AI face recognition cloud service system supporting elastic attendance checking rules

Abstract

The invention belongs to the technical field of biological information recognition, and discloses an AI face recognition cloud service system supporting an elastic attendance checking rule. Comprising the following steps: collecting working behavior data of each employee, performing time period efficiency analysis, generating a working efficiency curve, and performing intelligent optimization on a configured elastic attendance checking rule; future weather information is obtained, whether the weather belongs to extreme weather is judged, and if yes, a temporary attendance checking rule is intelligently generated; after each time of face recognition, the appearance difference degree is calculated in real time, the change degree is distinguished, if the change degree is slight change, the feature template is adjusted, and if the change degree is significant change, a new feature template is generated; obtaining a pre-attendance set, pre-caching a feature template of each employee in the pre-attendance set, and simultaneously performing face recognition on multiple employees; the attendance checking efficiency and the working efficiency of the employees can be improved, and the identification challenges caused by appearance changes of the employees can be effectively handled, so that the management cost is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biological information recognition, more specifically, the present application relates to an AI face recognition cloud service system supporting flexible attendance rules. BACKGROUND

[0002] With the development of artificial intelligence and face recognition technology, more and more enterprises begin to apply face recognition attendance systems to improve attendance efficiency and accuracy; the existing face recognition attendance system mainly adopts face detection and recognition technology based on deep learning, collects the face image of the employee through the camera, extracts the face feature vector and compares it with the pre-stored feature template, so as to realize identity verification; in practical application, compared with traditional punch card and manual check-in mode, this kind of face recognition attendance system significantly improves the attendance efficiency, reduces the human error rate, and at the same time reduces the work burden of the management personnel.

[0003] However, with the deepening of enterprise digital transformation and the diversified development of work mode, the traditional attendance management system has been difficult to meet the management needs of modern enterprises, and many shortcomings have been exposed: on the one hand, the attendance rules lack flexibility and cannot adapt to the increasingly diversified work mode of modern enterprises, especially for creative positions, R&D positions and other positions that require flexible working hours, fixed attendance time limits the work efficiency of employees; on the other hand, the environmental adaptability of face recognition is poor, and normal changes in the appearance of employees (such as changing hair style, wearing glasses, and making up, etc.) often lead to recognition failure, so that the features need to be re-entered, increasing the management cost.

[0004] In view of this, the present application proposes an AI face recognition cloud service system supporting flexible attendance rules to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: an AI face recognition cloud service system supporting flexible attendance rules, comprising: a rule configuration module for configuring flexible attendance rules; an efficiency driving module for collecting the work behavior data of each employee, performing time period efficiency analysis on the work behavior data, generating the work efficiency curve of each employee, and intelligently optimizing the flexible attendance rules based on the work efficiency curve to form the individualized attendance rules of each employee; a weather driving module for obtaining future weather information and determining whether it belongs to extreme weather, if it belongs to extreme weather, temporarily covering the corresponding individualized attendance rules based on the future weather information to intelligently generate temporary attendance rules for each employee; The feature management module is used to calculate the appearance difference in real time after each face recognition using a feature template, and to distinguish the degree of change based on the appearance difference. If the degree of change is slight, the feature template is adjusted; if the degree of change is significant, a new feature template is generated. The concurrent recognition module is used to obtain the pre-attendance set for the next time period and pre-cache the feature templates of each employee in the pre-attendance set to form a template set. Based on the template set, face recognition is performed on multiple employees simultaneously.

[0006] Furthermore, methods for performing time-based efficiency analysis on work behavior data include: Work behavior data includes employees' work activity data at different times throughout the day; each set of work behavior data corresponds to one employee, and the time periods are divided according to a preset granularity. Based on the work behavior data of each employee, obtain the work activity data of each employee in each time period; input the work activity data corresponding to each time period into the trained index analysis model to predict the set of efficiency indices corresponding to each employee in each time period. A preset efficiency weight set is used, which includes the weight coefficient of each index in the efficiency index set under different positions. Based on the efficiency weight set, the efficiency index set of the same employee in the same time period is weighted and summed to obtain the comprehensive work efficiency of each employee in each time period.

[0007] Furthermore, methods for generating the work efficiency curve for each employee include: Collect work behavior data of each employee during the previous week, and perform time-period efficiency analysis on the daily work behavior data to obtain the comprehensive work efficiency of each employee during each time period during the previous week; for each employee's comprehensive work efficiency, calculate the average comprehensive work efficiency of the same employee in the same time period to obtain the average work efficiency of each employee in each time period. An efficiency coordinate system is constructed for each employee; the midpoint of the time period corresponding to each average work efficiency is taken as the time point of the corresponding average work efficiency, and each average work efficiency and its corresponding time point are taken as a set of efficiency sampling points; in the efficiency coordinate system corresponding to each employee, the corresponding efficiency sampling points are marked in sequence; a smooth fitting algorithm is used to fit the efficiency sampling points in each efficiency coordinate system in sequence to generate the work efficiency curve of each employee.

[0008] Furthermore, methods for intelligently optimizing flexible attendance rules include: Flexible attendance rules include clock-in time windows, core working periods, and total working hours requirements; Based on the check-in time window, the covered time duration interval is obtained and marked as the check-in time range. Multiple optimized time windows are constructed based on the core work period, total working hours requirements, and the check-in time range. For each optimized time window, the overall work efficiency and low-efficiency time duration for each employee are calculated sequentially based on their work efficiency curve. Using a preset weighting coefficient, the overall work efficiency and low-efficiency time duration for each employee are weighted to obtain their comprehensive efficiency score for each optimized time window. The comprehensive efficiency scores of the same employee in different optimized time windows are compared, and the optimized time window with the highest comprehensive efficiency score is taken as the optimal time window for that employee. Based on the optimal time window for each employee, the check-in time windows within the flexible attendance rules are intelligently optimized to form personalized attendance rules for each employee.

