Teller sign-in method, system and equipment based on signature track recognition and medium
By collecting and analyzing the dynamic characteristics of signature trajectory data in real time, and combining them with machine learning algorithms, the security and false recognition issues of existing teller check-in methods have been resolved, achieving efficient and secure identity authentication.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INSPUR FINANCIAL INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2025-11-28
- Publication Date
- 2026-04-10
AI Technical Summary
Existing teller sign-in methods pose risks of information leakage and loss. Biometric technology has a high false recognition rate and is affected by environmental factors, while static electronic signatures are difficult to prevent forgery.
By collecting signature trajectory data in real time, dynamic features such as stroke start and end time, average speed, trajectory curvature changes and pressure distribution features are extracted. Machine learning algorithms are then used to compare these features with historical features to determine the user's identity.
It improves the security and convenience of identity authentication, reduces the false recognition rate, prevents signature forgery, simplifies the check-in process, and improves operational efficiency.
Smart Images
Figure CN121834291A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of signature trajectory recognition, and in particular to a teller check-in method and system based on signature trajectory recognition, a device and a medium. BACKGROUND
[0002] With the development of financial technology, bank business gradually evolves towards intelligence and unmanned. As the first step of business, the safety and efficiency of teller check-in is crucial. The common teller check-in methods currently include: Password login: the teller inputs the username and password on the terminal device to complete login; IC card / magnetic stripe card check-in: using physical cards for identity authentication; fingerprint recognition check-in: verifying the identity of the teller through the fingerprint recognition module; face recognition check-in: recognizing through the camera to collect facial features.
[0003] In the above methods, the password and card methods have the risk of leakage and loss; although the biometric recognition (such as fingerprint and face) improves the security, it has certain misrecognition rate and inconvenience due to the limitation of hardware cost, environmental factors (such as wet fingers and face recognition affected by wearing masks), etc. In addition, some banks have tried to introduce electronic signature for business confirmation, but the electronic signature is in the form of static signature collection, which cannot effectively prevent the behavior of forging signature. SUMMARY
[0004] The present application provides a teller check-in method and system based on signature trajectory recognition, a device and a medium to solve the problems of the risk of leakage and loss of password and card methods, certain misrecognition rate and inconvenience of biometric recognition, and the electronic signature being in the form of static signature collection which cannot effectively prevent the behavior of forging signature.
[0005] In the first aspect, the present application provides a teller check-in method based on signature trajectory recognition, the method comprising: real-time collection of original trajectory data of the current operation user in the signature process through a preset signature area touch interface; preprocessing of all the collected original trajectory data to obtain preset key dynamic features; wherein the preset key dynamic features at least include: stroke start and end time, average speed, maximum speed, trajectory curvature change, pressure distribution feature, and stroke pattern; comparison of the extracted preset key dynamic features with the historical key dynamic features of the teller in the background database using a machine learning algorithm to obtain a matching degree; if the matching degree exceeds a set threshold, it is determined that the user is legal, and the check-in is allowed to be successful; otherwise, it is prompted to sign again or the check-in is rejected.
[0006] In an implementation of the present application, before the original trajectory data of the current operating user in the signature process is collected in real time through the preset signature area touch interface, the method further comprises: Displaying the preset signature area touch interface by triggering a preset key by the user.
[0007] In an implementation of the present application, the original trajectory data of the current operating user in the signature process is collected in real time through the preset signature area touch interface, specifically comprising: Collecting the timestamp, logical coordinates (X, Y), pressure value, and touch point area; and storing the single-frame data by using a structure array; During the collection, the data is collected by using an interrupt triggering mode. When the user's finger touches the touch screen, the touch screen driver generates an interrupt signal, the main control module responds to the interrupt and reads the current frame data, and each frame data is immediately stored in a ring buffer after being collected. At the same time, a timer is started, and if no new data is detected for 500 ms continuously, it is determined that the signature is completed. After the signature is completed, the buffer data is processed in JSON format to obtain the JSON format original data; and the teller ID and device number fields are added to obtain the original trajectory data. The original trajectory data includes the primary key, teller ID, device number, JSON format original data, and collection time.
