Remote video monitoring system, method and electronic device

Abstract

This application provides a remote video supervision system, method, and electronic device, relating to the field of financial technology. Specifically: the head office device is configured to issue video supervision tasks to branch office devices. The branch office device is configured to collect monitoring videos from branch outlets. Upon receiving a video supervision task, the branch office device, based on the corresponding video supervision task requirements, applies an edge computing algorithm to extract and store corresponding monitoring video segments from the branch outlet's monitoring videos; it then organizes the monitoring video segments into a video file list and transmits it to the head office device. The head office device is also configured to, based on the video file list, retrieve corresponding target monitoring video segments from the branch office devices as needed; apply a large-scale visual understanding model to perform compliance identification based on multimodal semantic analysis on the target monitoring video segments, obtaining and outputting the corresponding compliance identification results. This application can improve the quality inspection efficiency and accuracy of the remote video supervision process.

Description

Technical Field

[0001] This application relates to the field of financial technology, and in particular to a remote video monitoring system, method and electronic device. Background Technology

[0002] In the banking and other financial industries, due to internal risk control, operational compliance checks, and external regulatory requirements, the head office needs to conduct regular and standardized sampling quality inspections of monitoring videos of branch outlets, important business areas, and key positions. By verifying the video content, the head office focuses on supervising whether frontline staff are compliant in business processing, service standards, safe operations, and risk prevention.

[0003] In the existing remote video monitoring process, the head office issues a task list to the branch offices through the management system. The branch offices then copy the monitoring videos from the security system or an outsourced company and manually upload them to the cloud drive. Head office quality control personnel manually download the monitoring videos provided by the branches from the cloud drive and perform manual quality checks on each video. If any quality issues are found, the head office notifies the branch offices via phone, email, or the system to verify and rectify the issues. If no quality issues are found, the process ends.

[0004] However, the aforementioned remote video supervision method suffers from low quality inspection efficiency and is prone to errors. Summary of the Invention

[0005] This application provides a remote video supervision system, method, and electronic device to improve the quality inspection efficiency and accuracy of remote video supervision processes.

[0006] In a first aspect, this application provides a remote video monitoring system, comprising: head office terminal equipment, branch office terminal equipment, and sub-branch terminal equipment, wherein:

[0007] The head office terminal equipment is configured to send video monitoring tasks to the branch terminal equipment;

[0008] The branch-level equipment is configured to collect monitoring videos from branch outlets.

[0009] The branch terminal device is configured to respond to the acquisition of a video supervision task, and according to the video supervision task requirements, apply an edge algorithm to extract and store monitoring video segments corresponding to the video supervision task requirements from the monitoring videos collected by the lower-level branch terminal device; and organize the monitoring video segments into a video file list and transmit the video file list to the head office terminal device.

[0010] The head office terminal equipment is also configured to retrieve the corresponding target monitoring video segments from the branch terminal equipment on demand based on a list of video files; and to apply a large visual understanding model to perform compliance identification on the target monitoring video segments based on multimodal semantic analysis, and to obtain and output the compliance identification results corresponding to the target monitoring video segments.

[0011] In one possible implementation, the branch office equipment includes: a security network and a production network, wherein:

[0012] The security network is configured to respond to a video surveillance task, and according to the video surveillance task requirements, apply edge computing algorithms to extract and store the surveillance video segments corresponding to the video surveillance task requirements from the surveillance videos collected by the lower-level branch terminal devices; and organize the surveillance video segments into a video file list and transmit the video file list to the head office terminal device.

[0013] The production network is configured to interact with the head office equipment to receive the head office equipment's request for acquiring target monitoring video clips; and to interact with the security network to acquire target monitoring video clips from the security network and send the target monitoring video clips to the head office equipment.

[0014] In one possible implementation, the security network and the production network communicate based on an encryption protocol, which includes the Financial Secure Transport Protocol (FSTP).

[0015] In one possible implementation, when the security network extracts and stores monitoring video segments corresponding to the requirements of the video monitoring task by applying an edge computing algorithm from the monitoring video collected by the lower-level branch terminal equipment, the specific configuration is as follows:

[0016] Based on the video supervision task requirements corresponding to the video supervision task, and based on the mapping relationship between the video supervision task requirements and the monitoring probes in the lower-level branch terminal equipment, the target monitoring probes in the lower-level branch terminal equipment corresponding to the video supervision task requirements are determined.

[0017] An edge detection algorithm is applied to extract and store monitoring video segments that correspond to the requirements of the video supervision task from the monitoring videos collected by the target monitoring probe.

