Non-contact double-recording signature method and device based on terminal intelligence

Abstract

The invention provides a non-contact double-recording signature method and device based on end intelligence, and the method comprises the steps: loading a face recognition model and a gesture recognition model preset in end side equipment, and initializing a real-time video stream of a camera; face detection, living body judgment and facial feature comparison are carried out based on the real-time video stream, and an identity verification result is generated; synchronously recognizing a gesture of a user or an object moving track, matching a standard signature template through dynamic time warping, and generating a track consistency verification result; the identity verification result, the track consistency verification result and the signature track data are locally encrypted and stored, and are asynchronously uploaded to a server when the network is recovered; and through a timestamp alignment mechanism, audio and video streams in a signature process, an identity verification result and track data are synchronously recorded, encrypted and stored. According to the invention, the authenticity and security of signature are effectively improved, the risk of data leakage is reduced, and the system stability is enhanced.

Description

Technical Field

[0001] This invention relates to the fields of intelligent interaction technology and financial technology, and in particular to a contactless dual-recording signature method and device based on edge intelligence. Background Technology

[0002] Dual recording systems, as a crucial technical means for compliance management of financial transactions, are widely used in the banking, insurance, and other financial sectors. With the rapid expansion of remote transaction scenarios, existing technologies utilize the collaborative operation of electronic signatures, paper signatures, and facial recognition to construct a basic identity verification and signature process. Specifically, this technical system covers the entire process from user identity collection to signature recording, including key steps such as camera data acquisition, feature extraction, and cloud comparison. Electronic signature technology relies on touchscreens or electronic pens to complete the signing operation, while facial recognition achieves identity verification through cloud processing. Based on this, related systems typically use the HTTPS protocol for data transmission and combine it with H.264 / H.265 encoding for audio and video recording to meet regulatory requirements for transaction traceability.

[0003] However, existing dual-recording methods directly adopt a cloud-based processing model without fully considering the collaborative optimization of edge computing capabilities. This may lead to a spatiotemporal separation between identity verification and signing, or security risks during data transmission. Specifically, in traditional electronic signature systems, users must complete the signing operation through a specific device, and identity verification relies on cloud processing, resulting in limitations such as poor device compatibility and transaction failure due to network interruptions. Furthermore, existing technologies typically employ a single process for identity verification and signing, lacking a real-time synchronization mechanism for multimodal data, leading to technical defects such as insufficient signature validity (risk of proxy signing >5%) and a high probability of data leakage (exposure rate of sensitive information transmission >10%). Due to the stringent requirements for legal validity and privacy protection in financial transactions, this separated processing model struggles to form a complete chain of evidence and suffers from system availability bottlenecks in weak network environments, ultimately affecting transaction compliance and business continuity. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the related art.

[0005] Therefore, the first objective of this invention is to propose a contactless dual-recording signature method based on edge intelligence.

[0006] The second objective of this invention is to propose a contactless dual-recording signature device based on edge intelligence.

[0007] The third objective of this invention is to provide an electronic device.

[0008] The fourth objective of this invention is to provide a computer-readable storage medium.

[0009] The fifth objective of this invention is to provide a computer program product.

[0010] To achieve the above objectives, a first aspect of the present invention proposes a contactless dual-recording signature method based on edge intelligence, comprising: S1, load the pre-installed face recognition model and gesture recognition model in the terminal device, and initialize the real-time video stream from the camera; S2, based on real-time video stream, performs face detection and liveness determination, extracts the current user's facial features and compares them with pre-stored authorized features to generate identity verification results; S3 synchronously recognizes the user's gestures or the trajectory of an object's movement, performs dynamic time normalization matching on the trajectory data and a preset standard signature template, and generates a trajectory consistency verification result. S4 encrypts and stores the identity verification result, trajectory consistency verification result, and signed trajectory data locally, and asynchronously uploads them to the server when the network is restored; S5 uses a timestamp alignment mechanism to synchronously record and encrypt the audio and video streams during the signing process, along with the identity verification results and trajectory data.

[0011] Optionally, the loading of the pre-set face recognition model and gesture recognition model in the loading end-side device, and initialization of the real-time video stream from the camera, further includes: S11 uses the MTCNN algorithm for face detection and combines it with the KCF tracking algorithm to achieve face position tracking in consecutive frames. S12 utilizes GPU or NPU hardware acceleration modules to perform inference calculations on face recognition models and gesture recognition models.

[0012] Optionally, the step of performing face detection and liveness determination based on real-time video stream, extracting the current user's facial features and comparing them with pre-stored authorized features to generate an identity verification result further includes: S21, by detecting the user's blink frequency and the amplitude of head micro-movements To determine if a person is alive, among other things... Must meet times / second Must meet ; S22, using cosine similarity Perform feature comparison, when When determined to be an authorized user, among which There are two eigenvectors. This is a preset threshold.

[0013] Optionally, the step of synchronously recognizing the user's gestures or object movement trajectories, dynamically time-normalizing and matching the trajectory data with a preset standard signature template, and generating a trajectory consistency verification result further includes: S31 uses the MediaPipe Hands model to detect the coordinates of 21 key points on the hand. And based on the SVM classifier, it recognizes specific gestures; S32 calculates trajectory similarity using a dynamic time warping algorithm. ,when The signature is determined to be valid at that time, where This is the preset maximum similarity threshold.

[0014] Optionally, the step of locally encrypting and storing the authentication result, trajectory consistency verification result, and signature trajectory data, and asynchronously uploading them to the server when the network is restored, further includes: S41 uses the AES encryption algorithm to encrypt the data, and the encryption key is generated by a local random number generator; S42 utilizes the iOS Keychain or Android Keystore system to implement sandbox storage, ensuring that encrypted data is only accessed through preset APIs.