[0008] In an implementation of the present application, the collected original trajectory data is preprocessed to obtain preset key dynamic features, specifically comprising: Abnormal data in the original trajectory data is removed by using the 3σ criterion; and the original trajectory data with the abnormal data removed is smoothed by using a sliding average filtering algorithm. The smoothed coordinate data is normalized to map the X and Y coordinates to the [0, 1] interval. The pressure data is normalized by using Z-Score standardization. The strokes are segmented based on the speed threshold method. The moving speed of the smoothed adjacent two frames of data is calculated. When the speed is less than the set threshold value for 3 frames continuously, it is determined that the stroke is completed, and the trajectory data is segmented into several independent strokes. According to the stroke segmentation result, the starting frame timestamp and ending frame timestamp of each stroke are recorded, and the stroke duration is calculated. Based on the normalized coordinate data and timestamp, the moving speed of each adjacent two frames is calculated. The average speed is the arithmetic average of all frame speeds, and the maximum speed is the maximum value in all frame speeds. The trajectory curvature is calculated by using the circular arc fitting algorithm. The coordinates of each continuous 5 frames are fitted by using a circular arc. The circular arc radius R is solved by using the least square method, and the curvature K = 1 / R. The average value of the absolute value of the curvature difference of the adjacent two fitted points is the curvature change rate. The signature area is divided into 16 grids of 4x4, and the proportion of the integral of the pressure value in each grid in the total integral is counted to form a 16-dimensional pressure distribution vector; wherein, the integral = the average of all pressure values in the current grid x the residence time, and the total integral is the cumulative value of all integrals; The segmented strokes are directionally encoded, the average direction of each stroke is mapped to 8 directions to form a stroke direction sequence; then the direction change relationship of adjacent strokes is calculated to construct a stroke order pattern feature vector; at the same time, the number of strokes and the length proportion of each stroke are recorded to form multi-dimensional stroke order features; wherein, the length proportion = single stroke length / total trajectory length.
[0009] In an implementation manner of the present application, the extracted preset key dynamic features are compared with the historical key dynamic features of the teller in the background database using a machine learning algorithm to obtain a matching degree, specifically including: When the preset key dynamic features of the current user are received, the historical key dynamic feature set of the teller is queried from the MySQL database according to the teller's job number; The current preset key dynamic features and each group of historical key dynamic features are respectively executed DTW alignment, and then the aligned vectors are input into the trained SVM model to obtain a classification confidence; According to the collection time of the historical key dynamic features, the weights of the historical key dynamic features are determined, the weights and the classification confidence are summed by weighting and then averaged to obtain the matching degree.
[0010] In an implementation manner of the present application, if the matching degree exceeds a set threshold, it is determined that the user is legal, and the sign-in is allowed to be successful; otherwise, the user is prompted to re-sign or the sign-in is rejected, and the method further includes: The original trajectory data of this time of signature, the preset key dynamic features, and the matching degree are uploaded to the background server to automatically generate a sign-in log.
[0011] In a second aspect, the present application provides a teller sign-in system based on signature trajectory recognition, which includes: The acquisition module is configured to acquire real-time original trajectory data of a current operation user in a signature process through a preset signature area touch interface; The obtaining module is configured to pre-process all the acquired original trajectory data to obtain preset key dynamic features; wherein, the preset key dynamic features at least include: stroke start and end time, average speed, maximum speed, trajectory curvature change, pressure distribution feature, and stroke order pattern; the extracted preset key dynamic features are compared with the historical key dynamic features of the teller in the background database using a machine learning algorithm to obtain a matching degree; The determination module is configured to determine that the user is a legal user and allow the sign-in to be successful if the matching degree exceeds a preset threshold, or prompt the user to re-sign or reject the sign-in.
