A temporary bank branch security monitoring method and system

Abstract

The present application provides a temporary bank branch security monitoring method and system, which relates to the field of bank security monitoring technology. The method includes: first acquiring a monitoring image sequence to obtain the number of customers, then collecting business types and business amounts, obtaining business types and business amounts, analyzing the number of customers, clustering business types and business amounts, obtaining anomaly coefficients, determining the security monitoring level through the anomaly coefficients, performing face recognition on high-quality images to obtain anomaly levels, and performing real-time anomaly monitoring. The present application mainly solves the problem of being unable to perform advance warning or real-time analysis, unable to predict abnormal time and prevention in real time, and also solves the problem of high personnel dependence. By analyzing the number of customers, clustering business types and business amounts, setting security monitoring levels, and performing real-time anomaly monitoring, the security and reliability of temporary bank branches are improved, and various security risks are effectively prevented and responded to.

Description

Technical Field

[0001] The present invention relates to the technical field of bank security monitoring, and in particular to a temporary bank branch security monitoring method and system. Background Art

[0002] With the continuous development of technology, the monitoring system of temporary bank branches is also constantly improving. From the initial analog monitoring to digital monitoring, and then to the current intelligent monitoring, the clarity, storage capacity and intelligence of the monitoring system have been significantly improved. In general, the security monitoring work of temporary bank branches has received full attention. Banks will continue to improve the security monitoring system, strengthen security drills and training, and maintain close contact with public security organs to ensure the safety of temporary bank branches.

[0003] The existing technology is to install a full-range, all-around monitoring system at temporary bank outlets, covering all doors, windows and passages of the outlets to ensure that there are no blind spots. Conduct safety drills and training to enhance employees' safety awareness and emergency response capabilities. In response to possible emergencies at temporary bank outlets, corresponding emergency plans are formulated, and emergency plans are regularly drilled and revised.

[0004] Existing technologies are usually "post-event video query systems". Although they can record and trace back events that have already occurred, they cannot provide advance warnings or real-time analysis, and cannot predict and prevent events that may occur but have not yet occurred. High dependence on personnel The execution of high-security monitoring often requires human resources, such as viewing monitoring images, judging and handling abnormal situations, etc. This not only increases labor costs, but may also lead to missed reports or false reports due to human factors. Summary of the invention

[0005] This application mainly solves the problem of being unable to conduct advance warning or real-time analysis, and being unable to predict and prevent events that may happen but have not yet occurred. It is highly dependent on personnel, and the execution of security monitoring often requires manpower, such as viewing monitoring images, judging and handling abnormal situations, etc. This not only increases labor costs, but may also lead to problems of missed reports or false reports due to human factors.

[0006] In view of the above problems, the embodiments of the present application provide a temporary bank branch security monitoring method and system. In a first aspect, the embodiments of the present application provide a temporary bank branch security monitoring method, the method comprising: collecting monitoring images of the temporary bank branch to be monitored through a branch information collection module to obtain a monitoring image sequence; identifying the number of customers in the monitoring image according to the monitoring image sequence, obtaining a customer number sequence, performing customer number anomaly analysis and customer flow anomaly analysis, and calculating the customer anomaly coefficient; collecting business types and business amounts in the temporary bank branch, obtaining a business type sequence and a business amount sequence, performing clustering anomaly analysis according to the business type sequence and the business amount sequence, and calculating the business anomaly coefficient; determining the security monitoring level according to the business anomaly coefficient and the customer anomaly coefficient through a monitoring requirement analysis module, and obtaining monitoring image quality information, face recognition number and recognition anomaly threshold according to the security monitoring level mapping; controlling the collection of high-quality monitoring images in the temporary bank branch according to the monitoring image quality information through a security monitoring module, searching and identifying faces in the high-quality monitoring images according to the face recognition number, obtaining anomaly levels, and marking faces greater than the recognition anomaly threshold as monitoring results.

[0007] In a second aspect, an embodiment of the present application provides a temporary bank branch security monitoring system, the system comprising: a monitoring image sequence acquisition module, the monitoring image series acquisition module is to collect monitoring images of the temporary bank branch to be monitored through the branch information acquisition module, and obtain a monitoring image sequence; a customer abnormality coefficient calculation module, the customer abnormality coefficient calculation module is to identify the number of customers in the monitoring image according to the monitoring image sequence, obtain a customer quantity sequence, perform customer quantity abnormality analysis and customer flow abnormality analysis, and calculate to obtain a customer abnormality coefficient; a business sequence acquisition module, the business sequence acquisition module is used to collect business types and business amounts in the temporary bank branch, obtain a business type sequence and a business amount sequence, and a business abnormality coefficient acquisition module, the business abnormality coefficient acquisition module The acquisition module is used to perform cluster anomaly analysis according to the business type sequence and the business amount sequence, and calculate the business anomaly coefficient. The monitoring security level acquisition module is used to make a decision to obtain the security monitoring level according to the business anomaly coefficient and the customer anomaly coefficient through the monitoring requirement analysis module, and obtain the monitoring image quality information, the number of face recognition and the recognition anomaly threshold according to the security monitoring level mapping. The monitoring result acquisition module is used to control the collection of high-quality monitoring images in temporary bank branches according to the monitoring image quality information through the security monitoring module, and search and identify the faces in the high-quality monitoring images according to the face recognition number, obtain the anomaly level, and mark the faces whose level is greater than the recognition anomaly threshold as the monitoring result.

[0008] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0009] The present application provides a temporary bank branch security monitoring method and system, which relates to the technical field of bank security monitoring. The method comprises: first acquiring a monitoring image sequence to obtain the number of customers, then collecting business types and business amounts to obtain business types and business amounts, analyzing the number of customers, and clustering the business types and business amounts to obtain anomaly coefficients, determining the security monitoring level based on the anomaly coefficients, performing face recognition on high-quality images to obtain anomaly levels, and performing real-time anomaly monitoring.

[0010] This application mainly solves the problem of being unable to conduct advance warning or real-time analysis, unable to predict abnormal time and prevention in real time, and also solves the problem of high dependence on personnel. By analyzing the number of customers, clustering analysis of business types and business amounts, setting security monitoring levels, and conducting real-time abnormal monitoring, the security and reliability of temporary bank branches are improved, and various security risks are effectively prevented and responded to.

[0011] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the present disclosure or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0013] Figure 1 A schematic diagram of a temporary bank branch security monitoring method flow is provided for an embodiment of the present application;

[0014] Figure 2 A schematic flow chart of a method for obtaining a customer quantity sequence by identifying multiple monitoring images in a temporary bank branch security monitoring method is provided in an embodiment of the present application;

[0015] Figure 3 A schematic flow chart of a method for calculating and obtaining a business abnormality coefficient by combining a type business abnormality coefficient and an amount business abnormality coefficient in a temporary bank branch security monitoring method is provided in an embodiment of the present application;

[0016] Figure 4A structural schematic diagram of a temporary bank branch security monitoring system is provided for an embodiment of the present application.