[0015] Optionally, the method further includes: S6 uses asynchronous I / O and message queue mechanisms for multi-threaded resource allocation, with the face recognition thread having a priority of [priority value missing]. The priority of the gesture recognition thread is And satisfy This is to ensure that real-time authentication takes precedence over the processing delay of trajectory capture.

[0016] To achieve the above objectives, a second aspect of the present invention provides a contactless dual-recording signature device based on edge intelligence, comprising: The model loading and camera initialization module is used to load the pre-installed face recognition model and gesture recognition model in the terminal device, and initialize the real-time video stream of the camera; The face detection and liveness verification module is used to perform face detection and liveness determination based on real-time video streams, extract the facial features of the current user and compare them with pre-stored authorized features to generate identity verification results; The gesture trajectory recognition and consistency verification module is used to synchronously recognize the user's gestures or the trajectory of an object's movement, dynamically time-normalize and match the trajectory data with a preset standard signature template, and generate trajectory consistency verification results. The data encryption storage and asynchronous upload module is used to encrypt and store the identity verification results, trajectory consistency verification results and signed trajectory data locally, and asynchronously upload them to the server when the network is restored; The timestamp alignment and synchronous recording module is used to synchronously record and encrypt the audio and video streams during the signing process, along with the identity verification results and trajectory data, through a timestamp alignment mechanism.

[0017] To achieve the above objectives, a third aspect of the present invention provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of the first aspects.

[0018] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of the first aspects.

[0019] To achieve the above objectives, a fifth aspect of the present invention provides a computer program product that, when executed by a processor, implements the method described in any one of the first aspects.

[0020] This invention provides a contactless dual-recording signature method, device, electronic device, and storage medium based on edge intelligence, which realizes contactless signature and real-time identity verification without network dependence, improves signature validity, data security, and system stability, and ensures compliance and traceability of financial transaction processes.

[0021] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a contactless dual-recording signature method based on edge intelligence provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of a system flow diagram of a contactless dual-recording signature method based on edge intelligence provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a contactless dual-recording signature device based on edge intelligence, provided in an embodiment of the present invention. Detailed Implementation

[0023] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0024] This invention provides a contactless dual-recording signature method based on edge intelligence. Figure 1 This is a flowchart illustrating a contactless dual-recording signature method based on edge intelligence provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of a system flow for a contactless dual-recording signature method based on edge intelligence, provided in an embodiment of the present invention. Figure 1 , Figure 2 As shown, the method includes the following steps: Step S1: Load the pre-set face recognition model and gesture recognition model in the terminal device, and initialize the real-time video stream from the camera.

[0025] Specifically, in one embodiment of the present invention, loading the pre-set face recognition model and gesture recognition model in the end-side device and initializing the real-time video stream of the camera are the basic steps for the startup and operation of the system. The technical implementation involves the model loading mechanism, the camera initialization process and the resource scheduling strategy.

[0026] Specifically, the system first loads the MobileFaceNet face feature extraction model and the MediaPipe Hands gesture recognition model locally on the device. Both models have undergone quantization and pruning to adapt to the computing power of the device. During model loading, the system uses memory mapping technology to map the model files into memory, avoiding frequent disk reads and thus improving loading efficiency and runtime response speed. After the model is loaded, the system initializes the camera module, calling the device's camera API (such as Android's CameraX or iOS's AVCaptureSession) to set the resolution, frame rate, and image format of the video stream. In one embodiment of the invention, the camera is typically configured with 1080p resolution and 30fps frame rate to ensure both recognition accuracy and real-time performance.

[0027] Furthermore, the system employs a multi-threaded architecture, deploying the face and gesture recognition modules in independent threads to achieve parallel processing. Camera video streams are distributed to each recognition module via shared memory or a queue mechanism, ensuring data synchronization and low-latency processing. This step provides a stable data input source for subsequent identity verification and contactless signature trajectory capture, and is a crucial step in achieving real-time interaction on the client side. Through local model loading and camera initialization, the system can quickly enter a working state in a network-free environment, significantly improving system availability and security.

[0028] Furthermore, S1 includes: S11 uses the MTCNN algorithm for face detection and combines it with the KCF tracking algorithm to achieve face position tracking in consecutive frames.

[0029] Specifically, in one embodiment of the present invention, the step of "using the MTCNN algorithm for face detection and combining it with the KCF tracking algorithm to achieve face position tracking in continuous frames" is a key step in realizing continuous face recognition on the edge. Its technical implementation principle is based on the multi-task convolutional neural network (MTCNN) and the Kernelized Correlation Filter (KCF) tracking algorithm. The two work together to improve the real-time performance and stability of the system in a network-free environment.

[0030] At the technical implementation level, the MTCNN algorithm first performs face detection on each frame of the image. This algorithm consists of three cascaded convolutional neural networks: P-Net for initial face region detection, R-Net for refining candidate boxes, and O-Net for further filtering and localization. The final O-Net outputs the bounding box of the face and five key points (eyes, nose tip, and corners of the mouth). When deployed on edge devices, MTCNN typically employs quantization compression techniques to reduce the model accuracy from FP32 to INT8, thereby reducing model size and inference latency, making it suitable for mobile devices or embedded platforms. The detection frequency is usually set to 15 frames per second (15 FPS) to balance real-time performance and computational load.