[0012] In an implementation form of the application, the obtaining module comprises an obtaining unit, for when the preset key dynamic feature of the current user is received, querying the historical key dynamic feature set of the teller from the MySQL database according to the teller ID; performing DTW alignment on the current preset key dynamic feature and each group of historical key dynamic features respectively, and then inputting the aligned vectors into the trained SVM model to obtain a classification confidence; determining the weight of each historical key dynamic feature according to the collection time of the historical key dynamic feature, and obtaining a matching degree by weighted summation and then averaging the weights and the classification confidence.
[0013] In a third aspect, the application provides a teller sign-in device based on signature trajectory recognition, comprising: a processor; and a memory having executable code stored thereon, when the executable code is executed, causing the processor to perform any one of the above teller sign-in methods based on signature trajectory recognition.
[0014] In a fourth aspect, the application provides a non-volatile computer storage medium having computer instructions stored thereon, the computer instructions being executed to implement any one of the above teller sign-in methods based on signature trajectory recognition.
[0015] From the above technical solutions, the application has the following advantages: By collecting dynamic features in the signature process in real time, the security and convenience of identity authentication are improved. Compared with traditional passwords and cards, this method avoids the risk of information leakage and loss, because the signature trajectory is a unique behavior feature of the user and is difficult to be copied or forged. At the same time, compared with biometric recognition technology, this method is not affected by hardware costs or environmental factors (such as wet fingers and wearing masks), reduces the false recognition rate, and improves the user experience. In addition, by comparing dynamic features such as stroke start and end time, average speed, and trajectory curvature change, the system can more accurately identify the user's identity, reduce authentication failures caused by environmental or physiological changes, and thus improve the sign-in efficiency.
[0016] Another advantage is its dynamic signature collection mechanism, effectively solving the problem of easy forgery of static electronic signatures. By analyzing the preset key dynamic features through machine learning algorithms, the system can verify the authenticity of the signature behavior in real time, preventing forged signature behavior. For example, the combination of pressure distribution characteristics and stroke pattern makes the signature trajectory highly unique and difficult to imitate. In addition, the design of the user triggering the preset key further enhances security, ensuring that only the user who actively operates can enter the signature interface, avoiding unauthorized access. This dynamic, real-time authentication method not only improves the security of bank business, but also simplifies the teller sign-in process, reduces business interruptions due to authentication failures, and improves overall operational efficiency. BRIEF DESCRIPTION OF DRAWINGS
[0017] In order to more clearly illustrate the technical solutions of the present application, the drawings required to be used in the description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.
[0018] Figure 1 is a flow chart of a teller sign-in method based on signature trajectory recognition provided by an embodiment of the present application.
[0019] Figure 2 is a schematic diagram of the internal structure of a teller sign-in system based on signature trajectory recognition provided by an embodiment of the present application.
[0020] Figure 3 is a schematic diagram of the internal structure of a teller sign-in device based on signature trajectory recognition provided by an embodiment of the present application. DETAILED DESCRIPTION
[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0022] Those skilled in the art should understand that the embodiments described below are only preferred embodiments of the present disclosure, and do not represent that the present disclosure can only be implemented by this preferred embodiment. The preferred embodiment is only used to explain the technical principles of the present disclosure, and is not used to limit the protection scope of the present disclosure. Based on the preferred embodiments provided by the present disclosure, all other embodiments obtained by those skilled in the art without creative labor are still within the scope of protection of the present disclosure.
[0023] It is also to be noted that the terms "comprising", "comprises" or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a" does not, without further constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.
[0024] The technical solutions of the embodiments of the present application are described in detail below with reference to the drawings.
[0025] The embodiments provide a teller check-in method based on signature trajectory recognition, as shown in Figure 1 The method provided by the embodiments of the present application mainly includes the following steps: Step 110, acquiring, through a preset signature area touch interface, original trajectory data of a current operation user in a signature process in real time.
[0026] In some embodiments, before acquiring, through a preset signature area touch interface, original trajectory data of a current operation user in a signature process in real time, the method further includes: Displaying the preset signature area touch interface by triggering a preset key by the user.