[0009] Furthermore, methods for intelligently generating temporary attendance rules for each employee include: Based on future weather information, calculate the weather risk value and determine whether to implement work-from-home arrangements based on the weather risk value. If work-from-home arrangements are not implemented, obtain the delay coefficient for different commuting modes based on future weather information. Obtain each employee's commuting mode and home address, and calculate each employee's normal commuting time. Calculate the delayed commuting time based on each employee's normal commuting time and the delay coefficient for the corresponding commuting mode. Apply an uncertainty buffer to each delayed commuting time to obtain each employee's final commuting time. Calculate the difference between each employee's final commuting time and the corresponding normal commuting time to obtain each employee's delay time. Obtain the check-in time window from each employee's personalized attendance rules and mark it as the actual check-in window; based on the delay time, shift the actual check-in window for each employee to obtain the delayed check-in window; based on the delayed check-in window, recalculate each employee's comprehensive efficiency score and mark it as the delayed efficiency score; based on each employee's comprehensive efficiency score and delayed efficiency score for the optimal time window, determine again whether to implement work-from-home; if work-from-home is still not implemented, replace the actual work window in the corresponding employee's personalized attendance rules with the corresponding delayed work window to obtain the corresponding temporary attendance rules for the employee.

[0010] Furthermore, the method for obtaining each employee's final commute time is as follows: Future weather information refers to meteorological parameters for the area where the company is located for the next day, including the type and level of the warning. Collect the commuting times of employees with different commuting methods under different warning types and warning levels, and mark them as historical delay times; from all historical delay times for each employee, obtain the historical delay times corresponding to meteorological parameters in future weather information, and mark them as current delay times; combine each employee's delayed commuting time and corresponding current delay time as a delay set; calculate the 80th percentile for each delay set and mark it as a safety buffer time; calculate the sum of each employee's safety buffer time and the preset fixed buffer time to obtain each employee's final commuting time.

[0011] Furthermore, methods for real-time calculation of appearance differences include: The process involves acquiring face images used in face recognition, calculating the texture complexity and self-similarity matrix of the face images, and extracting face feature vectors from the face images. The texture complexity, self-similarity matrix, and face feature vectors are then integrated to obtain a comprehensive feature vector. The values ​​in the comprehensive feature vector are then normalized sequentially to obtain a standard feature vector. Finally, the standard feature vectors are compared with the employee's feature templates, and Euclidean distance is used to calculate the degree of appearance difference.

[0012] Furthermore, methods for adjusting the feature template include: The feature template compared with the standard feature vector is marked as the current template, and the feature template located one position before the current template is marked as the predecessor template; the method used to adjust the predecessor template is obtained. A feature template is selected and marked as an adjustment template. The value is an integer greater than 2; obtain the acquisition time corresponding to each adjustment template, and delete the adjustment template with the earliest acquisition time; where, if the number of adjustment templates is less than 2... If the earliest acquisition time is not deleted, the acquisition time corresponding to the standard feature vector is marked as the current time, and the time difference of each acquisition template is calculated in turn according to the acquisition time of each acquisition template. A preset time decay function is used to substitute each time difference value into the time decay function to obtain the time weight of each adjustment template. Based on the time weight, a weighted average is calculated for all adjustment templates and the standard feature vector to obtain a weighted template. Based on the weighted template, the current template is adjusted.

[0013] Furthermore, methods for obtaining the pre-attendance set include: Obtain the real-time time, calculate the sum of the real-time time and the granularity of the segmentation to obtain the cache time; construct the cache period based on the real-time time and the cache time; mark the clock-in time window in the personalized attendance rules or temporary attendance rules corresponding to each employee as an analysis window, compare each analysis window with the cache period to determine whether there is an intersection between each analysis window and the cache period; if there is an intersection, mark the employee corresponding to the analysis window as a cache employee; combine all cache employees to obtain the pre-attendance set.

[0014] Furthermore, methods for simultaneously performing facial recognition on multiple employees include: When multiple employees pass through the face recognition area simultaneously, multiple images of the face recognition area are continuously acquired. Employees passing through the face recognition area are marked as employees to be identified. Each area image is then segmented sequentially to obtain multiple sets of employees to be identified. Each set of employees to be identified includes... Zhang is an image to be identified. The number of images in the region; each set of images to be identified corresponds to one employee to be identified. The standard feature vectors corresponding to each image to be identified are extracted sequentially and marked as recognition feature vectors. Each recognition feature vector is compared with each feature template in the template set to calculate the appearance difference degree corresponding to each image to be identified and marked as the recognition difference degree. The average recognition difference degree of the corresponding feature template and the employee to be identified is calculated to obtain the matching difference degree between each employee to be identified and each feature template in the template set. The feature template with the smallest matching difference degree is used as the recognition template of the corresponding employee to be identified, and the identity information of all employees to be identified is determined based on the recognition template.

[0015] The technical effects and advantages of the AI ​​facial recognition cloud service system supporting flexible attendance rules of this invention are as follows: By deeply analyzing employee work behavior data, the system achieves accurate prediction of work efficiency curves for employees in different positions and intelligent optimization of personalized attendance rules, thereby improving employee work efficiency and effectively meeting the management needs of modern enterprises for flexible working hours. Dynamic risk prediction and temporary attendance rule generation based on future weather conditions enable intelligent adjustment of employee work hours and methods, ensuring work efficiency while improving the enterprise's adaptability to extreme weather. The system employs a mechanism for monitoring facial appearance changes and adaptively adjusting feature templates, significantly reducing the decrease in facial recognition accuracy due to changes in employee appearance, thus improving the stability and reliability of the intelligent attendance system. Pre-caching and concurrent recognition technologies enable efficient processing of simultaneous facial recognition for multiple people, avoiding queuing delays caused by crowds, thereby improving the response speed and throughput of the intelligent attendance system. By combining AI empowerment, big data analysis, and dynamic optimization technologies, the system not only improves attendance efficiency and accuracy but also effectively addresses the recognition challenges brought about by changes in employee appearance, thereby reducing management costs and increasing employee satisfaction. This effectively adapts to the diversified work patterns of modern enterprises and promotes the digital and intelligent upgrade of attendance management. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of an AI face recognition cloud service system supporting flexible attendance rules, as described in Embodiment 1 of the present invention. Detailed Implementation

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

[0018] Example 1 Please see Figure 1 As shown in this embodiment, the AI ​​face recognition cloud service system supporting flexible attendance rules includes a rule configuration module, an efficiency-driven module, a weather-driven module, a feature management module, and a concurrent recognition module. Each module is connected via wired and / or wireless means to achieve data transmission between modules. Each module is deployed within a cloud service platform, which provides centralized data storage, computing processing, and AI model running capabilities to support high-concurrency access, remote management, and cross-regional deployment of the system.