[0017] Explanation of reference numerals: monitoring image sequence acquisition module 10 , customer anomaly coefficient calculation module 20 , business sequence acquisition module 30 , business anomaly coefficient acquisition module 40 , monitoring security level acquisition module 50 , monitoring result acquisition module 60 . DETAILED DESCRIPTION

[0018] This application mainly solves the problem of being unable to conduct advance warning or real-time analysis, and being unable to predict and prevent events that may happen but have not yet occurred. It is highly dependent on personnel, and the execution of security monitoring often requires manpower, such as viewing monitoring images, judging and handling abnormal situations, etc. This not only increases labor costs, but may also lead to problems of missed reports or false reports due to human factors.

[0019] The technical solution in the embodiment of the present application is to solve the above problems, and the overall solution idea is as follows:

[0020] Through the branch information collection module, surveillance images of temporary bank branches to be monitored are collected to obtain a surveillance image sequence. According to the surveillance image sequence, the number of customers in the surveillance image is identified. According to the surveillance data records of the temporary bank branches, sample surveillance image records are obtained, the number of customers in the sample surveillance image records are identified and marked, and sample customer number records are obtained. Based on a convolutional neural network, a customer number recognition channel is constructed. The sample surveillance image records and sample customer number records are used to train the customer number recognition channel. The customer number recognition channel is used to recognize multiple surveillance images in the surveillance image sequence, and a customer number sequence is obtained to perform customer number anomaly analysis and customer flow anomaly analysis.

[0021] According to the monitoring time window of the monitoring image sequence, the monitoring data records of the temporary bank outlets are retrieved, and the historical average customer flow and the historical average customer flow change rate are calculated. According to the customer quantity sequence, the average customer flow and the average customer flow change rate are calculated. The average customer flow and the historical average customer flow, as well as the deviations of the average customer flow change rate and the historical average customer flow change rate are calculated respectively to obtain the flow anomaly coefficient and the change rate anomaly coefficient. The flow anomaly coefficient and the change rate anomaly coefficient are weighted to obtain the customer anomaly coefficient. And the customer anomaly coefficient is calculated;

[0022] The business types and business amounts in the temporary bank outlets are collected to obtain the business type sequence and the business amount sequence. According to the business type sequence and the business amount sequence, clustering anomaly analysis is performed, multiple business types in the business type sequence are clustered to obtain multiple type clustering results, and multiple business amounts in the business amount sequence are clustered using the K-means algorithm to obtain multiple amount clusters. According to the time window of the business type sequence and the business amount sequence, the business data records of the temporary bank outlets are retrieved to obtain multiple historical average business quantities and multiple historical average amounts. The deviations between the business type quantities in the multiple type clustering results and the multiple historical average business quantities are calculated respectively, and the type business anomaly coefficient is obtained by weighted calculation. The deviations between the amounts in the multiple amount clusters and the multiple historical average amounts are calculated respectively, and the amount business anomaly coefficient is obtained by weighted calculation. The type business anomaly coefficient and the amount business anomaly coefficient are combined to calculate the business anomaly coefficient. And the business anomaly coefficient is calculated.

[0023] Through the monitoring requirement analysis module, according to the business abnormality coefficient and the customer abnormality coefficient, a decision is made to obtain the security monitoring level. According to the monitoring data records of temporary bank outlets, sample customer abnormality coefficient records and sample business abnormality coefficient records are processed to obtain sample security monitoring level records. The size of the sample security monitoring level is positively correlated with the size of the sample customer abnormality coefficient and the sample business abnormality coefficient. The sample customer abnormality coefficient record and the sample business abnormality coefficient record are used as decision inputs, and the sample security monitoring level record is used as a decision result. Based on the decision tree, a monitoring requirement decision classifier is constructed. The monitoring requirement decision classifier is used to make decision classifications on the business abnormality coefficient and the customer abnormality coefficient to obtain the security monitoring level.

[0024] According to the high-quality surveillance image records of temporary bank outlets, sample high-quality surveillance image records are obtained, and the faces of each sample high-quality surveillance image in the sample high-quality surveillance image records are identified and marked to obtain face recognition result records. Semantic segmentation is used to construct a face recognition channel, and the sample high-quality surveillance image records and face recognition result records are used for training until convergence. The face recognition channel is used to identify and segment the face images in the high-quality surveillance images to obtain face recognition results. The face recognition results include multiple face images. According to the number of face recognitions, the face images in the face recognition results are searched and identified, and the face abnormality levels are obtained by analysis. Face images greater than the recognition abnormality threshold are marked. According to the high-quality surveillance image records of temporary bank outlets, sample face image records are obtained, and expression abnormality analysis and marking are performed to obtain sample face abnormality level records. Sample face image records and sample face abnormality level records are used to construct and train a face abnormality identifier based on a convolutional neural network. According to the number of face recognitions, multiple face images are randomly searched in the face recognition results according to multiple initial directions.

[0025] A face anomaly identifier is used to identify multiple face images, obtain multiple face anomaly levels, and determine whether they are greater than an anomaly recognition threshold. If so, mark them and continue to search in an initial direction. If not, do not mark them, adjust the initial direction, obtain an adjusted direction, and search. According to the inverse of the ratio of the multiple face anomaly levels to the anomaly recognition threshold, adjust the preset search step length to obtain an adjusted search step length. According to the adjusted search step length and the number of face recognitions, continue to search and identify in the face recognition results until a preset number of searches is reached as a monitoring result. According to the security monitoring level mapping, monitoring image quality information, the number of face recognitions, and the anomaly recognition threshold are obtained. Through the security monitoring module, according to the monitoring image quality information, the collection of high-quality monitoring images in temporary bank outlets is controlled. According to the number of face recognitions, faces in high-quality monitoring images are searched and identified to obtain anomaly levels, and faces greater than the anomaly recognition threshold are marked as monitoring results.