[0027] Acquiring, through a preset signature area touch interface, original trajectory data of a current operation user in a signature process in real time, specifically includes: Acquiring a timestamp, a logical coordinate (X, Y), a pressure value, and a touch point area; and storing a single frame of data by using a structure array; During the acquisition, the data is acquired by using an interrupt triggering mode. When the user's finger touches the touch screen, the touch screen driver generates an interrupt signal, the main control module responds to the interrupt and reads the current frame of data, and each frame of data is immediately stored in a ring buffer area after being acquired. At the same time, a timer is started, and if no new data is detected for 500 ms continuously, it is determined that the signature is completed. After the signature is completed, the data in the buffer area is processed in a JSON format to obtain JSON format original data; a teller ID and a device number field are added to obtain original trajectory data; wherein the original trajectory data includes a primary key, a teller ID, a device number, JSON format original data, and acquisition time.
[0028] It needs to be added that in step 110, when the original trajectory data is collected in real time by the preset signature area touch interface, the touch screen has high-precision pressure sensing capability, the pressure value collection range is 0-1024 levels, and the change of the user's writing force can be accurately reflected. The area threshold of the touch point area is set to 5-50 mm². When the structure array stores a single frame of data, each data packet contains a timestamp (accuracy 1 ms), logical coordinates (resolution 0.1 mm), pressure value, and touch point area, forming a complete behavior feature record.
[0029] In the signature end determination mechanism, the 500 ms timeout threshold can be adjusted according to the actual application scenario. For busy bank outlets, it can be shortened to 300-400 ms to improve efficiency; for the elderly user group, it can be extended to 600-800 ms. The size of the ring buffer is set to 1.5 times the maximum estimated data volume of a single signature to ensure complete data storage.
[0030] In the JSON formatting processing link, the original data field naming uses general identifiers such as "timestamp", "x", "y", "pressure", and "area" to facilitate subsequent data analysis. The added teller ID field uses 8-digit numeric encoding, and the device number field uses a 6-digit encoding rule of "branch code + device serial number" to form a complete business trace chain.
[0031] Step 120, pre-processing all collected original trajectory data to obtain preset key dynamic features.
[0032] Among them, the preset key dynamic features at least include: stroke start and end time, average speed, maximum speed, trajectory curvature change, pressure distribution characteristics, and pen order mode.
[0033] The pre-processing of all collected original trajectory data to obtain the preset key dynamic features specifically includes: Remove abnormal data in the original trajectory data by 3σ criterion; use a sliding average filtering algorithm to smooth the original trajectory data of the abnormal data; Normalize the smoothed coordinate data, and map the X and Y coordinates to the [0, 1] interval; Standardize the pressure data, and use Z-Score standardization for normalization; Segment the strokes based on the speed threshold method, calculate the moving speed of the adjacent two frames of smoothed data, and when the speed is less than the set threshold for 3 consecutive frames, determine that the stroke is ended, and the trajectory data is divided into several independent strokes; According to the stroke segmentation result, record the start frame timestamp and end frame timestamp of each stroke, and calculate the stroke duration; Based on the standardized coordinate data and time stamp, the moving speed of each adjacent two frames is calculated, the average speed is the arithmetic mean of all frame speeds, and the maximum speed is the maximum value in all frame speeds; The trajectory curvature is calculated by using the circular arc fitting algorithm, the circular arc fitting is performed on each continuous 5 frames of coordinate points, the circular arc radius R is solved by using the least square method, and the curvature K = 1 / R; the average value of the absolute value of the curvature difference of adjacent two fitting points is the curvature change rate; The signature area is divided into 16 grids of 4x4, the proportion of the integral of the pressure value in each grid to the total integral is counted to form a 16-dimensional pressure distribution vector; wherein, the integral = the average value of all pressure values in the current grid x the residence time, and the total integral is the cumulative value of all integrals; The segmented strokes are directionally encoded, the average direction of each stroke is mapped to 8 directions to form a stroke direction sequence; then the direction change relationship of adjacent strokes is calculated to construct a stroke order mode feature vector; at the same time, the number of strokes and the length proportion of each stroke are recorded to form a multi-dimensional stroke order feature; wherein, the length proportion = single stroke length / total trajectory length.