[0031] In continuous frame processing, to avoid resource waste caused by repeated detection, the system introduces the KCF tracking algorithm to track the location of detected faces. The KCF algorithm is based on the principle of correlation filtering, achieving efficient tracking by calculating the correlation between the target region and the template in each frame. Its core formula is:

[0032] in, The Fourier transform features of the current frame. For the frequency domain representation of the training filter, This is a regularization parameter used to prevent division overflow. When running the KCF algorithm on the client side, it is usually implemented in conjunction with the OpenCV library, achieving a tracking accuracy of over 95% and a response time of less than 50ms, meeting the requirements for real-time interaction.

[0033] In this application scenario, this step is primarily used for continuous identity verification during the dual-recording signing process. When a user begins signing, the system captures images in real time via a camera and detects and tracks the face in each frame to ensure the signer remains within the authorized range. If the face is lost or identity verification fails, the system will automatically pause the signing process to prevent proxy signing.

[0034] The technical advantage of this step is that by combining MTCNN and KCF, the system can achieve efficient and stable face detection and tracking on the edge without relying on cloud computing, which significantly reduces network latency and data leakage risks, while improving the real-time performance and accuracy of identity verification, providing a reliable identity foundation for subsequent signature validity judgment.

[0035] S12 utilizes GPU or NPU hardware acceleration modules to perform inference calculations on face recognition models and gesture recognition models.

[0036] Specifically, the step of "using GPU or NPU hardware acceleration modules to perform inference calculations on the face recognition model and gesture recognition model" in this invention is a key step in achieving real-time intelligent recognition on the device side. By calling dedicated hardware acceleration modules such as the graphics processing unit (GPU) or neural network processing unit (NPU) on the device, efficient inference calculations are performed on the locally deployed MobileFaceNet face feature extraction model and MediaPipe Hands gesture key point detection model, thereby significantly improving recognition speed and system responsiveness.

[0037] At the technical implementation level, both GPUs and NPUs possess parallel computing capabilities, enabling efficient handling of large-scale matrix operations in convolutional neural networks (CNNs). The MobileFaceNet model employs a lightweight design, typically inputting images of 112×112 pixels and outputting a 128-dimensional facial feature vector. With GPU acceleration, the model's inference latency can be controlled to within 30ms, meeting the requirements of real-time authentication. The MediaPipe Hands model, based on the MobileNetV2 backbone network, outputs coordinates of 21 hand keypoints. Its inference latency, supported by the NPU, can be further reduced to below 20ms, ensuring continuous capture of gesture trajectories.

[0038] In terms of parameters, during hardware acceleration, the system typically sets the image frame rate of the model input to 15-30 FPS to balance recognition accuracy and computational load. For GPU acceleration, optimization frameworks such as TensorRT or OpenCL can be enabled to perform graph optimization and memory management on the model, improving inference efficiency. NPU acceleration uses Quantized Awareness Training (QAT) to compress model weights into INT8 format, reducing computational resource consumption while maintaining a recognition accuracy of over 95%.

[0039] In application scenarios, this step is widely applicable to dual-recording systems in finance, government affairs, and other sectors that have high requirements for real-time performance, privacy, and security. For example, during remote account opening or contract signing, users do not need to wear additional devices; they can complete identity verification and contactless signing simply through a camera. The system completes all recognition tasks locally, without needing to upload original images or trajectory data, effectively reducing the risk of data leakage.

[0040] The technical benefits of this step are that it significantly improves model inference efficiency through hardware acceleration, reduces power consumption and latency of edge devices, enhances the availability and stability of the system in a network-free environment, and provides solid technical support for realizing contactless dual-recording signature driven by edge intelligence.

[0041] Step S2: Perform face detection and liveness determination based on real-time video stream, extract the facial features of the current user and compare them with pre-stored authorized features to generate identity verification results.

[0042] Specifically, the face detection and liveness detection steps based on real-time video streams are the core components of this invention to ensure the legality and security of signatures. This step achieves real-time verification of the current user's identity by deploying a lightweight face detection and recognition model on the edge device. Specifically, the system uses the MTCNN (Multi-task Cascaded Convolutional Networks) algorithm for face detection. This algorithm can efficiently detect face regions in a single frame image with an accuracy of over 95%, and the average detection latency on mobile devices is less than 100ms. After detection, the system uses the MobileFaceNet model to extract face feature vectors. This model achieves a recognition accuracy of over 99.5% on the LFW (Labeled Faces in the Wild) dataset, and its size is kept below 2MB, facilitating deployment on the edge.

[0043] Furthermore, to prevent liveness detection attacks (such as those involving photos, videos, or screen captures), the system introduces a multimodal liveness detection mechanism. In addition to a visible light camera, an infrared camera can be optionally combined to collect facial thermal radiation information, enhancing anti-spoofing capabilities. Simultaneously, the system analyzes behavioral characteristics such as the user's blinking frequency (typically set at 3-8 times per minute) and the amplitude of subtle head movements (e.g., vertical movement not exceeding 5°, horizontal movement not exceeding 3°) to determine if the user is a genuine live subject. The confidence threshold for liveness detection can be set to 0.85; exceeding this threshold indicates successful liveness detection.

[0044] During the identity verification phase, the system performs a cosine similarity comparison between the feature vector extracted in real time and the pre-stored authorized feature vector. The calculation formula is as follows: ,in This represents the feature vector extracted in real time. This indicates pre-stored authorization features. When the similarity is greater than a set threshold (e.g., 0.92), the system determines the user as legitimate; otherwise, verification is rejected.

[0045] This step plays a crucial role in identity authentication within the system, ensuring that the signature is completed by the authorized user, thereby enhancing the legal validity of the signature and system security. Simultaneously, since all processing is completed on the client side, there is no need to upload original images, effectively reducing the risk of privacy breaches and meeting the high data security requirements of the financial industry.