[0026] In order to better understand the above technical solution, the above solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification:

[0027] Embodiment 1

[0028] like Figure 1 A temporary bank branch security monitoring method is shown, the method is applied to a temporary bank branch security monitoring device, the device includes a branch information collection module, a monitoring requirement analysis module and a security monitoring module, the method includes:

[0029] Through the branch information collection module, the monitoring images of the temporary bank branches to be monitored are collected to obtain a monitoring image sequence;

[0030] Specifically, the monitoring images of the temporary bank branches to be monitored are collected through the branch information collection module. This module is mainly used to collect information about the inside and surrounding environment of the branch, including the collection of video, audio and other information. It is mainly composed of front-end collection equipment and back-end processors. The front end includes a high-definition camera with intelligent fill light or a combination device of a low-light night vision device and a CCD image sensor. The intelligent fill light can be used to fill light when there is a shadow or the light intensity is low, so that the captured video information is clear and visible. The low-light night vision device and the CCD image sensor combination device, in which the low-light night vision device can make the dark environment visible to the human eye, and the CCD image sensor can capture high-definition image information when the light intensity is sufficient. It also includes smoke sensors, temperature sensors, microphones and other equipment, and transmits the collected information to the back-end processing system through the data transmission network. The back-end processing system can store, analyze, process, and display the collected information. The monitoring image sequence refers to a series of orderly monitoring images. The monitoring image sequence can be organized in time sequence. A unique number can be set for each channel or business hall. In the collected image information, the sequence is organized according to the time sequence, and can be viewed by querying a specific time period. The monitoring center of the bank branch is equipped with multiple monitors or screens to display the monitoring image sequence of different channels or areas, so that security personnel can find abnormal situations in time and deal with them. In order to achieve real-time monitoring and management of the internal and surrounding environment of the branch.

[0031] According to the monitoring image sequence, identify the number of customers in the monitoring image, obtain the customer number sequence, perform customer number anomaly analysis and customer flow anomaly analysis, and calculate and obtain the customer anomaly coefficient;

[0032] Specifically, according to the monitoring image sequence, the number of customers in the monitoring image is identified. First, the foreground target and the background in the image are separated by using image segmentation technology, and then the customer target in the image is detected by using the target detection algorithm. The target detection algorithm divides the image into regions with different gray levels by setting different thresholds, and then extracts the customer target, and extracts and identifies the customer target. According to the processed image data, the number of customers in each monitoring image is counted by using computer vision technology. This number can form a time series, that is, the customer number sequence. For the customer number sequence, some abnormal indicators are defined, such as the mean, standard deviation, maximum value, minimum value, etc. of the customer number. By comparing these indicators, abnormal situations in the customer number sequence can be found. By calculating and analyzing the speed and time of customers entering and leaving the outlets in the monitoring image sequence, the customer flow data for each time period can be obtained. Customer flow data can also form a time series. Combining the customer number sequence and the customer flow sequence, the customer abnormality coefficient can be calculated.

[0033] Collect business types and business amounts in temporary bank outlets, and obtain business type sequences and business amount sequences;

[0034] Specifically, the business types and business amounts of temporary bank outlets are collected through the business system of temporary bank outlets. The business types of banks include traditional business and complex business. Traditional business includes loans, foreign exchange trading, trade financing, etc. Complex business includes derivatives, structured financing, leasing, etc. It also includes personal business, which includes deposits and withdrawals, remittances and transfers, credit and debit card business, loan business, wealth management business, etc. The amount of business recorded in the business handling process can be arranged according to time. It means that the bank has processed 100 businesses in a period of time, and the total amount of these businesses is 200,000 yuan, of which the amount of deposit business is 30,000 yuan, the amount of withdrawal business is 20,000 yuan, the amount of remittance business is 10,000 yuan, the amount of transfer business is 10,000 yuan, the amount of loan business is 100,000 yuan, and the amount of wealth management business is 30,000 yuan. By obtaining the bank's business type sequence and business amount sequence, it is possible to compare the popularity of various types of business in the bank, such as which amount is higher and which amount is lower, and these sequence data can be visualized.

[0035] Perform clustering anomaly analysis based on the business type sequence and business amount sequence, and calculate and obtain the business anomaly coefficient;

[0036] Specifically, before clustering anomaly analysis, the business type sequence and business amount sequence are first cleaned and processed, invalid data is removed, missing data is filled, and outlier data is standardized. Then, representative features are extracted from the business type sequence and business amount sequence, such as calculating the number and amount ratio of each business type, or calculating the change trend of the business amount, etc. Cluster analysis is then performed, and the extracted features are used as input. Clustering algorithms (such as K-means, DBSCAN, etc.) are used to cluster the data, and similar business types or business amounts are clustered into one category. In the clustering results, abnormal data are found through some anomaly detection algorithms (such as Z-score, IQR, etc.). For each clustering category, its anomaly coefficient is calculated. The anomaly coefficient can be defined as the ratio of the number or amount of abnormal data in the category to the total number or amount of the category. A large amount of data can be effectively processed, and some potential business risks and abnormal situations can be found.

[0037] Through the monitoring requirement analysis module, according to the business abnormality coefficient and the customer abnormality coefficient, the security monitoring level is determined, and the monitoring image quality information, the number of face recognition and the recognition abnormality threshold are obtained according to the security monitoring level mapping;

[0038] Specifically, through the monitoring requirement analysis module, the decision to obtain the security monitoring level is made based on the business abnormality coefficient and the customer abnormality coefficient. The business abnormality coefficient and the customer abnormality coefficient are first used as inputs, and a decision tree or classification algorithm (such as Logistic regression, support vector machine, etc.) is used for training to determine the security monitoring level. For each security monitoring level, it is mapped to the corresponding monitoring image quality information, face recognition number and recognition abnormality threshold. For image quality information, the corresponding monitoring image quality requirements can be defined, such as resolution, clarity, color, etc. The parameters of the monitoring image are obtained or adjusted to ensure that the image quality meets the requirements. For the number of face recognition, the corresponding number of face recognition can be defined. Based on this number, face detection and recognition are performed in the monitoring image, and face features are extracted and analyzed. For the abnormal recognition threshold, the corresponding recognition abnormality threshold can be defined. This threshold can be one or more indicators, such as face similarity, feature difference, etc. Based on this threshold, it is judged whether the face recognition result is normal or abnormal.

[0039] Through the security monitoring module, according to the monitoring image quality information, the collection of high-quality monitoring images in temporary bank branches is controlled, and according to the face recognition quantity, faces in the high-quality monitoring images are searched and recognized to obtain the abnormality level, and faces with a value greater than the recognition abnormality threshold are marked as monitoring results.