[0034] It should be noted that in the preprocessing process of step 120, when the 3σ criterion removes abnormal data, the σ value is the average of the standard deviation of the last 100 signature trajectories. The sliding average filter adopts 7-frame window for data smoothing processing, and the window size is set according to the sampling rate of the touch screen. When the sampling rate is lower than 100Hz, a 5-frame window is used, and when the sampling rate is higher than 200Hz, a 10-frame window is used.
[0035] The coordinate normalization processing adopts the Min-Max normalization method to linearly map the X and Y coordinates to the [0, 1] interval, and retains the integer bit precision of the original coordinates. The pressure data standardization adopts the Z-Score standardization method, and the μ value and σ value are calculated independently according to the device. The pressure sensors of different models of touch screens use independent parameters.
[0036] The stroke segmentation speed threshold is set to 10mm / s, the business skilled teller uses a 12mm / s threshold, and the new employee uses a 6mm / s threshold. The adjacent stroke direction encoding adopts a 45° division method, which maps the average direction of each stroke to 8 directions to form a stroke direction sequence.
[0037] The pressure distribution grid adopts a 4x4 equal-area division method, and the integral calculation of each grid adopts an exponential decay function to give an early frame a weight coefficient of 0.8. In the construction of the stroke order mode feature vector, the relative angle difference Δθ = θn- θn-1 is used to encode the direction change relationship to form a 16-dimensional direction sensitive feature representation.
[0038] Step 130, using machine learning algorithm to compare the extracted preset key dynamic features with the historical key dynamic features of the teller in the background database, and obtain the matching degree.
[0039] This step can be specifically: When receiving the preset key dynamic features of the current user, the historical key dynamic feature set of the teller is queried from the MySQL database according to the teller's job number; The current preset key dynamic features and each group of historical key dynamic features are respectively executed DTW alignment, and then the aligned vectors are input into the trained SVM model to obtain the classification confidence; According to the collection time of the historical key dynamic features, the weight of each historical key dynamic feature is determined, and the weight and the classification confidence are summed by weighting and then averaged to obtain the matching degree.
[0040] It needs to be supplemented that in step 130, when using machine learning algorithm for feature comparison, the system queries the historical key dynamic feature set of the teller from the signature_features table of the MySQL database according to the teller's job number. The query result is arranged in descending order according to the collection time to ensure that the latest signature feature data is obtained. The current preset key dynamic features and each group of historical key dynamic features are respectively executed DTW dynamic time warping alignment. The DTW algorithm uses Euclidean distance as local distance measurement, and the window constraint is set to 10% of the sequence length. The regular path uses Itakura parallelogram constraint. The aligned feature vector dimension is unified to 128 dimensions, including stroke start and end time difference, average speed ratio, maximum speed ratio, trajectory curvature change rate difference, pressure distribution vector cosine similarity and stroke order mode edit distance.
[0041] The 128-dimensional feature vector after DTW alignment is input into the pre-trained SVM support vector machine model. The model uses RBF kernel function, the kernel parameter γ is set to 0.1, and the penalty factor C is set to 1.0. The SVM model outputs the classification confidence in the range of [0, 1]. According to the collection time of the historical key dynamic features, the weight is determined, and the exponential decay function is used. The feature weight of the last 7 days is 1.0, the feature weight of 7-30 days is 0.7, and the feature weight of more than 30 days is 0.3. The weight of each historical feature and the corresponding classification confidence are summed by weighting to calculate the total score, and then divided by the weight sum to obtain the final matching degree. The matching degree threshold is set to 0.85. If the matching degree exceeds the threshold, it is determined that the user is legal; otherwise, it is prompted to re-sign or refuse to sign in.