[0019] The rules configuration module is used to configure flexible attendance rules.

[0020] Flexible attendance rules are a dynamic attendance management mechanism based on working hours. They do not rely on fixed working hours, but instead achieve flexible management of employees' working hours by setting clock-in time windows, core working periods and total working hours requirements. This gives employees reasonable time autonomy and improves work flexibility and satisfaction. Flexible attendance rules are defined by enterprises based on job attributes (i.e., the characteristics or responsibilities of the employee's position), business needs (i.e., the specific work requirements or business rhythm of the enterprise in the actual operation process), and regional time zone (i.e., the time zone where the enterprise is located). The defined flexible attendance rules are configured in the cloud service platform for unified management. For example, a flexible attendance rule is implemented for R&D employees, specifically: allowed clock-in time window: 08:00-10:00 can start work and 17:00-19:00 can leave work; core working period: 10:00-15:00 is the critical working time during which employees must be on duty; total working hours requirement: employees must complete a total of 8 hours of work per day, and can freely allocate their clock-in and clock-out time within the clock-in time window.

[0021] The efficiency-driven module is used to collect work behavior data of each employee, perform time-period efficiency analysis on the work behavior data, generate work efficiency curves for each employee, and intelligently optimize the flexible attendance rules based on the work efficiency curves to form personalized attendance rules for each employee.

[0022] Work behavior data includes employee work activity data at different times throughout the day. Work activity data includes system operation data, communication and collaboration data, and business output data. Each set of work behavior data corresponds to one employee. The time period is divided according to a preset granularity, which is preset by those skilled in the art based on actual conditions. For example, if the granularity is set to one hour, the system records the employee's work activity data for that hour every hour, so the time period is one hour. System operation data consists of the specific operational behavior data of employees on their work terminals, such as the usage time and frequency of office applications, the number of mouse clicks, and the number of keyboard keystrokes. This data is obtained by calling the operating system (such as SetWindowsHookEx in Windows) and the office software interface. Communication and collaboration data refers to data on employees' interactive behaviors in team collaboration, communication, and meetings, such as the number of emails sent and received, meeting participation time, and speaking frequency. This data is obtained by integrating open APIs of collaboration platforms such as enterprise email, instant messaging, and video conferencing. Business output data reflects the actual business results created by employees. For example, sales positions include the number of customer visits and sales revenue; customer service positions include the number of customer inquiries handled and customer satisfaction; and design positions include the number of design drafts and the number of revisions. This data is obtained through interfaces with business systems (such as CRM, ERP, etc.), version control platforms (such as Git, SVN, Figma, etc.), and task management tools (such as Jira, Trello, Asana, etc.).

[0023] It should be noted that by using system operation data, communication and collaboration data, and business output data, employee work efficiency can be comprehensively evaluated from different dimensions: system operation data reflects the employee's work activity and level of engagement, and reflects the employee's level of focus during working hours; communication and collaboration data measures the timeliness and smoothness of information transmission and teamwork, affecting task flow efficiency; business output data directly reflects the quantity and quality of results per unit of time, reflecting the ability to convert input into value; combining these three types of data allows for a comprehensive evaluation of employee work efficiency from the complete input-collaboration-output chain.

[0024] It should be understood that the scope of work behavior data collection should be clearly defined, limited to information directly related to job responsibilities, in order to protect employees' personal privacy and data security.

[0025] Methods for performing time-based efficiency analysis on work behavior data include: Based on each employee's work behavior data, we obtain each employee's work activity data for each time period. We then input the work activity data for each time period into a pre-trained index analysis model to predict the efficiency index set for each employee in each time period. The index analysis model includes an input analysis model, a collaboration analysis model, and an output analysis model. The efficiency index set includes a work input index, a collaboration efficiency index, and an output efficiency index. The work input index is predicted by inputting system operation data into the input analysis model; the collaboration efficiency index is predicted by inputting communication and collaboration data into the collaboration analysis model; and the output efficiency index is predicted by inputting business output data into the output analysis model. The index analysis model is specifically a deep neural network model, which includes an input layer, hidden layers, and an output layer. Each hidden layer contains multiple neurons, and each neuron is connected to the neurons in the next layer. The connections contain weights that determine the importance and influence of data transmitted in the neural network. An activation function is applied to each neuron between the hidden layer and the output layer. The activation function introduces non-linearity, allowing the network to learn more complex patterns and features. The deep neural network model is a current technology, and the specific training process will not be elaborated on here.

[0026] A preset set of efficiency weights is provided, which includes the weight coefficients of each index in the efficiency index set for different positions. These weights are preset by those skilled in the art based on the business characteristics of different positions. For example, for the customer service position, the core of the customer service position is to communicate with customers and respond to problems quickly. Therefore, the collaboration efficiency index has the highest weight coefficient, followed by the output efficiency index, and the work input index has the lowest weight coefficient. Based on the efficiency weight set, the work input index, collaboration efficiency index and output efficiency index of the same employee in the same time period are weighted and summed to obtain the comprehensive work efficiency of each employee in each time period.

[0027] Methods for generating the productivity curve for each employee include: Collect work behavior data of each employee during the previous week, and perform time-period efficiency analysis on the daily work behavior data to obtain the comprehensive work efficiency of each employee during each time period during the previous week; for each employee's comprehensive work efficiency, calculate the average comprehensive work efficiency of the same employee in the same time period to obtain the average work efficiency of each employee in each time period. An efficiency coordinate system is constructed for each employee, where the horizontal axis represents a time point and the vertical axis represents the average work efficiency. The midpoint of the time period corresponding to each average work efficiency is taken as the time point of the corresponding average work efficiency, and each average work efficiency and its corresponding time point are used as a set of efficiency sampling points. For example, if the time period corresponding to an average work efficiency is 09:00-10:00, then the time point corresponding to the average work efficiency is 09:30. In the efficiency coordinate system for each employee, the corresponding efficiency sampling points are marked sequentially. A smoothing fitting algorithm (such as spline interpolation, local weighted regression, moving average, etc.) is used to fit the efficiency sampling points in each efficiency coordinate system sequentially to generate the work efficiency curve for each employee.

