-dimensional matching analysis on the identity feature identifiers, generating the identity confirmation result and abnormal feature area identifiers of the target person; The identity verification result is subjected to scene factor dynamic calibration processing to generate a calibrated identity verification result. The scene factor dynamic calibration processing is based on the correlation between the light intensity sequence and the device acquisition parameters. A set of personnel management strategies is generated based on the abnormal feature region identifiers. The set of personnel management strategies includes feature collection location adjustment schemes and database update optimization schemes.

2. The method for rapid identification of personnel in a case-handling center according to claim 1, characterized in that, The step of performing multimodal feature fusion processing on the biometric data set to generate the identity feature identifier of the target person includes: The facial image sequence is divided into multiple feature sub-sequences according to a time window, and each feature sub-sequence corresponds to a collection period; For each of the aforementioned feature subsequences, the following processing is performed: Based on the voiceprint frequency monitoring data, a multimodal feature topology structure of the target person is constructed. The multimodal feature topology structure includes spatial distribution data of facial angle field, fingerprint contact field and voiceprint energy field. The multimodal feature topology and the feature subsequence are coupled and analyzed to generate the feature fusion result of the current time window. The feature fusion result includes the spatial distribution matrix of facial contour features, fingerprint ridge features and voiceprint frequency features. The feature fusion results from multiple consecutive time windows are weighted and superimposed to calculate the facial contour matching degree, fingerprint ridge overlap rate, and voiceprint frequency consistency; wherein, The facial contour matching degree is the maximum similarity value of facial contour features along the feature extraction direction. The fingerprint ridge overlap rate is the percentage of overlap of fingerprint ridge features in the normal direction of the contact area. The voiceprint frequency consistency is the ratio of the variance to the mean of the voiceprint frequency characteristic quantity within a predetermined time interval.

3. The method for rapid identification of personnel in a case-handling center according to claim 2, characterized in that, The step of coupling the multimodal feature topology with the feature subsequence for analysis and processing to generate the feature fusion result for the current time window includes: Based on the correspondence between the facial angle field and the facial pose components in the feature subsequence, an angle-feature mapping equation is established, and the first distribution function of the facial contour feature quantity is obtained by solving the angle-feature mapping equation. Based on the correlation characteristics between the fingerprint contact field and the contact pressure, a fingerprint contact feature calculation model is constructed. The fingerprint contact feature calculation model includes dynamic correction parameters for the skin elastic coefficient and the sensitivity of the acquisition device. By combining the spatial rate of change of the voiceprint energy field with the environmental noise coefficient, an iterative calculation process for voiceprint features is established. The iterative calculation process includes a feedback correction mechanism for energy increment and voiceprint feature increment. The output results of the first distribution function, the fingerprint contact feature calculation model, and the voiceprint feature iterative calculation process are spatially interpolated and fused to generate multimodal feature distribution data containing facial, fingerprint, and voiceprint features.

4. The method for rapid identification of personnel in a case-handling center according to claim 1, characterized in that, The process of calling a pre-trained identity recognition model to perform multi-dimensional matching analysis on the identity feature identifiers, generating the identity confirmation result and abnormal feature region identifiers of the target person, including: The facial contour matching degree is input into the first feature analysis layer of the identity recognition model, and the distribution coordinates and similarity change curve of the facial feature difference region are determined by the feature difference factor calculation module. The fingerprint ridge overlap rate is input into the second feature analysis layer of the identity recognition model to perform contact feature matching calculation and generate the feature matching probability and matching deviation prediction value of the fingerprint contact area. The voiceprint frequency consistency is input into the third feature analysis layer of the identity recognition model, and the feature offset and feature stability evolution data of the voiceprint frequency are calculated based on the voiceprint feature degradation model. By integrating the similarity change curve, the feature matching probability, and the feature offset, a comprehensive matching index for the target person is generated, and the identity confirmation result is determined based on the comparison result between the comprehensive matching index and the preset recognition threshold. Based on the spatial overlay of the distribution coordinates, the predicted matching deviation values, and the feature stability evolution data, the geometric locations of the feature difference regions, matching deviation paths, and feature instability regions are identified.