[0046] Furthermore, S2 includes: S21, by detecting the user's blink frequency and the amplitude of head micro-movements To determine if a person is alive, among other things... Must meet times / second Must meet .

[0047] Specifically, in one embodiment of the present invention, the user's blink frequency is detected. and the amplitude of head micro-movements The liveness detection step is a key anti-spoofing mechanism in edge-based intelligent facial recognition modules, used to prevent attacks using non-liveness techniques such as photos, videos, or 3D masks. In some implementations, this step analyzes the dynamic changes of key points on the user's face in real time, extracting blink frequency and head micro-movement amplitude as biometric features for liveness detection.

[0048] In its specific technical implementation, the system first utilizes MTCNN (Multi-task Cascaded Convolutional Networks) for face detection and key point localization, obtaining the eye region and head pose angles. Blink frequency is also measured. The calculation is based on the periodic changes in eyelid closure and opening. By detecting changes in the vertical distance between key points on the eyelid and combining this with timestamp information, the number of blinks per unit time is counted. To ensure data stability, the system employs a sliding window mechanism, processing continuous data... The blink events within a second are averaged out, making The range of values ​​is limited to The blink rate is within the physiological range of normal human blinking frequency.

[0049] Head micro-movement amplitude This is achieved through head pose estimation, typically using Euler angles calculated from facial key points to represent the head's rotation angle in 3D space. The system calculates the micro-motion amplitude using the root mean square (RMS) of angle changes between consecutive frames, ensuring... satisfy This verifies whether the user is a real, living person, rather than a static image or video.

[0050] In practical applications, this step can be deployed on edge devices such as smartphones and embedded cameras. It combines the MobileFaceNet model for feature extraction and performs liveness detection locally, without relying on cloud computing, significantly improving the system's real-time performance and privacy protection capabilities. Furthermore, this invention effectively enhances the identity verification security during the dual recording process, ensuring that the signature is completed by a genuine user, thereby improving the legal validity of the signature and the overall compliance of the system.

[0051] S22, using cosine similarity Perform feature comparison, when When determined to be an authorized user, among which This is a preset threshold.

[0052] Specifically, this invention employs cosine similarity. Feature comparison is one of the key technical steps in achieving user authentication. This step is mainly used to compare the currently captured facial feature vectors in real time on the device. Feature vectors of pre-registered authorized users This allows us to determine whether the current user is a legitimate authorized person.

[0053] At the technical implementation level, the system first extracts feature vectors from face images using the MobileFaceNet model, which outputs a 128-dimensional normalized feature vector. In each frame, the system continuously detects and extracts facial features, then calculates cosine similarity with locally stored features of authorized users. Cosine similarity measures the cosine of the angle between two vectors, ranging from -1 to 1; a value closer to 1 indicates greater similarity. In this system, a similarity threshold is set. ,when If the user is authorized, the authentication is considered complete; otherwise, the authentication is rejected.

[0054] Regarding parameter metrics, threshold The settings need to be adjusted according to the actual application scenario. In scenarios with high security requirements, such as financial dual recording, it is usually... Set the threshold to 0.85 or higher to ensure accurate recognition. Furthermore, the system supports dynamic threshold adjustment; for example, it can be appropriately lowered in cases of changing lighting or occlusion. To improve recognition robustness, while further enhancing anti-counterfeiting capabilities through liveness detection (such as blinking and head micro-movements).

[0055] In this application scenario, this step is integrated into the continuous facial recognition module of the edge device to verify the user's identity in real time when making contactless signatures. The system does not rely on cloud servers; all comparison calculations are completed locally, meeting the low latency and high privacy requirements of edge AI. It is suitable for financial transaction scenarios with unstable networks or those sensitive to data security.

[0056] The technical advantage of this step lies in its ability to effectively prevent proxy signing through a localized, real-time identity verification mechanism, thereby enhancing the legitimacy and security of the signing process. Simultaneously, by incorporating a cosine similarity algorithm, the system can achieve efficient identity comparison while maintaining recognition accuracy, providing reliable identity protection for the entire dual-recording signing process.

[0057] Step S3: Simultaneously identify the user's gestures or object movement trajectory, perform dynamic time normalization matching between the trajectory data and the preset standard signature template, and generate trajectory consistency verification results.

[0058] Specifically, in this invention, synchronously recognizing the user's gestures or object movement trajectories and matching them with a preset standard signature template using dynamic time warping (DTW) is one of the core steps in achieving contactless signature verification. This step uses the camera on the edge device to capture the user's gestures or the movement trajectory of an object (such as a pen) in space in real time, forming a continuous coordinate sequence, which is then compared with a pre-stored standard signature template to generate a trajectory consistency verification result, ensuring the authenticity and legality of the signing behavior.

[0059] At the technical implementation level, gesture or object recognition uses the MediaPipe Hands model to detect key hand points, locating a total of 21 key points for extracting motion features of the gesture. For object recognition, the Tiny-YOLO model is used for target detection, identifying pen-like objects of a specific shape. The positional information of the recognized gesture or object is continuously tracked using Kalman filtering and optical flow methods to form trajectory data. The trajectory data is synchronized based on timestamps to ensure consistency with the output of the face recognition module in the time dimension. After trajectory capture, the system performs smoothing processing, typically using a low-pass filter or moving average algorithm, to eliminate noise caused by camera shake or minor hand movements of the user.

[0060] At the parameter level, trajectory data is typically presented as a two-dimensional coordinate sequence. It means that among them This is a timestamp. The DTW algorithm is used to calculate the similarity between a user's trajectory and a standard template. Its core formula is:

[0061] in Indicates the first The user trajectory point and the first The cumulative distance between template trajectory points This involves using Euclidean distance or other similarity metrics. Ultimately, the DTW distance is calculated and compared to a preset threshold. Compare the trajectories to determine if they match.