[0028] Methods for intelligently optimizing flexible attendance rules include: The system obtains the core working hours, total working hours requirements, and check-in time range from the flexible attendance rules. The check-in time range is the duration of the time covered by the check-in time window. Based on the core working hours, total working hours requirements, and check-in time range, multiple optimized time windows are constructed. For each optimized time window, the overall work efficiency and low-efficiency time for each employee are calculated sequentially based on their work efficiency curve. Using a preset weighting coefficient, the overall work efficiency and low-efficiency time for each employee are weighted to obtain their comprehensive efficiency score for each optimized time window. The comprehensive efficiency scores of the same employee in different optimized time windows are compared, and the optimized time window with the highest comprehensive efficiency score is taken as the optimal time window for that employee. Based on the optimal time window for each employee, the check-in time windows within the flexible attendance rules are intelligently optimized to form personalized attendance rules for each employee. The weighting coefficients are preset by those skilled in the art based on actual conditions.

[0029] The clock-in time window includes the work start time window and the get off work end time window. The work start time window indicates the period during which employees can clock in when they start work, and the get off work end time window indicates the period during which employees can clock in when they leave work. For optimizing time windows, the start time window is marked as the start optimization window, and the end time window is marked as the end optimization window. The range of both the start and end optimization windows is the same as the corresponding clock-in / out time range. The time period between the end time of the start optimization window and the start time of the end optimization window should completely cover the core work period. The start time of the end optimization window is equal to the sum of the start time of the morning time window and the total working hours requirement, and the end time of the end optimization window is equal to the sum of the end time of the start optimization window and the total working hours requirement. For example, if the clock-in time window is 08:00-10:00 and 17:00-19:00, the clock-in time range is 2 hours; the core working period is 11:00-15:00, and the total working hours requirement is 9 hours, then the optimized time window could be 08:01-10:01 and 17:01-19:01, 08:02-10:02 and 17:02-19:02, etc.

[0030] The overall work efficiency is calculated as follows: obtain the start time corresponding to the start time optimization window and mark it as the start point; obtain the end time corresponding to the end time optimization window and mark it as the end point; obtain the average work efficiency between the start point and the end point in the efficiency coordinate system, and calculate the mean to obtain the overall work efficiency.

[0031] The method for calculating the duration of inefficiency is as follows: A low-efficiency threshold is preset, which is determined by those skilled in the art based on actual conditions; efficiency sampling points between the start and end points in the efficiency coordinate system are marked as analysis points, and analysis points with average working efficiency equal to the low-efficiency threshold are marked as critical points; every two adjacent critical points are considered as a critical set, and the center point of each critical set is calculated; the center point is the midpoint between the two critical points in the corresponding critical set; if the average working efficiency corresponding to the center point is lower than the low-efficiency threshold, the corresponding critical set is marked as an inefficient set; if the average working efficiency corresponding to the center point is higher than the low-efficiency threshold, the corresponding critical set is not marked; the time difference between the two critical points in each inefficient set is calculated and summed sequentially to obtain the duration of inefficiency.

[0032] The comprehensive efficiency score is calculated as follows: Subtract the weighting coefficient from 1 to obtain the supplementary coefficient; mark the earliest efficiency sampling point with a non-zero average work efficiency in the efficiency coordinate system as the starting point, and mark the latest efficiency sampling point with a non-zero average work efficiency as the ending point; calculate the time difference between the starting point and the ending point to obtain the total working hours; calculate the ratio between the low-efficiency time and the total working hours to obtain the standard low-efficiency time; subtract the product of the supplementary coefficient and the standard low-efficiency time from the product of the weighting coefficient and the overall work efficiency to obtain the comprehensive efficiency score.

[0033] The weather-driven module is used to obtain future weather information and determine whether it is extreme weather. If it is extreme weather, it intelligently generates temporary attendance rules for each employee based on the future weather information and temporarily overrides the corresponding personalized attendance rules.

[0034] Future weather information refers to the meteorological parameters for the enterprise's location within the next day, obtained through meteorological APIs (such as Hefeng Weather, Moji Weather, etc.); meteorological parameters include warning type and warning level, warning types such as rainstorm, snowstorm, high temperature, typhoon, etc., and warning levels include blue, yellow, orange and red; Methods for determining whether something is extreme weather include: Different numerical labels are assigned to different warning levels and marked as level labels. The higher the severity of the warning level, the larger the value of the level label; that is, a red level label is greater than an orange level label, an orange level label is greater than a yellow level label, and a yellow level label is greater than a blue level label. The level labels corresponding to the warning levels in future weather information are compared with preset label thresholds. If the level label is greater than or equal to the label threshold, it is considered extreme weather; if the level label is less than the label threshold, it is not considered extreme weather. The label thresholds are preset by those skilled in the art based on the numerical range of the level labels.