[0040] Specifically, according to the monitoring image quality information, the collection of high-quality monitoring images in temporary bank branches is controlled. The parameters of the camera can be adjusted to ensure that the collected images meet the requirements. High quality is generally an abnormal situation. Among them, the higher the security monitoring level, the more abnormal the current branch is, and riots or accidents may occur. It is necessary to collect higher quality (such as higher resolution) images, perform more face search, recognition and recording, and the lower the recognition abnormality threshold is, in order to maintain subsequent order. When the security monitoring level is general, there is no need to collect high-quality images, the number of face recognition is low, and the recognition abnormality threshold is high, saving computing power. For each collected monitoring image, face search and recognition algorithms are used to detect and recognize faces. Feature extraction and matching technology is to extract feature vectors from face images to distinguish different individuals. These feature vectors can include global features such as color, shape, texture, and local features such as corners of the eyes and mouth. The method that can be used is LDA based on PCA for feature extraction. According to the number of face recognition and the recognition abnormality threshold, the abnormality level of each face is calculated. The threshold is set as the normal face without other expressions, and the face can be divided into different abnormal levels. For all faces whose abnormality level is greater than the recognition abnormality threshold, those with expressions exceeding the threshold, such as panic or panic, are marked and used as monitoring results. This can effectively monitor abnormal people.

[0041] Furthermore, if Figure 2 As shown, the present application method also includes:

[0042] Obtain sample surveillance image records based on surveillance data records of temporary bank branches;

[0043] Identify and mark the number of customers in the sample surveillance image records to obtain the sample customer number records;

[0044] Specifically, according to the monitoring data records of temporary bank branches, a certain number of monitoring image records are obtained from the monitoring system of temporary bank branches. These records are monitoring images obtained in chronological order, by region, by specific business and other standards. The foreground target and background in the image are separated by image segmentation technology, and then the target detection algorithm is used to detect the customer target in the image. The target detection algorithm divides the image into regions with different gray levels by setting different thresholds, and then extracts the customer target, and performs feature extraction and recognition on the customer target. According to the processed image data, the number of customers in each monitoring image is counted using computer vision technology. It is necessary to use edge detection, shape recognition, and object tracking image processing technology to accurately identify each customer, and a sample customer quantity record containing customer quantity information will be obtained.

[0045] Based on the convolutional neural network, a customer quantity recognition channel is constructed, and the customer quantity recognition channel is trained using the sample monitoring image records and the sample customer quantity records;

[0046] The customer quantity recognition channel is used to recognize multiple surveillance images in the surveillance image sequence to obtain a customer quantity sequence.

[0047] Specifically, based on the convolutional neural network, a customer quantity recognition channel is constructed, and the sample surveillance image records and sample customer quantity records are sorted into a format suitable for CNN input. The image data is converted into a numerical array, and the customer quantity information is marked as the corresponding category label. The CNN model is used as the basic model of the customer quantity recognition channel. One or more convolutional layers, pooling layers, and fully connected layers are used to construct the model. The neurons output by the model correspond to a category of customer quantity. The sample surveillance image records and sample customer quantity records are used as training data to train the CNN model. During the training process, an optimization algorithm (such as stochastic gradient descent) is used to minimize the difference between the model output and the actual number of customers. The trained CNN model is deployed to the monitoring system of the bank branch to identify the number of customers in the surveillance image in real time. The trained customer quantity recognition channel is used to identify multiple surveillance images in the image sequence, and the image size and pixel value of each input image are adjusted. The processed image is input into the customer quantity recognition channel to obtain the customer quantity sequence, and the final customer quantity sequence is obtained after the statistical average of these values ​​and the removal of outliers, thereby improving the accuracy and reliability of the results.

[0048] Furthermore, the present application method also includes:

[0049] According to the monitoring time window of the monitoring image sequence, the monitoring data records of the temporary bank outlets are retrieved, and the historical average customer flow and the historical average customer flow change rate are calculated;

[0050] According to the customer quantity sequence, calculate and obtain the average customer flow and the average customer flow change rate;

[0051] Specifically, according to the monitoring time window of the monitoring image sequence, the monitoring data records of the temporary bank branch are retrieved. In order to calculate the historical average customer flow and the historical average customer flow change rate, a suitable monitoring time window is first determined. The length of this time window should be determined according to specific needs and the availability of data records. The time window can be set freely, 10 minutes during the week and 50 minutes on weekends. According to the determined monitoring time window, the corresponding monitoring image sequence is retrieved from the monitoring system of the temporary bank branch. This sequence includes all monitoring image data within the specified time window. The trained customer number recognition channel is used to identify the number of customers in the image. According to the output of the model, the customer number information in each monitoring image is extracted to calculate the average customer flow in each time window. The average number of customers is obtained by adding the number of customers in all monitoring images and then dividing it by the number of monitoring images in the time window. The average customer flow data in adjacent time windows can be used for calculation. First, the difference between the average customer flow in two adjacent time windows is calculated, and then the flow change rate is obtained by dividing it by the time interval between the time windows. Then the historical average customer flow change rate and the average customer flow are obtained.

[0052] Calculate the average customer flow and the historical average customer flow, as well as the deviation between the average customer flow change rate and the historical average customer flow change rate, respectively, to obtain a flow anomaly coefficient and a change rate anomaly coefficient;

[0053] The traffic anomaly coefficient and the change rate anomaly coefficient are weighted to obtain the customer anomaly coefficient.

[0054] Specifically, the average customer flow and the historical average customer flow are calculated respectively, the image information of the same time period of all historical times is selected, the number of customers is identified through the customer number recognition channel, the average number of customers at this time is calculated, the average number of customers is obtained by adding the number of customers in all windows and dividing it by the number of customers in all historical windows, as the historical average number of customers, the difference in the number of customers in adjacent windows is calculated, divided by the time difference of the windows, and the historical average customer flow change rate is obtained. By comparing the numerical values ​​of the historical and real-time average customer flow and change rate, the two can be subtracted and the absolute value is taken to obtain a deviation value. This deviation value is the abnormal coefficient of the change rate and the average number. The flow abnormal coefficient and the change rate coefficient are weighted and calculated to obtain the customer abnormal coefficient, so as to evaluate and analyze the customer flow and customer flow change of bank branches.

[0055] Furthermore, if Figure 3 As shown, the present application method also includes:

[0056] Clustering multiple service types in the service type sequence to obtain multiple type clustering results;

[0057] Clustering multiple business amounts in the business amount sequence using a K-means algorithm to obtain multiple amount clusters;

[0058] Specifically, firstly, the business type sequence and business amount sequence are cleaned and processed, invalid data is removed, missing data is filled, and outlier data is standardized. Then, representative features are extracted from the business type sequence and business amount sequence, such as calculating the number and amount proportion of each business type, or calculating the change trend of the business amount, etc. Then, cluster analysis is performed, and the extracted features are used as input. Cluster analysis is performed on the data using clustering algorithms (such as K-means, DBSCAN, etc.), and similar business types or business amounts are clustered into one category. One category is an amount cluster, and multiple amount clusters are obtained according to different categories. Transactions in the same amount cluster may have similar transaction amounts, while transactions in different amount clusters may have different transaction amounts.