[0042] In addition, the machine learning algorithm selection and model training process can be specifically: Algorithm selection: Support vector machine (SVM) combined with dynamic time warping (DTW) hybrid algorithm is selected, in which DTW is used to solve the feature alignment problem caused by inconsistent signature trajectory length, and SVM is used to realize classification and judgment. DTW finds the optimal matching path of two feature sequences through dynamic programming, and the calculation formula is: DTW(i,j) = min(DTW(i-1,j), DTW(i,j-1), DTW(i-1,j-1)) + d(i,j), where d(i,j) is the Euclidean distance of two frames of features; SVM adopts RBF kernel function (radial basis function), and the kernel function parameter γ is optimized by grid search method (value range 10 - ³~10³), and the penalty coefficient C is set to 10 to improve the generalization ability of the model.
[0043] Model training process: Collect the historical signature data of each teller (at least 50 valid signature samples per person), extract the feature vector of each sample, and divide it into training set and test set according to the ratio of 7:3; use the training set to train the SVM model, adjust the parameters through 5-fold cross-validation, so that the accuracy of the model on the test set is more than 95%; store the trained model parameters (such as support vector, kernel function parameter) to the model file (.model format) and deploy it to the background server.
[0044] Step 140, if the matching degree exceeds the set threshold, it is determined that the user is legal, and the sign-in is successful; otherwise, prompt to re-sign or refuse to sign in.
[0045] In some embodiments, after the matching degree exceeds the set threshold, it is determined that the user is legal, and the sign-in is successful; otherwise, prompt to re-sign or refuse to sign in, the method further comprises: Upload the original trajectory data of this signature, the preset key dynamic feature, and the matching degree to the background server to automatically generate the sign-in log.
[0046] It needs to be supplemented that in step 140, when the matching degree exceeds the set threshold, the system executes the sign-in success process: displays the "sign-in success" prompt information through the teller terminal, and activates the operation permission of the business system; if the matching degree does not reach the threshold, display the "signature verification failed, please re-sign" prompt, and keep the system in a locked state. The system sets a three-level threshold mechanism: when the matching degree is greater than or equal to 0.85, it is directly authenticated; when the matching degree is in the range of 0.75-0.85, the user is allowed to choose to re-sign or use the backup verification method; when the matching degree is less than 0.75, the sign-in is directly refused and a security event is recorded. All authentication results are synchronized to the bank business system in real time to ensure the timely update of the operation permission.
[0047] After the determination is completed, the system uploads the complete data packet of this signature to the background server through the HTTPS protocol. The data packet contains the original trajectory data JSON file, the preset key dynamic feature vector, the matching degree value, the timestamp, the device identifier, and the session ID. After the server receives the data, it automatically generates a sign-in log record in the sign_log table of the Oracle database, and the log fields include: log ID, teller ID, sign-in time, matching degree, authentication result (success / failure), device number, and data check code. At the same time, the system starts a security warning mechanism for accounts that have failed verification for three consecutive times, marks the account status as "manual review required", and notifies the risk control department for subsequent processing.
[0048] In addition, the present application Figure 2 A teller sign-in system based on signature trajectory recognition is provided in the embodiments of the present application. As shown in Figure 2 The system provided by the embodiments of the present application mainly includes: The acquisition module 210 is configured to acquire, in real time, original trajectory data of a current operation user in a signature process through a preset signature area touch interface.
[0049] The obtaining module 220 is configured to pre-process all the acquired original trajectory data to obtain preset key dynamic features, wherein the preset key dynamic features at least include: stroke start and end time, average speed, maximum speed, trajectory curvature change, pressure distribution feature, and stroke mode; and use a machine learning algorithm to compare the extracted preset key dynamic features with historical key dynamic features of the teller in a background database to obtain a matching degree.
[0050] The obtaining module 220 includes an obtaining unit, configured to query the historical key dynamic feature set of the teller from a MySQL database according to the teller ID after receiving the preset key dynamic features of the current user; configured to perform DTW alignment on the current preset key dynamic features and each group of historical key dynamic features respectively, and then input the aligned vectors into a trained SVM model to obtain a classification confidence; configured to determine the weight of each historical key dynamic feature according to the acquisition time of the historical key dynamic feature, and obtain the matching degree by weighted summation and then averaging of the weights and the classification confidence.