[0062] At the application level, this procedure is widely applicable to dual-recording systems in the financial sector, especially in scenarios such as remote account opening and electronic contract signing. Users do not need to touch the screen; they can complete the signature simply by gestures or the movement of objects. The system completes the recognition and verification locally, without uploading the original trajectory data, thereby reducing the risk of privacy leaks.

[0063] The technical advantage of this step lies in achieving high-precision matching of contactless signature trajectories through the DTW algorithm, effectively preventing proxy signing and enhancing the legal validity of the signature. Simultaneously, since all processing is completed on the edge, the system features low latency, high privacy, and strong stability, making it suitable for scenarios with complex network environments or high real-time requirements.

[0064] Furthermore, S3 includes: S31 uses the MediaPipe Hands model to detect the coordinates of 21 key points on the hand. It also identifies specific gestures based on an SVM classifier.

[0065] Specifically, in the intelligent contactless dual-recording signature system of the present invention, the MediaPipe Hands model is used to detect the coordinates of 21 key points on the hand. The process of recognizing specific gestures using an SVM classifier is one of the core steps in achieving contactless signature trajectory capture and gesture semantic parsing. This step involves acquiring real-time hand images through the device's camera, processing the images using the MediaPipe Hands model, and outputting the 3D coordinate information of 21 key points on the hand, including... , and Three dimensions, among which Indicates the first The spatial coordinates of key points. The MediaPipe Hands model is based on the MediaPipe framework and uses a lightweight machine learning model to achieve low-latency, high-precision hand key point detection on mobile devices or embedded systems. Its detection frequency can typically reach more than 30 FPS, meeting the needs of real-time interaction.

[0066] In terms of gesture recognition, the system further extracts the relative position and angle features between key points to construct a gesture feature vector. This feature vector serves as the input to the SVM classifier, used to recognize preset specific gestures (such as "clenching a fist," "spreading five fingers," "writing," etc.). During the training phase, the SVM classifier uses the Support Vector Machine algorithm, employing kernel functions (such as the RBF kernel) to model the classification boundaries of high-dimensional features. The classification accuracy can reach over 95%, and the model size is kept within 5MB, facilitating deployment on edge devices.

[0067] In practical applications, this step can be widely used in scenarios with high requirements for identity verification and signature compliance, such as finance, government affairs, and healthcare. Through localized processing, the system does not rely on cloud computing, reducing data transmission risks and improving response speed and privacy protection capabilities. Simultaneously, gesture recognition and facial recognition operate in parallel, ensuring continuous identity verification during the signing process and enhancing the traceability and legal validity of the signing action.

[0068] S32 calculates trajectory similarity using a dynamic time warping algorithm. ,when The signature is determined to be valid at that time, where This is the preset maximum similarity threshold.

[0069] Specifically, this step involves calculating the similarity of the user's signature trajectory based on the Dynamic Time Warping (DTW) algorithm to determine whether it is a valid signature. DTW is a non-linear algorithm used to align and compare two time series, particularly suitable for situations where there are time offsets or speed differences in trajectory data. In one embodiment of the invention, the user completes a contactless signature through gestures or the movement trajectory of an object. The system performs DTW matching on this trajectory and calculates a similarity index with a preset standard signature trajectory. ,in and These represent the user's current trajectory and the standard trajectory at the [number]th ... The coordinate difference of each sampling point The total number of trajectory points. This formula sums the absolute differences between trajectory points and takes the reciprocal, so that the closer the similarity value is to 1, the more matched the trajectories are, and the closer it is to 0, the greater the difference.

[0070] At the implementation level, the system first normalizes the user's gestures or the motion trajectory of objects, mapping the trajectory points to a unified coordinate space to eliminate the influence of differences in device resolution or viewing angle. Then, the DTW algorithm is used to align the user's trajectory with a standard template trajectory and calculate the distance, ultimately obtaining a similarity value. .when When the signature is deemed valid, it is considered to be one of the following: The preset maximum similarity threshold is usually set within the range of [0.8, 0.95] based on actual test data to balance the false positive rate and the recognition accuracy.

[0071] This step plays a crucial role in contactless signature systems, particularly in enabling offline verification on edge devices without relying on cloud computing. By locally determining trajectory similarity, the system can verify signature validity even without a network connection. Combined with real-time facial recognition, it ensures that the signing process and identity verification occur simultaneously, significantly improving the compliance and security of the signature. Furthermore, this method supports various signature tools (such as gestures and pen-like objects), enhancing the system's applicability and user-friendliness.

[0072] Step S4: The identity verification result, trajectory consistency verification result, and signed trajectory data are encrypted and stored locally, and then asynchronously uploaded to the server when the network is restored.

[0073] Specifically, in one embodiment of the present invention, the step of "locally encrypting and storing the authentication result, trajectory consistency verification result, and signed trajectory data, and asynchronously uploading them to the server when the network recovers" is a key step in ensuring the security and availability of system data. This step achieves local protection and reliable transmission of sensitive data through end-side encryption mechanisms and asynchronous upload strategies.

[0074] At the technical implementation level, the system first structurally encapsulates the identity verification results (such as cosine similarity values) output by the face recognition module and the signature trajectory data generated by the gesture recognition module. The trajectory data is typically represented in the form of a timestamp sequence, with each trajectory point containing coordinates. timestamp and velocity vector Information such as... The consistency verification results are then calculated using the Dynamic Time Warping (DTW) algorithm to determine the similarity between the trajectory and the standard template. ,in This represents the DTW distance. This data is organized locally in JSON or binary format and encrypted using algorithms such as AES-256 or RSA-2048. The encryption key is generated and stored by the device's local security module (such as iOS Keychain or Android Keystore) to ensure that the key is not accessed externally.