[0018] In one possible implementation, when the security network applies an edge detection algorithm to extract monitoring video segments corresponding to the video supervision task requirements from the monitoring videos collected by the target monitoring probes, the specific configuration is as follows:

[0019] Based on the scene features, time range, and action rules preset in the edge algorithm, the initial interception rules are determined;

[0020] Based on the dynamic personnel density in the real-time scene from the monitoring video collected from the target monitoring probe, and / or based on the environmental perception information of the target monitoring probe, optimize the initial interception rules;

[0021] Based on the optimized extraction rules, the monitoring videos collected by the target monitoring probes are filtered, and monitoring video segments that correspond to the requirements of the video supervision task are extracted.

[0022] In one possible implementation, the head office terminal equipment includes a head office control system, a head office imaging platform, and a head office data department. The head office control system is equipped with a video surveillance module, wherein:

[0023] The head office control system is configured to send video surveillance tasks to branch office devices;

[0024] The video monitoring module is configured to retrieve the corresponding target monitoring video clips from the branch terminal devices as needed, based on a list of video files.

[0025] The head office's imaging platform is configured to interact with the video supervision module, obtain target monitoring video segments from the video supervision module, assign video segment identifiers to the target monitoring video segments, and store the target monitoring video segments based on the video segment identifiers;

[0026] The head office data department is configured to interact with the head office image platform, retrieve target monitoring video segments based on video segment identifiers, and apply a large visual understanding model to perform compliance identification based on multimodal semantic analysis on the target monitoring video segments, obtaining and outputting the compliance identification results corresponding to the target monitoring video segments.

[0027] In one possible implementation, the head office data department is also configured as follows:

[0028] After outputting the compliance identification results corresponding to the target monitoring video clips, obtain the results of manual review;

[0029] Based on the results of manual review, the prompt words and sub-models of the visual understanding model are optimized and adjusted to generate an optimized visual understanding model.

[0030] Secondly, this application provides a remote video supervision method, applied to a branch office device in the remote video supervision system of the first aspect above. The remote video supervision method includes:

[0031] Upon receiving a video monitoring task from the head office terminal device, the system applies an edge algorithm to extract and store monitoring video segments corresponding to the video monitoring task requirements from the monitoring videos collected by the lower-level branch terminal devices, based on the video monitoring task requirements.

[0032] The monitored video clips are compiled into a list of video files, and the list of video files is transmitted to the head office terminal device;

[0033] Additionally, it provides target monitoring video clips to the head office terminal equipment. These target monitoring video clips are obtained on demand by the head office terminal equipment based on a list of video files. The head office terminal equipment is used to perform compliance identification based on multimodal semantic analysis on the target monitoring video clips, and obtain and output the compliance identification results corresponding to the target monitoring video clips.

[0034] Thirdly, this application also provides a remote video supervision method, applied to the head office device in the remote video supervision system of the first aspect above, the remote video supervision method comprising:

[0035] Send video surveillance tasks to branch office equipment;

[0036] Based on the video file list provided by the branch terminal device, the corresponding target monitoring video segments are obtained from the branch terminal device as needed. The video file list contains relevant information about the monitoring video segments. The monitoring video segments are extracted by the branch terminal device from the monitoring videos collected by the lower-level branch terminal device according to the video supervision task requirements corresponding to the video supervision task.

[0037] By applying a large-scale visual understanding model, compliance identification based on multimodal semantic analysis is performed on target monitoring video clips, and the compliance identification results corresponding to the target monitoring video clips are obtained and output.

[0038] Fourthly, this application provides an electronic device, including: a memory and a processor;

[0039] The memory stores the instructions that the computer executes;

[0040] The processor executes computer execution instructions stored in memory, causing the processor to perform various possible implementations of the second or third aspect above.

[0041] Fifthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a device such as a processor, are used to implement various possible embodiments of the second or third aspect above.

[0042] In a sixth aspect, this application provides a computer program product, including a computer program that, when executed by a device such as a processor, implements various possible implementations of the second or third aspect described above.

[0043] The remote video monitoring system, method, and electronic equipment provided in this application issue video monitoring tasks to branch office devices via head office equipment. Branch office devices automatically collect monitoring videos from their locations. Upon receiving the video monitoring task, the branch office devices, based on the requirements of the task, apply edge computing algorithms to extract and store monitoring video segments corresponding to the requirements from the monitoring videos collected by the lower-level branch office devices. This reduces the amount of video data to be processed by the head office equipment, improves the processing speed of quality inspection, and allows the extracted monitoring video segments to focus on the monitored scene, reducing invalid data and ensuring the accuracy of quality inspection from the data source level. The branch office devices automatically organize the extracted monitoring video segments into a video file list, forming a structured video index. This enables the head office equipment to quickly locate and call target video segments as needed, and transmits the video file list to the head office equipment, reducing manual video downloading and uploading. The head office terminal equipment retrieves corresponding target monitoring video segments from branch terminal equipment on demand based on a video file list, achieving fully automated workflow from task to data to result. This effectively improves the efficiency of data flow and quality inspection initiation. The head office terminal equipment applies a large-scale visual understanding model to perform compliance identification on the target monitoring video segments based on multimodal semantic analysis. This significantly shortens the quality inspection cycle while accurately identifying compliance issues in complex scenarios, reducing information gaps and misjudgments caused by single-modal analysis. Ultimately, it obtains and outputs the compliance identification results corresponding to the target monitoring video segments, improving the quality inspection efficiency and accuracy of the remote video supervision process. Attached Figure Description