[0035] Methods for intelligently generating temporary attendance rules for each employee include: Based on future weather information, a weather risk value is calculated, and a decision is made on whether to implement work-from-home measures based on the weather risk value. That is, the weather risk value is compared with a preset risk threshold. If the weather risk value is greater than or equal to the risk threshold, work-from-home measures are implemented; if the weather risk value is less than the risk threshold, work-from-home measures are not implemented. The risk threshold is preset by a person skilled in the art based on the actual situation. If working from home is implemented, the attendance method will be changed from facial recognition to online check-in, and no temporary attendance rules will be generated. If working from home is not implemented, the delay coefficients for different commuting methods will be obtained based on future weather information. The commuting method and home address of each employee will be obtained, and the normal commuting time for each employee will be calculated. Based on the normal commuting time of each employee and the delay coefficient of the corresponding commuting method, the delayed commuting time will be calculated. An uncertainty buffer will be applied to each delayed commuting time to obtain the final commuting time for each employee. The difference between the final commuting time and the corresponding normal commuting time for each employee will be calculated to obtain the delay time for each employee. Obtain the clock-in time window from each employee's personalized attendance rules and mark it as the actual clock-in window. Based on the delay time, shift the actual clock-in window for each employee to obtain the delayed clock-in window. That is, based on the delay time, shift both the start and end time windows within the actual clock-in window. Based on the delayed clock-in window, recalculate each employee's overall efficiency score and mark it as the delayed efficiency score. Based on each employee's overall efficiency score and delayed efficiency score for the optimal time window, determine again whether to implement work-from-home. If work-from-home is still not implemented, replace the actual start window in the employee's personalized attendance rules with the corresponding delayed start window to obtain the employee's temporary attendance rules.

[0036] For example, if the actual check-in windows are 08:00-10:00 and 17:00-19:00, and the delay time is half an hour, then the delayed check-in windows are 08:30-10:30 and 17:30-19:30.

[0037] The method for calculating weather risk values ​​is as follows: different risk coefficients are assigned to different warning types, with higher risk coefficients for warning types that cause more severe commuting disruptions. For example, the risk coefficient for typhoons is greater than that for rainstorms, and the risk coefficient for rainstorms is greater than that for high temperatures. Based on future weather information, the corresponding level labels and risk coefficients are obtained and multiplied. The product is then multiplied by a preset risk adjustment coefficient to obtain the weather risk value. The risk adjustment coefficient is preset by those skilled in the art based on actual conditions and is used to adjust the sensitivity of the weather risk value.

[0038] The method for obtaining delay coefficients under different commuting modes is as follows: A delay coefficient table is constructed, which includes the delay coefficients corresponding to different commuting modes under different warning types and warning levels. Based on future weather information, the delay coefficients for different commuting modes are obtained from the delay coefficient table. The method for constructing the delay coefficient table is as follows: The commuting times of employees using different commuting modes under different warning types and warning levels are collected and marked as historical delay times. These historical delay times are uploaded by each employee to the company's employee information management system. The ratio between each historical delay time and its corresponding normal commuting time is calculated to obtain the sub-delay coefficients for different commuting modes under different warning types and warning levels. The sub-delay coefficients with the same warning type, warning level, and commuting mode are averaged to obtain the delay coefficients for different commuting modes under different warning types and warning levels.

[0039] The method for calculating each employee's normal commute time is as follows: based on each employee's commute method, home address, and company address, the commute time of each employee under normal weather conditions is calculated through map APIs (such as Gaode Maps, Baidu Maps, etc.). The commute method and home address are uploaded by each employee to the company's employee information management system, while the company address is pre-entered by the company.

[0040] The method for calculating delayed commuting time is as follows: calculate the product of the normal commuting time and the delay coefficient of the corresponding commuting mode to obtain the delayed commuting time.

[0041] The method for obtaining the final commute time for each employee is as follows: From all historical delay times corresponding to each employee, obtain the historical delay time corresponding to the meteorological parameters in the future weather information and mark it as the current delay time; treat each employee's delayed commute time and the corresponding current delay time as a delay set; calculate the 80th percentile corresponding to each delay set and mark it as a safety buffer time; calculate the sum of each employee's safety buffer time and the preset fixed buffer time to obtain each employee's final commute time, thereby achieving a buffer against the uncertainty of delayed commute time; the fixed buffer time is preset by those skilled in the art based on actual conditions.

[0042] The calculation method for the safety buffer time is as follows: Sort all values ​​in the delay set from smallest to largest to generate a delay sequence; count the number of values ​​in the delay sequence and mark them as the delay quantity; subtract one from the delay quantity, multiply by 0.8, and add one to obtain the safety position value; determine whether the safety position value is an integer. If it is, obtain the value corresponding to the safety position value in the delay sequence and use it as the safety buffer time; if not, the integer part of the safety position value is called the safety integer value; obtain the value corresponding to the safety integer value in the delay sequence and mark it as the safety value; mark the value one position after the safety value in the delay sequence as the safety post-value; perform linear interpolation between the safety value and the safety post-value based on the safety position value to obtain the safety buffer time; specifically: the decimal part of the safety position value is called the safety decimal value; subtract the safety value from the safety post-value, multiply by the safety decimal value, and add the safety value to obtain the safety buffer time.

[0043] The method for determining whether to implement work-from-home is as follows: calculate the difference between the overall efficiency score and the delayed efficiency score to obtain the efficiency decline rate; compare the efficiency decline rate with a preset decline threshold, which is preset by those skilled in the art based on the actual situation; if the efficiency decline rate is greater than or equal to the decline threshold, then work-from-home is implemented; if the efficiency decline rate is less than the decline threshold, then work-from-home is not implemented.

[0044] The feature management module is used to calculate the appearance difference in real time after each face recognition using a feature template, and to distinguish the degree of change based on the appearance difference. If the degree of change is slight, the feature template is adjusted; if the degree of change is significant, a new feature template is generated.

[0045] Methods for real-time calculation of appearance differences include: The process involves acquiring face images used in face recognition, calculating the texture complexity and self-similarity matrix of the face images, and extracting face feature vectors from the face images. The texture complexity, self-similarity matrix, and face feature vectors are then integrated to obtain a comprehensive feature vector. The values ​​in the comprehensive feature vector are then normalized sequentially to obtain a standard feature vector. The standard feature vector is compared with the employee's feature template, and Euclidean distance is used to calculate the difference in appearance. The feature template also includes the normalized texture complexity, self-similarity matrix, and face feature vectors, and is extracted from the face images used when employees register their face information.

[0046] The texture complexity is calculated as follows: the face image is converted to grayscale to obtain a grayscale image; the occurrence probability of each grayscale value in the grayscale image is calculated, and the information entropy corresponding to all occurrence probabilities is calculated to obtain the texture complexity; the occurrence probability is calculated as follows: the number of pixels corresponding to each grayscale value is counted and marked as the number of pixels; the total number of pixels in the grayscale image is counted and marked as the total number of pixels; the ratio between the number of each pixel and the total number of pixels is calculated to obtain the occurrence probability of the grayscale value corresponding to each pixel.