[0075] At the parameter level, the system employs the AES-256 encryption algorithm with a block size of 128 bits and a key length of 256 bits, meeting financial-grade data encryption standards (such as FIPS 140-2). The asynchronous upload mechanism is based on the HTTP / HTTPS protocol and uses a segmented upload strategy, keeping the size of a single upload data packet under 1MB to adapt to weak network environments. Upload priority is determined by both the timestamp and the data integrity hash value (such as SHA-256), ensuring data order and consistency.

[0076] At the application level, this step applies to scenarios where users remotely sign contracts in a financial dual-recording system. When a user completes a signature without a network or with an unstable network, the system ensures that the data is not lost and automatically uploads it to the cloud server after the network is restored, achieving final data archiving and compliance recording.

[0077] The technical benefits of this step are that local encrypted storage effectively prevents data leakage or tampering during the storage phase; and the asynchronous upload mechanism ensures the system continues to operate even during network outages, improving overall robustness and user experience. Furthermore, this design complies with data privacy regulations such as GDPR and CCPA, enhancing the system's security and compliance.

[0078] Furthermore, S4 includes: S41 uses the AES encryption algorithm to encrypt the data, and the encryption key is generated by a local random number generator.

[0079] Specifically, in the local storage and encryption module, the AES encryption algorithm is used to encrypt data. The encryption key is generated by a local random number generator, which is one of the key steps to ensure system data security. The technical implementation of this step is based on a symmetric encryption mechanism, specifically using the AES-256 standard. Its encryption strength is a 256-bit key length, which conforms to internationally accepted encryption security standards (such as NIST SP800-38D) and can effectively resist known cryptographic attack methods.

[0080] In some implementations, the system generates a 256-bit key that meets cryptographic security requirements by calling the secure random number generation interface provided by the device's operating system (such as SecRandomCopyBytes in iOS or the SecureRandom class in Android). This key is used to encrypt sensitive information, including facial recognition feature vectors, gesture trajectory data, OCR recognition results, and audio / video file metadata. The encryption process uses AES's CBC (Cipher Block Chaining) mode to ensure dependencies between data blocks, thereby improving the unpredictability and security of the encrypted data. The initialization vector (IV) is also generated by a random number generator and stored along with the ciphertext to support subsequent decryption operations.

[0081] Furthermore, the encrypted data is written to a secure local storage area on the device, such as iOS's Keychain or Android's Keystore system. These mechanisms prevent unauthorized access or data leakage through sandbox isolation and hardware-level encryption. The system can still operate normally in a network-free environment; all encrypted data is cached locally and uploaded to the server via a network synchronization module once the network is restored, ensuring data integrity and traceability.

[0082] This step plays a crucial role in the overall system's security. By generating local keys and using AES encryption, it effectively reduces the risk of data exposure during transmission, enhances the privacy protection capabilities of edge data processing, and meets financial-grade data security requirements.

[0083] S42 utilizes the iOS Keychain or Android Keystore system to implement sandbox storage, ensuring that encrypted data is only accessed through preset APIs.

[0084] Specifically, in the local storage and encryption module of this invention, sandbox storage is implemented using the iOS Keychain or Android Keystore system, which is a key technical means to ensure the local security of users' sensitive data. The core of this step lies in using the secure storage mechanism provided by the operating system to encrypt and store key information, including facial recognition features, gesture trajectories, OCR recognition results, and audio and video metadata, ensuring that the data can only be accessed through preset API interfaces, thereby preventing unauthorized reading or tampering.

[0085] At the technical implementation level, the iOS platform uses the Keychain Services API, which is based on Apple's cryptographic services framework and supports the AES-256 encryption algorithm. Data is stored in key-value pairs within a system-level encrypted sandbox. Keychain features access control policies, allowing users to set biometric authentication (Touch ID / Face ID) or device unlock status as a prerequisite for access, further enhancing security. The Android platform uses the Android Keystore system, which is based on the Java Security API, supports encryption algorithms such as RSA and AES, and implements hardware-level encryption key protection through the Android Keymaster module. This ensures that keys are not extracted into memory, thereby preventing man-in-the-middle attacks or malware theft.

[0086] At the parameter level, the key length of the encryption algorithm is recommended to be no less than 256 bits (e.g., AES-256) to meet financial-grade security standards (e.g., FIPS 140-2). The data storage format uses Encrypted Data Blocks (EDBs), with each block containing a timestamp, user ID, encrypted data, and a hash checksum to ensure data integrity and traceability. Access control policies can be configured for iOS or Android to achieve secure access based on user identity.

[0087] In application scenarios, this module is widely applicable to dual-recording systems in finance, law, and healthcare, where data privacy and security requirements are high. For example, during remote account opening or contract signing, user identity verification data and signature traces can be encrypted and stored locally in real time, ensuring the continuity of the transaction process and the integrity of the data even in offline environments. Once the network is restored, the system uploads the data to the server through the network synchronization module, realizing a secure end-to-cloud collaborative mechanism.

[0088] The technical benefits of this step lie in significantly enhancing the system's data security and privacy protection capabilities. Through operating system-level encrypted storage mechanisms, not only is data leakage prevented, but the controllability and compliance of data access are also ensured. Furthermore, the combination of the sandbox storage mechanism and pre-defined APIs gives the system a robust modular design and maintainability, providing a solid security foundation for the edge-to-edge intelligent dual-recording system.

[0089] Step S5: Through a timestamp alignment mechanism, the audio and video streams during the signing process are synchronously recorded and encrypted with the identity verification results and trajectory data.