[0059] According to the time window of the business type sequence and the business amount sequence, retrieve the business data records of the temporary bank outlets to obtain multiple historical average business quantities and multiple historical average amounts;

[0060] Specifically, according to the time window of the business type sequence and the business amount sequence, the business data records of the temporary bank outlets are retrieved. The time window of the business type and the time window of the business amount sequence represent the business handled and the amount used for the related business handling within a period of time. The business data records include deposits and withdrawals, remittances and transfers, credit and debit card businesses, loan businesses, wealth management businesses and other types. The amount of the business implemented in the business handling record, the business type sequence and the business amount sequence can be arranged according to time, which means that the bank handled 100 businesses within a period of time, and the total amount of these businesses was 200,000 yuan, of which the deposit business amount was 30,000 yuan, the withdrawal business amount was 20,000 yuan, the remittance business amount was 10,000 yuan, the transfer business amount was 10,000 yuan, the loan business amount was 100,000 yuan, and the wealth management business amount was 30,000 yuan. The historical average business quantity is obtained by adding up all the historical business quantities and the corresponding amounts involved at this time, and dividing by the total number of windows or the number of time to obtain the average daily transaction business quantity and amount.

[0061] Calculating the deviations between the number of service types in the multiple type clustering results and the multiple historical average service numbers respectively, and performing weighted calculation to obtain a type service anomaly coefficient;

[0062] Calculate the deviations between the amounts in multiple amount clusters and multiple historical average amounts respectively, and perform weighted calculation to obtain the amount business abnormality coefficient;

[0063] The business anomaly coefficient is calculated by combining the business anomaly coefficient of type and the business anomaly coefficient of amount.

[0064] Specifically, for each type of clustering result, calculate the deviation between the number of business types and the historical average number of business. By comparing the numerical values ​​of the historical and implemented business and amount, the two can be subtracted and then the absolute value is taken to obtain a deviation value. To obtain the type business anomaly coefficient. The deviation of each clustering result is weighted averaged (for example, based on the proportion of the number of business types in each cluster) to obtain the type business anomaly coefficient and the amount business anomaly coefficient. Combine the type business anomaly coefficient and the amount business anomaly coefficient to calculate a comprehensive business anomaly coefficient through weighted average or simple average. This business anomaly coefficient can reflect the degree of abnormality of the overall business.

[0065] Furthermore, the present application method also includes:

[0066] According to the monitoring data records of temporary bank branches, process and obtain the abnormal coefficient records of sample customers and sample business abnormal coefficient records;

[0067] According to the sample customer abnormality coefficient record and the sample business abnormality coefficient record, the sample security monitoring level record is evaluated and obtained, and the size of the sample security monitoring level is positively correlated with the size of the sample customer abnormality coefficient and the sample business abnormality coefficient;

[0068] Specifically, the customer information in each surveillance image is extracted from the surveillance data records, including the customer's facial features, behavioral actions and other information. Using face recognition technology and behavioral analysis technology, abnormality detection is performed on the customer in each surveillance image. For example, if the customer's actions or expressions do not match the normal behavior pattern, they can be regarded as abnormal customers. For each abnormal customer in the surveillance image, their facial features, behavioral actions and other information can be recorded, and an abnormality coefficient can be calculated. According to the sample customer abnormality coefficient record and the sample business abnormality coefficient record, the sample security monitoring level record is evaluated and obtained. This coefficient can reflect the degree of abnormality of the customer. The size of the sample security monitoring level is positively correlated with the size of the sample customer abnormality coefficient and the sample business abnormality coefficient. For example, the larger the coefficient value, the higher the degree of abnormality of the customer.

[0069] Using the sample customer abnormal coefficient record and the sample business abnormal coefficient record as decision inputs, using the sample security monitoring level record as a decision result, and building a monitoring requirement decision classifier based on a decision tree;

[0070] A monitoring requirement decision classifier is used to perform decision classification on the business anomaly coefficient and the customer anomaly coefficient to obtain the security monitoring level.

[0071] Specifically, the sample customer abnormal coefficient record and the sample business abnormal coefficient record are used as decision inputs, and the sample customer abnormal coefficient record, the sample business abnormal coefficient record and the sample security monitoring level record are used as input data. The data is first preprocessed, including data cleaning, standardization, and normalization operations, to determine the features to be used, namely the business abnormal coefficient and the customer abnormal coefficient. Using the prepared data and the determined features, a decision tree algorithm is used to construct a monitoring requirement decision classifier. CART, ID3, etc. can be used for construction. The constructed decision tree is trained using a training data set, and the parameters are adjusted so that it can better classify new data. The trained decision tree is verified using a verification data set to evaluate its performance, such as accuracy and recall rate. The trained classifier is used to classify the new business abnormal coefficient and customer abnormal coefficient for decision-making to obtain the security monitoring level. The new data can be input into the classifier to obtain the output result, namely the security monitoring level.

[0072] Furthermore, the present application method also includes:

[0073] Obtain sample high-quality surveillance image records based on the high-quality surveillance image records of temporary bank branches;

[0074] Identify and mark the face of each sample high-quality surveillance image in the sample high-quality surveillance image record to obtain a face recognition result record;

[0075] Specifically, based on the high-quality surveillance image records of the temporary bank branch, high-definition, high-quality surveillance image records are obtained from the monitoring system of the temporary bank branch to obtain sample high-quality surveillance image records. The sample high-quality surveillance image records are historical image data. For each image in the sample high-quality surveillance image records, a face detection algorithm (such as Haar cascade or deep learning algorithm) is used to detect the face in the image. For each detected face, it is aligned and standardized. According to the results of face recognition, each face is matched with a specific person or facial feature, and these matching results are recorded. The data can be saved in a database to obtain a record of face recognition results.

[0076] Using semantic segmentation, we build a face recognition channel, and use sample high-quality surveillance image records and face recognition result records to train until convergence;

[0077] Using a face recognition channel, identifying and segmenting the face image in the high-quality surveillance image to obtain a face recognition result, wherein the face recognition result includes a plurality of face images;

[0078] Specifically, semantic segmentation technology is used to divide the image into multiple objects or regions, and each pixel is marked as belonging to the corresponding object or region, and a deep learning model (such as U-Net, Mask R-CNN, etc.) is built. This model will be used to recognize and segment faces, and the model is trained using sample high-quality surveillance image records and face recognition result records. The training process will be iterated until the model performance converges. When the model training is completed, it can be used to recognize and segment faces in high-quality surveillance images. For each high-quality surveillance image, the model will output a segmented image containing all the faces in the image. After obtaining the face recognition results, some post-processing steps are performed, such as filtering out faces that are too small or too blurry, or comparing the segmented faces with known faces in the database.