[0047] The method for calculating the self-similarity matrix is ​​as follows: divide the face image into... Personal face area, The integers are greater than 1; the HOG method is used to extract the directional gradient feature vector corresponding to each face region; based on the directional gradient feature vector, the cosine similarity is used to calculate the region similarity between each pair of face regions; all region similarities are arranged into a grid of size . A self-similar matrix; wherein the elements on the main diagonal of the self-similar matrix all have the value 1.

[0048] It should be noted that the facial feature vectors in the face image are extracted by a deep learning-based face recognition model. Both the face recognition model and the HOG method are existing technologies, and the specific process will not be elaborated on here. It should be understood that calculating appearance differences based on three methods—texture complexity, self-similarity matrix, and facial feature vector—can comprehensively reflect changes in a person's face: texture complexity captures subtle changes in overall texture, such as skin color, makeup, or beard; self-similarity matrix reflects changes in local structural relationships, such as glasses, accessories, or local makeup disrupting local regularity; and facial feature vector, extracted through deep learning, retains global identity information, ensuring accurate overall difference assessment. The combination of these three methods can sensitively capture both local and overall changes while maintaining stable identity recognition, thus more robustly judging the slight or significant degree of appearance changes.

[0049] Methods for distinguishing the degree of change based on differences in appearance include: A preset threshold set is provided, which includes a slight threshold and a significant threshold, with the significant threshold being greater than the slight threshold. The threshold set is preset by those skilled in the art based on actual conditions. The appearance difference is compared with the slight threshold and the significant threshold respectively. If the appearance difference is less than the slight threshold, the degree of change is determined to be no change. If the appearance difference is greater than or equal to the slight threshold but less than the significant threshold, the degree of change is determined to be a slight change. If the appearance difference is greater than or equal to the significant threshold, the degree of change is determined to be a significant change.

[0050] Methods for adjusting feature templates include: The feature template compared with the standard feature vector is marked as the current template, and the feature template located one position before the current template is marked as the predecessor template. That is, the current template is the feature template obtained by adjusting the predecessor template; the method used to adjust the predecessor template is also obtained. A feature template is selected and marked as an adjustment template. The value is an integer greater than 2; obtain the acquisition time (i.e., the time of acquiring the feature template) corresponding to each adjustment template, and delete the adjustment template with the earliest acquisition time; where, if the number of adjustment templates is less than If so, the adjustment template with the earliest acquisition time will not be deleted; Mark the acquisition time corresponding to the standard feature vector as the current time. Based on the acquisition time of each adjustment template, calculate the time difference value of each adjustment template in turn; that is, subtract the acquisition time of each adjustment template from the current time to obtain the time difference value of each adjustment template. A preset time decay function is used to successively substitute each time difference into the time decay function to obtain the time weight of each adjustment template. Based on the time weight, a weighted average is calculated for all adjustment templates and the standard feature vector to obtain a weighted template. The current template is then adjusted based on the weighted template. It should be noted that when there is no preceding template, the current template and the standard feature vector are calculated based on the time weight to obtain the weighted template. The time decay function is a function that monotonically decreases as the time difference increases. It is used to quantify the impact of the distance between the acquisition time of each adjustment template and the current time on the time weight. The specific form can be an exponential decay function, a linear decay function, etc., which can be pre-designed by those skilled in the art according to the actual situation. The time difference value corresponding to the standard feature vector is 0, so the corresponding time weight is 1.

[0051] It should be noted that when the degree of change is significant, the standard feature vector extracted from the face image is used as the new feature template for the corresponding employee, while the old feature template is still retained.

[0052] The concurrent recognition module is used to obtain the pre-attendance set for the next time period and pre-cache the feature templates of each employee in the pre-attendance set to form a template set. Based on the template set, face recognition is performed on multiple employees simultaneously.

[0053] Methods for obtaining the pre-attendance set include: Obtain the real-time time, calculate the sum of the real-time time and the partition granularity to obtain the cache time; construct a cache period based on the real-time time and cache time, which is the next time period; mark the clock-in time window in each employee's personalized attendance rule or temporary attendance rule as an analysis window, compare each analysis window with the cache period, and determine whether there is an intersection between each analysis window and the cache period, i.e., whether there is a time overlap between the analysis window and the cache period; if there is an intersection, mark the employee corresponding to the analysis window as a cache employee; if there is no intersection, do not mark the corresponding employee; combine all cache employees to obtain the pre-attendance set; Methods for performing facial recognition on multiple employees simultaneously include: When multiple employees simultaneously pass through the facial recognition area (i.e., a specific area within the company used to collect and recognize facial images), cameras deployed within the facial recognition area continuously capture multiple images of the area. Employees passing through the facial recognition area are marked as candidates for identification. Each image is then segmented sequentially to obtain multiple sets of candidates for identification. Each set of candidates for identification includes... Zhang is an image to be identified. The number of images in the region represents the number of images, and each set of images to be identified corresponds to one employee to be identified; image segmentation methods include Mask R-CNN, SOLOv2, etc. The standard feature vectors corresponding to each image to be identified are extracted sequentially and marked as recognition feature vectors. Each recognition feature vector is compared with each feature template in the template set to calculate the appearance difference degree corresponding to each image to be identified and marked as the recognition difference degree. The average recognition difference degree of the corresponding feature template and the employee to be identified is calculated to obtain the matching difference degree between each employee to be identified and each feature template in the template set. The feature template with the smallest matching difference degree is used as the recognition template of the corresponding employee to be identified, and the identity information of all employees to be identified is determined based on the recognition template.

[0054] It should be understood that in order to avoid unclear face recognition due to only collecting one regional image, it is necessary to continuously collect multiple regional images. Since the content of multiple regional images is similar, when performing image segmentation, multiple images to be identified that are in the same position in the regional images can be identified as corresponding to the same employee to be identified. This can result in multiple sets of to be identified, that is, multiple images to be identified in each set of to be identified belong to the same employee to be identified, but it has not yet been determined which specific employee they correspond to.