[0090] Specifically, the key step in achieving a highly secure and traceable signature process in this invention is to synchronously record and encrypt the audio and video streams during the signing process, along with the identity verification results and trajectory data, through a timestamp alignment mechanism. The core of this step lies in ensuring strict alignment of multi-source heterogeneous data in the time dimension, thereby enabling accurate reconstruction of the entire signing process during subsequent audits or compliance reviews.

[0091] At the technical implementation level, the system deploys a multi-threaded data acquisition and processing mechanism on the edge device, acquiring audio and video streams, trajectory data, and authentication results from the camera, microphone, gesture recognition module, and face recognition module, respectively. Each data source is appended with a high-precision timestamp during acquisition; the timestamp generation is based on the device's local system clock, with a time resolution reaching the microsecond level. (s) Ensures precise alignment of each data stream on the timeline. Furthermore, the system employs timestamp interpolation and synchronization compensation algorithms to correct timing deviations caused by processing delays or frame rate differences. For example, it uses linear interpolation to time-map trajectory points to video frames, as shown in the formula: ,in For video frame timestamps, For the timestamp of the trajectory data, For time intervals, The frame interval.

[0092] In terms of specifications, the system supports adjustable audio and video stream recording frame rates between 15 and 30 fps, trajectory data acquisition frequency of 60 Hz, and authentication result update frequency of 10 Hz. Timestamp synchronization error is controlled within... Within the millisecond range, it meets the compliance requirements of financial-grade dual-recording systems. Encrypted storage uses the AES-256 algorithm with a 256-bit key length. The encryption process is completed locally, ensuring that the data is tamper-proof before storage and transmission.

[0093] In terms of application scenarios, this procedure is widely applicable to remote business scenarios with high requirements for signature compliance, such as finance, law, and healthcare. For example, during remote bank account opening or loan signing, the system can record the user's identity verification status, signature trajectory, and audio / video content in real time, and store them locally in an encrypted manner. Once the network is restored, these records are uploaded to the backend server, achieving dual protection for offline operation and compliance auditing.

[0094] From a technical perspective, this step effectively solves the timing misalignment problem caused by asynchronous data collection from multiple sources through a timestamp alignment mechanism, ensuring the integrity and consistency of the signing process. Simultaneously, combined with local encrypted storage, it significantly improves data security, reduces the risk of data loss or tampering due to network interruptions or attacks, and provides reliable technical support for contactless signature systems.

[0095] Step S6: Multi-threaded resource allocation is performed based on asynchronous I / O and message queue mechanisms, with the face recognition thread having a priority of [priority value missing]. The priority of the gesture recognition thread is And satisfy This is to ensure that real-time authentication takes precedence over the processing delay of trajectory capture.

[0096] Specifically, this invention achieves parallel processing of face recognition and gesture recognition functions through a multi-threaded resource allocation strategy based on asynchronous I / O and message queue mechanisms, thereby improving the overall real-time performance and stability of the system. Specifically, the system constructs a multi-threaded architecture on the edge device, where the priority of the face recognition thread is set to... The priority of the gesture recognition thread is set to And satisfy This is to ensure that real-time authentication takes precedence over the processing delay of trajectory capture.

[0097] From a technical implementation perspective, the system employs asynchronous I / O mechanisms (such as those based on the Reactor pattern or event-driven models) to achieve non-blocking reading and distribution of camera data streams. Each recognition module (face recognition and gesture recognition) runs in an independent thread, receiving image frame data through a message queue (such as Ring Buffer or ZeroMQ) and asynchronously feeding back the processing results to the main thread. The face recognition thread has a higher priority, meaning it receives higher execution rights in CPU scheduling, thus ensuring that the system can continuously verify the user's identity when performing a signature operation, preventing unauthorized personnel from operating the system.

[0098] At the parameter level, priority and It can be dynamically configured according to the device's hardware performance and task requirements, such as on ARM-based embedded devices. It can be set to "real-time priority (SCHED_FIFO)" in the thread scheduler, and Then set it to "normal priority (SCHED_OTHER)" to ensure that face recognition tasks can still be executed with priority when resources are scarce. The message queue buffer size is usually set to Frames are used to balance real-time performance with system load.

[0099] In application scenarios, this mechanism is particularly suitable for dual-recording systems in the financial sector, where the system needs to simultaneously complete identity verification and trajectory capture when a user makes a contactless signature. Through priority scheduling and asynchronous processing, the system can operate stably on low-power devices, meeting the real-time and data integrity requirements of information security standards such as ISO / IEC 27001.

[0100] The technical effect of this step is that, through thread priority control and asynchronous message queue scheduling, the system achieves efficient collaboration between face recognition and gesture recognition, improves the real-time performance of identity verification and the smoothness of the signing process, thereby enhancing the system's security, stability, and user experience.

[0101] To achieve the above embodiments, the present invention also proposes a contactless dual-recording signature device based on edge intelligence. Figure 3 This is a schematic diagram of a contactless dual-recording signature device based on edge intelligence, provided as an embodiment of the present invention. Figure 3 As shown, the device includes: The model loading and camera initialization module 100 is used to load the pre-set face recognition model and gesture recognition model in the terminal device and initialize the real-time video stream of the camera. The face detection and liveness verification module 200 is used to perform face detection and liveness judgment based on real-time video stream, extract the facial features of the current user and compare them with pre-stored authorized features to generate identity verification results; The gesture trajectory recognition and consistency verification module 300 is used to synchronously recognize the user's gestures or the trajectory of an object's movement, dynamically time-normalize and match the trajectory data with a preset standard signature template, and generate a trajectory consistency verification result. The data encryption storage and asynchronous upload module 400 is used to locally encrypt and store the identity verification result, trajectory consistency verification result and signed trajectory data, and asynchronously upload them to the server when the network is restored; The timestamp alignment and synchronous recording module 500 is used to synchronously record and encrypt the audio and video streams, identity verification results, and trajectory data during the signing process through a timestamp alignment mechanism.