5. The method for rapid identification of personnel in a case-handling center according to claim 1, characterized in that, The step of performing dynamic scene factor calibration processing on the identity verification result to generate a calibrated identity verification result includes: Extract extreme light intensity values ​​and light intensity change frequency from the light intensity sequence, and calculate the dynamic adjustment amount of the device acquisition parameters as the light intensity changes; Based on the dynamic adjustment amount, the illumination feature compensation calculation is performed on the facial contour matching degree to generate a calibrated facial contour matching degree. Based on the correlation between the frequency of illumination changes and the stability of feature acquisition, the fingerprint ridge overlap rate is subjected to acquisition deviation correction processing to generate a calibrated fingerprint ridge overlap rate. Based on the equipment noise change data under extreme lighting conditions, the voiceprint frequency consistency is subjected to feature anti-noise adaptation adjustment processing to generate calibrated voiceprint frequency consistency. The calibrated facial contour matching degree, fingerprint ridge overlap rate, and voiceprint frequency consistency are input into the identity recognition model for recalculation, generating the identity confirmation result after calibrating the scene factors.

6. The method for rapid identification of personnel in a case-handling center according to claim 5, characterized in that, The step of performing illumination feature compensation calculation on the facial contour matching degree based on the dynamic adjustment amount to generate a calibrated facial contour matching degree includes: Obtain the initial acquisition parameters and dynamic adjustment amount of the target person under the reference illumination, and establish the acquisition parameter-illumination correlation function; The facial feature compensation amount is calculated based on the acquisition parameter-illumination correlation function, whereby the facial feature compensation amount is the product of the illumination change and the acquisition parameter change. The facial feature compensation amount is superimposed on the facial contour matching degree calculation process to generate a facial contour matching degree calibration value that includes the influence of lighting. The facial contour matching calibration value is subjected to noise suppression compensation processing, which is based on the product factor of the device noise curve and the illumination duration.

7. The method for rapid identification of personnel in a case-handling center according to claim 4, characterized in that, The step of generating a set of personnel management strategies based on the abnormal feature region identifiers includes: For the identification of the feature difference region, the optimal acquisition position is calculated, and the optimal acquisition position is achieved by adjusting the coverage ratio of adjacent acquisition devices; Based on the identifier of the matching deviation path, a database update optimization scheme is constructed, which includes the selection of feature sample supplementation regions and the optimized configuration of sample feature parameters. Based on the identification of the unstable region of the feature, a sampling frequency adjustment strategy is generated. The sampling frequency adjustment strategy dynamically adjusts the sampling interval and sampling duration according to the feature stability prediction value. The optimal acquisition location, the database update optimization scheme, and the acquisition frequency adjustment strategy are prioritized and sorted to generate a set of management strategies that include execution timing and implementation parameters.

8. The method for rapid identification of personnel in a case-handling center according to claim 7, characterized in that, The database update optimization scheme includes: Extract the geometric features of the matching deviation path, and calculate the path curvature radius and deviation direction angle; The coverage density of the feature samples is selected based on the bending radius, and the coverage density is inversely proportional to the bending radius. The acquisition direction of the sample feature parameters is adjusted based on the deviation direction angle so that the acquisition direction and the deviation direction form a predetermined angle. The frequency of sample replenishment is dynamically adjusted based on the feature acquisition accuracy test data to ensure that the replenished samples are below the critical value of the feature library capacity. Generate an optimized parameter configuration table that includes coverage density, acquisition direction, and supplementation frequency.

9. The method for rapid identification of personnel in a case-handling center according to claim 1, characterized in that, The method further includes: The actual identification results and feature matching deviations of the target personnel are collected within a preset calibration period; The actual identification result and the predicted confirmation result are subjected to deviation analysis to generate a first error correction coefficient. The feature matching deviation and the prediction deviation are compared in the time domain to generate a second error correction coefficient. The weight parameters of the identity recognition model are adjusted according to the first error correction coefficient and the second error correction coefficient to generate an optimized identity recognition model; The optimized identity recognition model was then applied to subsequent batches of personnel identity recognition tasks.

10. A rapid identification system for personnel in a case-handling center, characterized in that, The device includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the memory to implement the rapid identification method for personnel in a case-handling center as described in any one of claims 1-9.

Images