[0051] The determination module 230 is configured to determine that the user is a legal user and allow successful sign-in if the matching degree exceeds a set threshold; otherwise, prompt to re-sign or reject sign-in.
[0052] The above is a method embodiment in the present application. Based on the same inventive concept, the embodiments of the present application also provide a teller sign-in device based on signature trajectory recognition. As shown in Figure 3As shown, the device comprises: a processor; and a memory having stored thereon executable code that, when executed, causes the processor to perform a teller sign-in method based on signature trajectory recognition as in one of the above embodiments.
[0053] Specifically, the server side acquires real-time original trajectory data of the current operation user in the signature process through a preset signature area touch interface; pre-processes all the acquired original trajectory data to obtain preset key dynamic features; wherein the preset key dynamic features at least include: stroke start and end time, average speed, maximum speed, trajectory curvature change, pressure distribution feature, and stroke mode; uses a machine learning algorithm to compare the extracted preset key dynamic features with the teller's historical key dynamic features in the background database to obtain a matching degree; if the matching degree exceeds a set threshold, it is determined that the user is legal, and the sign-in is allowed to be successful; otherwise, it prompts to re-sign or refuses to sign in.
[0054] In addition, the embodiments of the present application also provide a non-volatile computer storage medium having executable instructions stored thereon, which, when executed, implement a teller sign-in method based on signature trajectory recognition as described above.
[0055] The above description of disclosed embodiments enables those skilled in the art to carry out 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 can 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 these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A teller check-in method based on signature trajectory recognition, characterized in that, The method includes: The system uses a preset signature area touch interface to collect real-time raw trajectory data of the current user during the signature process. All the collected raw trajectory data are preprocessed to obtain preset key dynamic features; among them, the preset key dynamic features include at least: stroke start and end time, average speed, maximum speed, trajectory curvature change, pressure distribution features, and stroke order pattern. Machine learning algorithms are used to compare the extracted key dynamic features with the teller's historical key dynamic features in the backend database to obtain the matching degree. If the matching degree exceeds the set threshold, the user is deemed a legitimate user and is allowed to sign in successfully; otherwise, the user is prompted to re-sign or sign in is rejected.
2. The teller check-in method based on signature trajectory recognition according to claim 1, characterized in that, Before collecting the original trajectory data of the current user during the signing process in real time through the touch interface of the preset signature area, the method further includes: The user can trigger a preset button to display a preset signature area touch interface.
3. The teller check-in method based on signature trajectory recognition according to claim 1, characterized in that, Through a preset signature area touch interface, the system collects real-time raw trajectory data of the current user during the signature process, specifically including: Collect timestamps, logical coordinates (X, Y), pressure values, and touch point areas; store single-frame data using a structure array; Data is collected using an interrupt-triggered method. When a user touches the touchscreen, the touchscreen driver generates an interrupt signal. The main control module responds to the interrupt and reads the current frame data. After each frame of data is collected, it is immediately stored in a circular buffer. At the same time, a timer is started. If no new data is detected for 500ms, the signature process is considered complete. After signing, the buffer data is formatted into JSON format to obtain the raw JSON data; the counter employee number and device number fields are added to obtain the raw trajectory data; the raw trajectory data includes: primary key, counter employee number, device number, raw JSON data, and collection time.