[0102] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0103] To implement the above embodiments, the present invention also proposes an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory to implement the method provided in the foregoing embodiments.

[0104] To implement the above embodiments, the present invention also proposes a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the methods provided in the foregoing embodiments.

[0105] To implement the above embodiments, the present invention also proposes a computer program product, including a computer program that, when executed by a processor, implements the methods provided in the foregoing embodiments.

[0106] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in this invention all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0107] It should be noted that personal information collected from users should be used for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. Furthermore, such collection / sharing should only be conducted after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign an agreement / authorization that includes authorization of relevant user information before the user uses the function. In addition, any necessary steps must be taken to protect and safeguard access to such personal information data and ensure that others with access to personal information data comply with their privacy policies and procedures.

[0108] This invention is intended to provide implementation schemes for users to selectively prevent the use or access to personal information data. That is, this disclosure is intended to provide hardware and / or software to prevent or block access to such personal information data. Once personal information data is no longer needed, risks can be minimized by restricting data collection and deleting data. Furthermore, where applicable, such personal information can be de-identified to protect user privacy.

[0109] In the foregoing descriptions of the embodiments, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0110] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0111] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0112] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0113] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0114] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0115] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0116] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

[0117] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0118] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A non-contact dual-recording signature method based on end intelligence, characterized in that, The method comprises the following steps: S1, loading the preset face recognition model and gesture recognition model in the terminal side device, and initializing the real-time video stream of the camera; S2, face detection and liveness judgment based on the real-time video stream, extracting the facial features of the current user and comparing them with the pre-stored authorized features to generate an identity verification result; S3, synchronously recognizing the gesture or object movement trajectory of the user, performing dynamic time warping matching of the trajectory data with the preset standard signature template to generate a trajectory consistency verification result; S4, locally encrypting and storing the identity verification result, the trajectory consistency verification result, and the signature trajectory data, and asynchronously uploading them to the server when the network is restored; S5, synchronously recording and encrypting the audio and video stream in the signature process with the identity verification result and the trajectory data through a timestamp alignment mechanism.

2. The method of claim 1, wherein, Loading the preset face recognition model and gesture recognition model in the terminal side device, and initializing the real-time video stream of the camera, further comprising: S11, face detection using the MTCNN algorithm, and continuous frame face position tracking combining the KCF tracking algorithm; S12, using a GPU or NPU hardware acceleration module to perform inference calculation on the face recognition model and the gesture recognition model.

3. The method of claim 1, wherein, Face detection and liveness judgment based on the real-time video stream, extracting the facial features of the current user and comparing them with the pre-stored authorized features to generate an identity verification result, further comprising: S21, by detecting the user's blink frequency and head micro-motion amplitude to determine the living body, wherein need to meet times per second, need to meet ; S22, cosine similarity is adopted characteristics are compared, and when the user is determined as authorized, wherein are two characteristic vectors, is a preset threshold.

4. The method of claim 1, wherein, Synchronously recognizing the gesture or object movement trajectory of the user, performing dynamic time warping matching of the trajectory data with the preset standard signature template to generate a trajectory consistency verification result, further comprising: S31, adopt MediaPipe Hands model to detect 21 key point coordinates of hand and identify specific gestures based on SVM classifier; S32, calculating the trajectory similarity by a dynamic time warping algorithm When is determined as a valid signature, wherein is a preset maximum similarity threshold.

5. The method of claim 1, wherein, Locally encrypting and storing the identity verification result, the trajectory consistency verification result, and the signature trajectory data, and asynchronously uploading them to the server when the network is restored, further comprising: S41, using the AES encryption algorithm to encrypt the data, and generating the encryption key through a local random number generator; S42, using the iOS Keychain or Android Keystore system to realize sandbox storage, and ensuring that the encrypted data is accessed only through a preset API.

6. The method of claim 1, wherein, Further comprising: S6, multi-threaded resource allocation based on asynchronous I / O and message queue mechanism, wherein the face recognition thread has a priority of , the gesture recognition thread has a priority of , and is satisfied to ensure that the real-time performance of identity verification is prioritized over the processing delay of trajectory capture.

7. A contactless dual-recording signature device based on end intelligence, characterized by, Comprising: A model loading and camera initialization module for loading the preset face recognition model and gesture recognition model in the terminal side device, and initializing the real-time video stream of the camera; A face detection and liveness verification module for face detection and liveness judgment based on the real-time video stream, extracting the facial features of the current user and comparing them with the pre-stored authorized features to generate an identity verification result; A gesture trajectory recognition and consistency verification module for synchronously recognizing the gesture or object movement trajectory of the user, performing dynamic time warping matching of the trajectory data with the preset standard signature template to generate a trajectory consistency verification result; A data encryption storage and asynchronous upload module for locally encrypting and storing the identity verification result, the trajectory consistency verification result, and the signature trajectory data, and asynchronously uploading them to the server when the network is restored; A timestamp alignment and synchronous recording module for synchronously recording and encrypting the audio and video stream in the signature process with the identity verification result and the trajectory data through a timestamp alignment mechanism.

8. An electronic device, comprising: Comprising: A processor, and a memory in communication connection with the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method of any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the method of any one of claims 1-6.

10. A computer program product, characterised in that, A computer program is included, which, when executed by a processor, implements the method of any one of claims 1-6.

Images