[0044] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0045] Figure 1 A schematic diagram comparing the existing technology with the remote video supervision process provided in this application;

[0046] Figure 2 A schematic diagram of the structure of the remote video monitoring system provided in the embodiments of this application;

[0047] Figure 3 This is an application diagram of the remote video monitoring system provided in the embodiments of this application;

[0048] Figure 4 A flowchart illustrating the remote video supervision method provided in this application embodiment;

[0049] Figure 5 Another flowchart illustrating the remote video supervision method provided in this application embodiment;

[0050] Figure 6This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0051] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0052] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0053] Figure 1 A schematic diagram comparing the existing technology with the remote video supervision process provided in this application. Figure 1 As shown, in the existing remote video monitoring process, during the video acquisition phase, the head office issues a task list to the branch office through the management system. The branch office copies the monitoring videos from the security system or an outsourced company and manually uploads them to the cloud drive. During this process, the branch office also needs to supplement the videos. During the video quality inspection phase, the head office's quality inspectors manually download the monitoring videos provided by the branch office from the cloud drive and perform manual quality inspection on each video. During the results feedback phase, if quality issues are found, the branch office is notified by phone, email, or the system to verify and rectify the issues. If no quality issues are found, the process ends.

[0054] To overcome the limitations of existing remote video supervision processes, this application provides a remote video supervision system that, through automated interaction and processing between head office equipment, branch office equipment, and sub-branch equipment, breaks down the entire process of video acquisition, transmission, and quality inspection into modules that can be automated, thereby achieving full automation of remote video supervision.

[0055] Specifically, still as Figure 1As shown, during the video acquisition phase, the head office equipment sends a video supervision task list to the branch office equipment. The branch office equipment then automates the flow of monitoring videos through video stream transmission and storage processing, optimizing the video supervision process. During the video quality inspection phase, the semantic analysis capabilities of a large-scale visual understanding model are used to automatically inspect the compliance of monitoring video segments collected at the branch level and processed at the branch level (70% model output results, 30% manual review). In the result feedback phase, if quality issues are found, the branch office equipment is notified to verify and rectify them. If no quality issues are found, the process ends. This remote video supervision process improves the efficiency and accuracy of quality inspection in remote video supervision.

[0056] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0057] Figure 2 This is a schematic diagram of the structure of a remote video monitoring system provided in an embodiment of this application. Figure 2 As shown, the remote video monitoring system 10 includes: a head office terminal device 101, a branch office terminal device 102, and a sub-branch terminal device 103, wherein:

[0058] Head office terminal device 101 is configured to send video monitoring tasks to branch terminal device 103.

[0059] For example, the head office terminal device 101 can automatically generate structured video supervision tasks based on the unified business rules and task scheduling engine, according to the supervision needs of different branches and different security scenarios (such as vault guarding, branch security, autonomous equipment inspection, etc.), and send the video supervision tasks to the corresponding branch terminal device 103 through a secure and encrypted internal communication link.

[0060] Optionally, the video supervision task may include task parameter information such as task type, supervision time period, target branch outlet, video acquisition requirements, key behavior types and task priority, so as to ensure that the branch terminal device 103 can accurately receive and execute the video supervision task requirements issued by the head office terminal device 101, laying the foundation for subsequent automated video acquisition and intelligent quality inspection.

[0061] Branch terminal device 102 is configured to collect monitoring videos of branch outlets.

[0062] For example, the branch terminal device 102 can be deployed in key security areas such as branch business premises, vaults, and self-service equipment areas. Through multiple high-definition monitoring probes, it can collect monitoring videos covering all dimensions of the branch in real time. The frame rate, resolution, and monitoring angle can be flexibly adjusted according to different scenario requirements to ensure complete recording of key images such as daily branch operations, personnel operations, equipment status, and security incidents.

[0063] Optionally, the monitoring video of the branch outlet collected by the branch terminal device 102 is first stored in a local multimedia terminal (such as a hard disk / terminal device) and then transmitted to the branch terminal device 103.

[0064] Branch terminal device 103 is configured to respond to a video supervision task, and according to the video supervision task requirements, apply an edge algorithm to extract and store monitoring video segments corresponding to the video supervision task requirements from the monitoring videos collected by the lower-level branch terminal devices; and organize the monitoring video segments into a video file list and transmit the video file list to the head office terminal device.