4. The teller check-in method based on signature trajectory recognition according to claim 1, characterized in that, All collected raw trajectory data are preprocessed to obtain preset key dynamic features, specifically including: Abnormal data in the original trajectory data is removed using the 3σ criterion; the original trajectory data with abnormal data is then smoothly removed using a moving average filtering algorithm. The smoothed coordinate data is normalized to map the X and Y coordinates to the [0,1] interval; The pressure data was standardized using Z-Score normalization. The strokes are segmented based on the speed threshold method. The movement speed of two adjacent frames of data after smoothing is calculated. When the speed is less than the set threshold for three consecutive frames, the stroke is determined to be over and the trajectory data is segmented into several independent strokes. Based on the stroke segmentation results, record the start frame timestamp and end frame timestamp of each stroke, and calculate the stroke duration; Based on the standardized coordinate data and timestamps, the movement speed between each two adjacent frames is calculated. The average speed is the arithmetic mean of the speeds of all frames, and the maximum speed is the maximum value among all frame speeds. The trajectory curvature is calculated using a circular arc fitting algorithm. Circular arc fitting is performed on the coordinate points of every 5 consecutive frames, and the radius R of the arc is solved by the least squares method. The curvature K = 1 / R. The average value of the absolute value of the curvature difference between two adjacent fitting points is calculated as the rate of curvature change. The signature area is divided into 16 4×4 grids. The proportion of the pressure value integral in each grid to the total integral is calculated to form a 16-dimensional pressure distribution vector. The integral is the average pressure value of the current grid × the dwell time, and the total integral is the sum of all integrals. The segmented strokes are oriented and encoded, and the average direction of each stroke is mapped to 8 directions to form a stroke direction sequence. Then, the direction change relationship between adjacent strokes is calculated to construct a stroke order pattern feature vector. At the same time, the number of strokes and the length ratio of each stroke are recorded to form a multi-dimensional stroke order feature. Among them, the length ratio = single stroke length / total trajectory length.
5. The teller check-in method based on signature trajectory recognition according to claim 1, characterized in that, Machine learning algorithms are used to compare the extracted key dynamic features with the teller's historical key dynamic features in the backend database to obtain the matching degree, specifically including: Upon receiving the current user's preset key dynamic characteristics, the system queries the MySQL database for the teller's historical key dynamic characteristic set based on the teller's employee number. Perform DTW alignment on the current preset key dynamic features and each group of historical key dynamic features respectively, and then input the aligned vectors into the trained SVM model to obtain the classification confidence. Based on the collection time of historical key dynamic features, the weights of each historical key dynamic feature are determined. The weights and classification confidence scores are then weighted, summed, and averaged to obtain the matching degree.
6. The teller check-in method based on signature trajectory recognition according to claim 1, characterized in that, If the matching degree exceeds the set threshold, the user is deemed a legitimate user and is allowed to check in successfully. Otherwise, after prompting for re-signing or refusing to sign in, the method further includes: The original trajectory data of this signature, preset key dynamic features, and matching degree are uploaded to the backend server to automatically generate a sign-in log.
7. A teller check-in system based on signature trajectory recognition, characterized in that, The system includes: The data acquisition module is used to collect the original trajectory data of the current user during the signing process in real time through the touch interface of the preset signature area; The acquisition module is used to preprocess all the collected raw trajectory data to obtain preset key dynamic features. The preset key dynamic features include at least: stroke start and end time, average speed, maximum speed, trajectory curvature change, pressure distribution features, and stroke order pattern. The extracted preset key dynamic features are compared with the teller's historical key dynamic features in the background database using machine learning algorithms to obtain the matching degree. The judgment module is used to determine that if the matching degree exceeds the set threshold, the user is deemed a legitimate user and the check-in is allowed to proceed successfully; otherwise, the user is prompted to re-sign or the check-in is rejected.
8. The teller check-in system based on signature trajectory recognition according to claim 7, characterized in that, The acquisition module includes acquisition units. This is used to query the historical set of key dynamic features of the current user from the MySQL database based on the teller's employee number after receiving the current user's preset key dynamic features; Perform DTW alignment on the current preset key dynamic features and each group of historical key dynamic features respectively, and then input the aligned vectors into the trained SVM model to obtain the classification confidence. Based on the collection time of historical key dynamic features, the weights of each historical key dynamic feature are determined. The weights and classification confidence scores are then weighted, summed, and averaged to obtain the matching degree.
9. A teller check-in device based on signature trajectory recognition, characterized in that, The device includes: processor; And a memory having executable code stored thereon, which, when executed, causes the processor to perform a teller check-in method based on signature trajectory recognition as described in any one of claims 1-6.
10. A non-volatile computer storage medium, characterized in that, It stores computer instructions, which, when executed, implement a teller check-in method based on signature trajectory recognition as described in any one of claims 1-6.