[0055] It should be noted that the face image mentioned in the feature management module is the image with the smallest recognition difference between the image of the employee to be identified and the recognition template.

[0056] This embodiment achieves accurate prediction of employee work efficiency curves and intelligent optimization of personalized attendance rules by deeply analyzing employee work behavior data, thereby improving employee work efficiency and effectively meeting the management needs of modern enterprises for flexible working hours. By combining future weather conditions for dynamic risk prediction and generating temporary attendance rules, it can intelligently adjust employee work hours and methods, ensuring work efficiency while improving the enterprise's adaptability to extreme weather. Employing a mechanism for monitoring facial appearance changes and adaptively adjusting feature templates significantly reduces the problem of decreased facial recognition accuracy due to changes in employee appearance, thus improving the stability and reliability of the intelligent attendance system. Through pre-caching and concurrent recognition technologies, it achieves efficient processing of multiple people's facial recognition simultaneously, avoiding queuing delays caused by crowds, thereby improving the response speed and throughput of the intelligent attendance system. By combining AI empowerment, big data analysis, and dynamic optimization technologies, it not only improves attendance efficiency and accuracy but also effectively addresses the recognition challenges brought about by changes in employee appearance, thereby reducing management costs and improving employee satisfaction. This effectively adapts to the diversified work patterns of modern enterprises and promotes the digital and intelligent upgrade of attendance management.

[0057] Example 2 This application also provides an electronic device. The electronic device may include one or more processors and one or more memories. The memories store computer-readable code, which, when executed by the one or more processors, can perform the AI ​​facial recognition cloud service system supporting flexible attendance rules as described above.

[0058] The methods or systems according to the embodiments of this application can also be implemented using the architecture of the electronic device shown in this application. The electronic device may include a bus, one or more CPUs, ROM, RAM, a communication port connected to a network, input / output, a hard disk, etc. The storage device in the electronic device, such as a ROM or hard disk, may store the AI ​​face recognition cloud service system supporting flexible attendance rules provided in this application. Furthermore, the electronic device may also include a user interface. Of course, the architecture shown in this application is merely exemplary; when implementing different devices, one or more components of the electronic device shown in this application may be omitted according to actual needs.

[0059] Example 3 One embodiment of this application discloses a computer-readable storage medium. The computer-readable storage medium stores computer-readable instructions. When the computer-readable instructions are executed by a processor, an AI face recognition cloud service system supporting flexible attendance rules, as described in the above figures according to an embodiment of this application, can be executed. The storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0060] Furthermore, according to embodiments of this application, the processes described in the above-referenced flowcharts can be implemented as computer software programs. For example, this application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be executed by a processor to perform instructions corresponding to the method steps provided in this application, such as an AI face recognition cloud service system supporting flexible attendance rules. When this computer program is executed by a central processing unit (CPU), it performs the functions defined in the method of this application.

[0061] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0062] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0063] In the description of this invention, it should be understood that the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0064] In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0065] In the description of this invention, "several" means one or more, and "a large number" means two or more.

[0066] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0067] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0068] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An AI facial recognition cloud service system supporting flexible attendance rules, characterized in that: include: The rules configuration module is used to configure flexible attendance rules; The efficiency-driven module is used to collect work behavior data of each employee, perform time-period efficiency analysis on the work behavior data, generate work efficiency curves for each employee, and intelligently optimize the flexible attendance rules based on the work efficiency curves to form personalized attendance rules for each employee. The weather-driven module is used to obtain future weather information and determine whether it is extreme weather. If it is extreme weather, it will intelligently generate temporary attendance rules for each employee based on the future weather information and temporarily override the corresponding personalized attendance rules. The feature management module is used to calculate the appearance difference in real time after each face recognition using a feature template, and to distinguish the degree of change based on the appearance difference. If the degree of change is slight, the feature template is adjusted; if the degree of change is significant, a new feature template is generated. The concurrent recognition module is used to obtain the pre-attendance set for the next time period and pre-cache the feature templates of each employee in the pre-attendance set to form a template set. Based on the template set, face recognition is performed on multiple employees simultaneously.

2. The AI ​​face recognition cloud service system supporting flexible attendance rules according to claim 1, characterized in that, Methods for performing time-based efficiency analysis on work behavior data include: Work behavior data includes employees' work activity data at different times throughout the day; each set of work behavior data corresponds to one employee, and the time periods are divided according to a preset granularity. Based on the work behavior data of each employee, obtain the work activity data of each employee in each time period; input the work activity data corresponding to each time period into the trained index analysis model to predict the set of efficiency indices corresponding to each employee in each time period. A preset efficiency weight set is used, which includes the weight coefficient of each index in the efficiency index set under different positions. Based on the efficiency weight set, the efficiency index set of the same employee in the same time period is weighted and summed to obtain the comprehensive work efficiency of each employee in each time period.

3. The AI ​​face recognition cloud service system supporting flexible attendance rules according to claim 2, characterized in that, Methods for generating the productivity curve for each employee include: Collect work behavior data of each employee during the previous week, and perform time-period efficiency analysis on the daily work behavior data to obtain the comprehensive work efficiency of each employee during each time period during the previous week; for each employee's comprehensive work efficiency, calculate the average comprehensive work efficiency of the same employee in the same time period to obtain the average work efficiency of each employee in each time period. An efficiency coordinate system is constructed for each employee; the midpoint of the time period corresponding to each average work efficiency is taken as the time point of the corresponding average work efficiency, and each average work efficiency and its corresponding time point are taken as a set of efficiency sampling points; in the efficiency coordinate system corresponding to each employee, the corresponding efficiency sampling points are marked in sequence; a smooth fitting algorithm is used to fit the efficiency sampling points in each efficiency coordinate system in sequence to generate the work efficiency curve of each employee.