[0079] According to the number of face recognitions, the face images in the face recognition results are searched and recognized, and the face anomaly levels are obtained through analysis. The face images with a level greater than the recognition anomaly threshold are marked as monitoring results.

[0080] Specifically, an abnormal threshold is set. The abnormal threshold may show other features on the face but not obvious features. For each pixel, the corresponding facial features are extracted, and the extracted feature vector is matched with the features of the target face. The face image with an abnormal level greater than the abnormal threshold is found and marked. For all faces with abnormal levels greater than the recognition abnormal threshold, faces with expressions exceeding the threshold, such as panic and panic, are marked and used as monitoring results. Abnormal people can be effectively monitored.

[0081] Furthermore, the present application method also includes:

[0082] Based on the high-quality surveillance image records of temporary bank branches, sample facial image records are obtained, and abnormal expression analysis and marking are performed to obtain abnormal level records of sample facial expressions;

[0083] Specifically, based on the high-quality surveillance image records of temporary bank outlets, sample facial image records are obtained, and expression abnormality analysis and marking are performed. High-quality facial images containing normal and abnormal expressions are selected from the surveillance data of temporary bank outlets as sample facial image records. Expression analysis is performed on each image in the sample facial image records, and which images contain abnormal expressions are marked to obtain sample facial abnormality level records.

[0084] Using sample face image records and sample face anomaly level records, a face anomaly identifier is constructed and trained based on a convolutional neural network.

[0085] Specifically, according to the labeling results, each image containing abnormal expressions is assigned an abnormality level, which can reflect the severity of the abnormal expression. Using the sample face image records and the sample face abnormality level records, a convolutional neural network (CNN) is used to build and train a face abnormality recognizer. For abnormal face images, they need to be labeled and classified, for example, according to different abnormal types (such as anger, fear, joy, etc.). Preprocess the data, such as resizing the image, normalizing the pixel values, etc. Use a deep learning framework (such as TensorFlow or PyTorch) to build a face abnormality recognizer. Use a convolutional neural network (CNN) to divide the prepared data set into a training set and a validation set. Use the training set to train the model, and use the validation set to evaluate the performance of the model. This recognizer will be able to distinguish between normal and abnormal expressions.

[0086] According to the number of face recognitions, a plurality of face images are randomly searched in the face recognition results according to a plurality of initial directions;

[0087] Specifically, in each search process, when a face image is found, it is added to a set. This set will contain all face images searched in multiple initial directions. If a face image is not found after searching for a long time in a certain direction, the search space is adjusted or the search direction is changed. This can be done by randomly selecting a new initial position or changing the search direction to search again.

[0088] A face anomaly identifier is used to identify multiple face images, obtain multiple face anomaly levels, and determine whether they are greater than the recognition anomaly threshold. If so, mark them and continue searching in the initial direction. If not, do not mark them and adjust the initial direction to obtain the adjusted direction and search;

[0089] Specifically, multiple randomly searched face images are input into the face anomaly recognizer to obtain multiple face anomaly levels. The multiple face anomaly levels are compared with the preset recognition anomaly threshold. If the face anomaly level of any person is greater than the threshold, the face image is marked and the search continues in the original initial direction. If all face anomaly levels are less than the threshold, the initial direction is adjusted, a new adjusted direction is obtained, and then the search continues in this new direction.

[0090] According to the inverse of the ratio of the multiple face abnormality levels to the recognition abnormality threshold, the preset search step length is adjusted to obtain the adjusted search step length;

[0091] According to the adjusted search step length and face recognition number, continue searching and identifying in the face recognition results until the preset search times are reached.

[0092] Specifically, for each of the multiple face abnormality levels, the ratio thereof to the recognition abnormality threshold is calculated. This ratio can reflect the gap between the severity of each abnormality level and the threshold. According to the calculated ratio, the preset search step is adjusted. The search step can be scaled according to the inverse of the ratio, that is, if the ratio is small (the abnormality level is more serious), the search step is reduced to perform a more detailed search in the nearby search space, and if the ratio is large (the abnormality level is lighter), the search step is increased to expand the search range and find a more distant abnormality level. According to the adjusted search step and the number of face recognitions, the search and recognition in the face recognition results are continued. The random search method or other search strategies described above can be used to search in the adjusted search space until the preset number of searches is reached. During the search process, it is checked whether the preset number of searches has been reached. If the preset number of searches has been reached, the search process is terminated, otherwise, the search is continued. After the search is terminated, the multiple face images searched and their corresponding abnormality levels are recorded. These results can be used for subsequent analysis and processing.

[0093] Embodiment 2

[0094] Based on the same inventive concept as the temporary bank branch security monitoring method in the aforementioned embodiment, Figure 4 As shown, the present application provides a temporary bank branch security monitoring system, the system comprising:

[0095] A monitoring image sequence acquisition module 10, wherein the monitoring image series acquisition module 10 acquires monitoring images of temporary bank outlets to be monitored through an outlet information acquisition module to acquire a monitoring image sequence;

[0096] A customer abnormality coefficient calculation module 20, wherein the customer abnormality coefficient calculation module 20 identifies the number of customers in the monitoring image according to the monitoring image sequence, obtains a customer number sequence, performs customer number abnormality analysis and customer flow abnormality analysis, and calculates and obtains a customer abnormality coefficient;

[0097] A business sequence acquisition module 30, which is used to collect business types and business amounts in temporary bank outlets, and acquire business type sequences and business amount sequences;

[0098] A business anomaly coefficient acquisition module 40, the business anomaly coefficient acquisition module 40 is used to perform cluster anomaly analysis according to the business type sequence and the business amount sequence, and calculate and obtain the business anomaly coefficient;

[0099] A monitoring security level acquisition module 50, the monitoring security level acquisition module 50 is used to make a decision to acquire a security monitoring level according to the business abnormality coefficient and the customer abnormality coefficient through the monitoring requirement analysis module, and acquire monitoring image quality information, face recognition quantity and recognition abnormality threshold according to the security monitoring level mapping;

[0100] The monitoring result acquisition module 60 is used to control the collection of high-quality monitoring images in temporary bank branches through the security monitoring module according to the monitoring image quality information, and search and identify faces in the high-quality monitoring images according to the face recognition quantity, obtain the abnormality level, and mark the faces whose recognition level is greater than the recognition abnormality threshold as the monitoring result.