4. The AI ​​face recognition cloud service system supporting flexible attendance rules according to claim 3, characterized in that, Methods for intelligently optimizing flexible attendance rules include: Flexible attendance rules include clock-in time windows, core working periods, and total working hours requirements; Based on the check-in time window, the covered time duration interval is obtained and marked as the check-in time range. Multiple optimized time windows are constructed based on the core work period, total working hours requirements, and the check-in time range. For each optimized time window, the overall work efficiency and low-efficiency time duration for each employee are calculated sequentially based on their work efficiency curve. Using a preset weighting coefficient, the overall work efficiency and low-efficiency time duration for each employee are weighted to obtain their comprehensive efficiency score for each optimized time window. The comprehensive efficiency scores of the same employee in different optimized time windows are compared, and the optimized time window with the highest comprehensive efficiency score is taken as the optimal time window for that employee. Based on the optimal time window for each employee, the check-in time windows within the flexible attendance rules are intelligently optimized to form personalized attendance rules for each employee.

5. The AI ​​face recognition cloud service system supporting flexible attendance rules according to claim 4, characterized in that, Methods for intelligently generating temporary attendance rules for each employee include: Based on future weather information, calculate the weather risk value and determine whether to implement work-from-home arrangements based on the weather risk value. If work-from-home arrangements are not implemented, obtain the delay coefficient for different commuting modes based on future weather information. Obtain each employee's commuting mode and home address, and calculate each employee's normal commuting time. Calculate the delayed commuting time based on each employee's normal commuting time and the delay coefficient for the corresponding commuting mode. Apply an uncertainty buffer to each delayed commuting time to obtain each employee's final commuting time. Calculate the difference between each employee's final commuting time and the corresponding normal commuting time to obtain each employee's delay time. Obtain the check-in time window from each employee's personalized attendance rules and mark it as the actual check-in window; based on the delay time, shift the actual check-in window for each employee to obtain the delayed check-in window; based on the delayed check-in window, recalculate each employee's comprehensive efficiency score and mark it as the delayed efficiency score; based on each employee's comprehensive efficiency score and delayed efficiency score for the optimal time window, determine again whether to implement work-from-home; if work-from-home is still not implemented, replace the actual work window in the corresponding employee's personalized attendance rules with the corresponding delayed work window to obtain the corresponding temporary attendance rules for the employee.

6. The AI ​​face recognition cloud service system supporting flexible attendance rules according to claim 5, characterized in that, The method for obtaining each employee's final commute time is as follows: Future weather information refers to meteorological parameters for the area where the company is located for the next day, including the type and level of the warning. Collect the commuting times of employees with different commuting methods under different warning types and warning levels, and mark them as historical delay times; from all historical delay times for each employee, obtain the historical delay times corresponding to meteorological parameters in future weather information, and mark them as current delay times; combine each employee's delayed commuting time and corresponding current delay time as a delay set; calculate the 80th percentile for each delay set and mark it as a safety buffer time; calculate the sum of each employee's safety buffer time and the preset fixed buffer time to obtain each employee's final commuting time.

7. The AI ​​face recognition cloud service system supporting flexible attendance rules according to claim 6, characterized in that, Methods for real-time calculation of appearance differences include: The process involves acquiring face images used in face recognition, calculating the texture complexity and self-similarity matrix of the face images, and extracting face feature vectors from the face images. The texture complexity, self-similarity matrix, and face feature vectors are then integrated to obtain a comprehensive feature vector. The values ​​in the comprehensive feature vector are then normalized sequentially to obtain a standard feature vector. Finally, the standard feature vectors are compared with the employee's feature templates, and Euclidean distance is used to calculate the degree of appearance difference.

8. The AI ​​face recognition cloud service system supporting flexible attendance rules according to claim 7, characterized in that, Methods for adjusting feature templates include: The feature template compared with the standard feature vector is marked as the current template, and the feature template located one position before the current template is marked as the predecessor template; the method used to adjust the predecessor template is obtained. A feature template is selected and marked as an adjustment template. The value is an integer greater than 2; obtain the acquisition time corresponding to each adjustment template, and delete the adjustment template with the earliest acquisition time; where, if the number of adjustment templates is less than 2... If the earliest acquisition time is not deleted, the acquisition time corresponding to the standard feature vector is marked as the current time, and the time difference of each acquisition template is calculated in turn according to the acquisition time of each acquisition template. A preset time decay function is used to substitute each time difference value into the time decay function to obtain the time weight of each adjustment template. Based on the time weight, a weighted average is calculated for all adjustment templates and the standard feature vector to obtain a weighted template. Based on the weighted template, the current template is adjusted.

9. The AI ​​face recognition cloud service system supporting flexible attendance rules according to claim 8, characterized in that, Methods for obtaining the pre-attendance set include: Obtain the real-time time, calculate the sum of the real-time time and the granularity of the segmentation to obtain the cache time; construct the cache period based on the real-time time and the cache time; mark the clock-in time window in the personalized attendance rules or temporary attendance rules corresponding to each employee as an analysis window, compare each analysis window with the cache period to determine whether there is an intersection between each analysis window and the cache period; if there is an intersection, mark the employee corresponding to the analysis window as a cache employee; combine all cache employees to obtain the pre-attendance set.

10. The AI ​​face recognition cloud service system supporting flexible attendance rules according to claim 9, characterized in that, Methods for performing facial recognition on multiple employees simultaneously include: When multiple employees pass through the face recognition area simultaneously, multiple images of the face recognition area are continuously acquired. Employees passing through the face recognition area are marked as employees to be identified. Each area image is then segmented sequentially to obtain multiple sets of employees to be identified. Each set of employees to be identified includes... Zhang is an image to be identified. The number of images in the region; each set of images to be identified corresponds to one employee to be identified. The standard feature vectors corresponding to each image to be identified are extracted sequentially and marked as recognition feature vectors. Each recognition feature vector is compared with each feature template in the template set to calculate the appearance difference degree corresponding to each image to be identified and marked as the recognition difference degree. The average recognition difference degree of the corresponding feature template and the employee to be identified is calculated to obtain the matching difference degree between each employee to be identified and each feature template in the template set. The feature template with the smallest matching difference degree is used as the recognition template of the corresponding employee to be identified, and the identity information of all employees to be identified is determined based on the recognition template.

Images