[0101] Furthermore, the system also includes:

[0102] The sample record acquisition module acquires sample monitoring image records based on the monitoring data records of temporary bank outlets;

[0103] The customer quantity record acquisition module identifies and marks the customer quantity in the sample monitoring image record and acquires the sample customer quantity record;

[0104] A channel training module, based on a convolutional neural network, constructs a customer quantity recognition channel, and uses the sample monitoring image records and sample customer quantity records to train the customer quantity recognition channel;

[0105] The quantity sequence acquisition module uses the customer quantity recognition channel to recognize multiple monitoring images in the monitoring image sequence to obtain a customer quantity sequence.

[0106] Furthermore, the system also includes:

[0107] A historical average flow and change rate calculation module, which retrieves monitoring data records of temporary bank outlets according to the monitoring time window of the monitoring image sequence, and calculates and obtains the historical average customer flow and the historical average customer flow change rate;

[0108] A real-time average flow and change calculation module, which calculates and obtains the average customer flow and the average customer flow change rate according to the customer quantity sequence;

[0109] The abnormal coefficient acquisition module calculates the average customer flow and the historical average customer flow, as well as the deviation between the average customer flow change rate and the historical average customer flow change rate, to obtain the flow abnormal coefficient and the change rate abnormal coefficient;

[0110] The customer anomaly coefficient acquisition module performs weighted calculation on the flow anomaly coefficient and the change rate anomaly coefficient to obtain the customer anomaly coefficient.

[0111] Furthermore, the system also includes:

[0112] A clustering result acquisition module clusters multiple business types in the business type sequence to obtain multiple type clustering results;

[0113] An amount cluster acquisition module, which clusters multiple business amounts in the business amount sequence using a K-means algorithm to obtain multiple amount clusters;

[0114] The average amount acquisition module retrieves the business data records of temporary bank outlets according to the time window of the business type sequence and the business amount sequence, and obtains multiple historical average business quantities and multiple historical average amounts;

[0115] An abnormal coefficient acquisition module calculates the deviations between the number of business types in the multiple type clustering results and the multiple historical average business numbers, and performs weighted calculation to obtain the type business abnormality coefficient;

[0116] The amount abnormality coefficient acquisition module calculates the deviations between the amounts in multiple amount clusters and multiple historical average amounts, and performs weighted calculation to obtain the amount business abnormality coefficient;

[0117] The business anomaly coefficient acquisition module calculates and obtains the business anomaly coefficient by combining the type business anomaly coefficient and the amount business anomaly coefficient.

[0118] Furthermore, the system also includes:

[0119] The coefficient record acquisition module processes and acquires sample customer abnormal coefficient records and sample business abnormal coefficient records based on the monitoring data records of temporary bank outlets;

[0120] A sample evaluation module evaluates and obtains a sample safety monitoring level record according to the sample customer abnormality coefficient record and the sample business abnormality coefficient record, wherein the size of the sample safety monitoring level is positively correlated with the size of the sample customer abnormality coefficient and the sample business abnormality coefficient;

[0121] A decision classifier acquisition module uses the sample customer abnormal coefficient record and the sample business abnormal coefficient record as decision inputs, uses the sample safety monitoring level record as a decision result, and constructs a monitoring requirement decision classifier based on a decision tree;

[0122] The security monitoring level acquisition module adopts a monitoring requirement decision classifier to make decision classification on the business abnormality coefficient and the customer abnormality coefficient to obtain the security monitoring level.

[0123] Furthermore, the system also includes:

[0124] A sample record acquisition module, which acquires sample high-quality surveillance image records based on the high-quality surveillance image records of temporary bank branches;

[0125] The face recognition result acquisition module recognizes and marks the face of each sample high-quality surveillance image in the sample high-quality surveillance image record to obtain the face recognition result record;

[0126] The face recognition channel construction module uses semantic segmentation to build the face recognition channel, and uses sample high-quality surveillance image records and face recognition result records for training until convergence;

[0127] A face recognition module, using a face recognition channel, recognizes and segments the face image in the high-quality surveillance image to obtain a face recognition result, wherein the face recognition result includes a plurality of face images;

[0128] The abnormality level judgment module searches and identifies the face images in the face recognition results according to the face recognition quantity, analyzes and obtains the face abnormality level, and marks the face images that are greater than the recognition abnormality threshold as the monitoring result.

[0129] Furthermore, the system also includes:

[0130] The abnormal level recording module obtains sample face image records based on the high-quality surveillance image records of temporary bank branches, analyzes and marks abnormal expressions, and obtains sample face abnormal level records;

[0131] The anomaly detector acquisition module uses sample face image records and sample face anomaly level records to construct and train a face anomaly detector based on a convolutional neural network;

[0132] The face image acquisition module randomly searches and obtains multiple face images in the face recognition results according to the number of face recognitions and multiple initial directions;

[0133] The direction adjustment module uses a face anomaly identifier to identify multiple face images, obtain multiple face anomaly levels, and determine whether they are greater than the recognition anomaly threshold. If so, it is marked and the search continues in the initial direction. If not, it is not marked and the initial direction is adjusted to obtain the adjusted direction and search;

[0134] A step length adjustment module adjusts the preset search step length according to the inverse of the ratio of the multiple face abnormality levels to the recognition abnormality threshold to obtain an adjusted search step length;

[0135] The number judgment module continues to search and identify the face recognition results according to the adjusted search step size and the number of face recognitions until the preset number of searches is reached.

[0136] Through the detailed description of the aforementioned temporary bank branch security monitoring method, the technical personnel in this field can clearly understand the temporary bank branch security monitoring system in this embodiment. For the system disclosed in the embodiment, since it corresponds to the device disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0137] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A temporary bank branch security monitoring method, characterized in that: The method is applied to a temporary bank branch security monitoring device, the device includes a branch information collection module, a monitoring requirement analysis module and a security monitoring module, and the method includes: Through the branch information collection module, the monitoring images of the temporary bank branches to be monitored are collected to obtain a monitoring image sequence; According to the monitoring image sequence, identify the number of customers in the monitoring image, obtain the customer number sequence, perform customer number anomaly analysis and customer flow anomaly analysis, and calculate and obtain the customer anomaly coefficient; Collect business types and business amounts in temporary bank outlets, and obtain business type sequences and business amount sequences; Perform clustering anomaly analysis based on the business type sequence and business amount sequence, and calculate the business anomaly coefficient; Through the monitoring requirement analysis module, according to the business abnormality coefficient and the customer abnormality coefficient, the security monitoring level is determined, and the monitoring image quality information, the number of face recognition and the recognition abnormality threshold are obtained according to the security monitoring level mapping; Through the security monitoring module, according to the monitoring image quality information, the collection of high-quality monitoring images in temporary bank branches is controlled, and according to the face recognition quantity, faces in the high-quality monitoring images are searched and recognized to obtain the abnormality level, and faces with a value greater than the recognition abnormality threshold are marked as monitoring results.

2. The method according to claim 1, characterized in that According to the surveillance image sequence, identifying the number of customers in the surveillance image and obtaining a customer number sequence includes: Obtain sample surveillance image records based on surveillance data records of temporary bank branches; Identify and mark the number of customers in the sample surveillance image records to obtain the sample customer number records; Based on the convolutional neural network, a customer quantity recognition channel is constructed, and the customer quantity recognition channel is trained using the sample monitoring image records and the sample customer quantity records; The customer quantity recognition channel is used to recognize multiple surveillance images in the surveillance image sequence to obtain a customer quantity sequence.

3. The method according to claim 1, characterized in that: Conduct customer quantity and customer flow anomaly analysis and calculate customer anomaly coefficients, including: According to the monitoring time window of the monitoring image sequence, the monitoring data records of the temporary bank outlets are retrieved, and the historical average customer flow and the historical average customer flow change rate are calculated; According to the customer quantity sequence, calculate and obtain the average customer flow and the average customer flow change rate; Calculate the average customer flow and the historical average customer flow, as well as the deviation between the average customer flow change rate and the historical average customer flow change rate, respectively, to obtain a flow anomaly coefficient and a change rate anomaly coefficient; The traffic anomaly coefficient and the change rate anomaly coefficient are weighted to obtain the customer anomaly coefficient.

4. The method according to claim 1, characterized in that: According to the business type sequence and business amount sequence, clustering anomaly analysis is performed, and the business anomaly coefficient is calculated, including: Clustering multiple service types in the service type sequence to obtain multiple type clustering results; Clustering multiple business amounts in the business amount sequence using a K-means algorithm to obtain multiple amount clusters; According to the time window of the business type sequence and the business amount sequence, retrieve the business data records of the temporary bank outlets to obtain multiple historical average business quantities and multiple historical average amounts; Calculating the deviations between the number of service types in the multiple type clustering results and the multiple historical average service numbers respectively, and performing weighted calculation to obtain a type service anomaly coefficient; Calculate the deviations between the amounts in multiple amount clusters and multiple historical average amounts respectively, and perform weighted calculation to obtain the amount business abnormality coefficient; The business anomaly coefficient is calculated by combining the business anomaly coefficient of type and the business anomaly coefficient of amount.

5. The method according to claim 1, characterized in that According to the business abnormality coefficient and the customer abnormality coefficient, a decision is made to obtain a security monitoring level, including: According to the monitoring data records of temporary bank branches, process and obtain the abnormal coefficient records of sample customers and sample business abnormal coefficient records; According to the sample customer abnormality coefficient record and the sample business abnormality coefficient record, the sample security monitoring level record is evaluated and obtained, and the size of the sample security monitoring level is positively correlated with the size of the sample customer abnormality coefficient and the sample business abnormality coefficient; Using the sample customer abnormal coefficient record and the sample business abnormal coefficient record as decision inputs, using the sample security monitoring level record as a decision result, and building a monitoring requirement decision classifier based on a decision tree; A monitoring requirement decision classifier is used to perform decision classification on the business anomaly coefficient and the customer anomaly coefficient to obtain the security monitoring level.

6. The method according to claim 1, characterized in that According to the monitoring image quality information, the collection of high-quality monitoring images in the temporary bank branch is controlled, and according to the face recognition quantity, the faces in the high-quality monitoring images are searched and recognized, the abnormality level is obtained, and the faces with a level greater than the recognition abnormality threshold are marked as monitoring results, including: Obtain sample high-quality surveillance image records based on the high-quality surveillance image records of temporary bank branches; Identify and mark the face of each sample high-quality surveillance image in the sample high-quality surveillance image record to obtain a face recognition result record; Using semantic segmentation, we build a face recognition channel, and use sample high-quality surveillance image records and face recognition result records to train until convergence; Using a face recognition channel, identifying and segmenting the face image in the high-quality surveillance image to obtain a face recognition result, wherein the face recognition result includes a plurality of face images; According to the number of face recognitions, the face images in the face recognition results are searched and recognized, and the face anomaly levels are obtained through analysis. The face images with a level greater than the recognition anomaly threshold are marked as monitoring results.

7. The method according to claim 6, characterized in that According to the number of face recognitions, the face images in the face recognition results are searched and recognized, and the face abnormality level is obtained by analysis, and the face images with a level greater than the recognition abnormality threshold are marked, including: Based on the high-quality surveillance image records of temporary bank branches, sample facial image records are obtained, and abnormal expression analysis and marking are performed to obtain abnormal level records of sample facial expressions; Using sample face image records and sample face anomaly level records, a face anomaly identifier is constructed and trained based on a convolutional neural network. According to the number of face recognitions, a plurality of face images are randomly searched in the face recognition results according to a plurality of initial directions; A face anomaly identifier is used to identify multiple face images, obtain multiple face anomaly levels, and determine whether they are greater than the recognition anomaly threshold. If so, mark them and continue searching in the initial direction. If not, do not mark them and adjust the initial direction to obtain the adjusted direction and search; According to the inverse of the ratio of the multiple face abnormality levels to the recognition abnormality threshold, the preset search step length is adjusted to obtain the adjusted search step length; According to the adjusted search step length and face recognition number, continue searching and identifying in the face recognition results until the preset search times are reached.

8. A temporary bank branch security monitoring system, characterized in that: The system comprises: A monitoring image sequence acquisition module, wherein the monitoring image series acquisition module acquires monitoring images of temporary bank outlets to be monitored through an outlet information acquisition module to acquire a monitoring image sequence; A customer anomaly coefficient calculation module, wherein the customer anomaly coefficient calculation module identifies the number of customers in the monitoring image according to the monitoring image sequence, obtains a customer quantity sequence, performs customer quantity anomaly analysis and customer flow anomaly analysis, and calculates and obtains a customer anomaly coefficient; A business sequence acquisition module, which is used to collect business types and business amounts in temporary bank outlets, and acquire business type sequences and business amount sequences; A business anomaly coefficient acquisition module, which is used to perform clustering anomaly analysis according to the business type sequence and the business amount sequence, and calculate the business anomaly coefficient; A monitoring security level acquisition module, which is used to make a decision to acquire a security monitoring level according to the business abnormality coefficient and the customer abnormality coefficient through the monitoring requirement analysis module, and acquire monitoring image quality information, face recognition quantity and recognition abnormality threshold according to the security monitoring level mapping; The monitoring result acquisition module is used to control the collection of high-quality monitoring images in temporary bank branches through the security monitoring module according to the monitoring image quality information, and search and identify faces in the high-quality monitoring images according to the face recognition quantity, obtain the abnormality level, and mark the faces whose recognition level is greater than the recognition abnormality threshold as the monitoring result.

Images