[0065] For example, after obtaining the video supervision task issued by the head office terminal device 101 through the secure communication link between the branch terminal device 103 and the head office terminal device 101, the branch terminal device 103 needs to first parse the video supervision task requirements corresponding to the video supervision task and clarify the core instructions such as the scope of the branch outlets to be supervised, the monitoring scenario, the time interval, and the behavior recognition standards.

[0066] Furthermore, by applying locally deployed edge algorithms, the monitoring videos collected in real time by the lower-level branch terminal devices and transmitted to the branch terminal device 103 are subjected to targeted and intelligent analysis and processing. The monitoring video segments that match the requirements of the video supervision task are automatically extracted and stored, and redundant video content with no supervision value is eliminated.

[0067] Meanwhile, the branch-level device 103 can be a hardware server and / or edge box deployed in the branch, each equipped with data transmission application software. This software standardizes and organizes the captured and stored monitoring video clips, generating a structured list of video files according to unified naming rules, classification dimensions, and data formats. This list is then transmitted to the head office-level device 101 in the form of a message, enabling cross-environment transmission. For example, this video file list includes key information such as the branch affiliation, monitoring scenario, timestamp, and file storage path for each monitoring video clip.

[0068] The head office terminal device 101 is also configured to obtain the corresponding target monitoring video segments from the branch terminal devices on demand based on the video file list; and to apply a large visual understanding model to perform compliance identification on the target monitoring video segments based on multimodal semantic analysis, and obtain and output the compliance identification results corresponding to the target monitoring video segments.

[0069] For example, after receiving a list of video files in the form of a message uploaded by the branch terminal device 103, the head office terminal device 101 parses the key information contained in the video file list, such as the branch affiliation, monitoring scene, timestamp, and file storage path corresponding to each monitoring video segment, and obtains the target monitoring video segment from the branch terminal device 103 on demand.

[0070] Furthermore, by applying a large-scale visual understanding model, a compliance identification process based on multimodal semantic analysis is executed on the acquired target monitoring video clips. This process comprehensively analyzes multi-dimensional information such as visual features, temporal actions, and scene environment in the video footage to identify compliant operations and violations during branch operations. Simultaneously, the identification results are structured to generate and output the compliance identification result corresponding to the target monitoring video clip. For example, this compliance identification result may include key information such as the type of violation, the time period of occurrence, the corresponding scene, and the identification confidence level.

[0071] For example, a large model of visual understanding can be obtained in the following ways:

[0072] Based on a mature, general-purpose visual understanding model in the industry, we collected a massive dataset of compliance / violation monitoring videos from scenarios such as bank branch vaults, business premises, and self-service equipment areas. Combining the compliance rules of video supervision tasks, we fine-tuned and optimized the general basic model for specific scenarios, optimized the model's loss function and recognition threshold, and made the model adaptable to the visual characteristics and compliance judgment requirements of bank security videos, thereby improving the recognition accuracy of specific scenarios.

[0073] It is important to note that this large-scale visual understanding model supports continuous iteration and optimization based on new scenarios and compliance rules in actual supervision operations, in order to improve the model's long-term adaptability.

[0074] The remote video supervision system provided in this application sends video supervision tasks from the head office terminal device to the branch terminal devices, and the branch terminal devices automatically collect monitoring videos of the branch outlets. Upon receiving the video supervision task, the branch terminal devices, based on the requirements of the video supervision task, apply edge computing algorithms to extract and store monitoring video segments corresponding to the requirements of the video supervision task from the monitoring videos collected by the lower-level branch terminal devices. This reduces the amount of video data to be processed by the head office terminal device, improves the processing speed of quality inspection, and allows the extracted monitoring video segments to focus on the supervised scene, reducing invalid data and ensuring the accuracy of quality inspection from the data source level. The branch terminal devices automatically organize the extracted monitoring video segments into a video file list, forming a structured video index, enabling the head office terminal device to quickly locate and call target video segments on demand, and transmit the video file list to the head office terminal device, reducing manual downloading and uploading of videos. Based on the video file list, the head office terminal device obtains the corresponding target monitoring video segments from the branch terminal devices on demand, realizing the fully automated flow of task-data-result, which can effectively improve the efficiency of data flow and quality inspection initiation. The head office terminal equipment uses a large visual understanding model to perform compliance identification on target monitoring video clips based on multimodal semantic analysis. This can significantly shorten the quality inspection cycle, accurately identify compliance issues in complex scenarios, reduce information loss and misjudgment caused by single-modal analysis, and finally obtain and output the compliance identification results corresponding to the target monitoring video clips, thereby improving the quality inspection efficiency and accuracy of the remote video supervision process.

[0075] Based on the above embodiments, the branch-level terminal equipment may include a security network and a production network. The security network is configured to respond to a received video surveillance task, and according to the video surveillance task requirements, apply an edge computing algorithm to extract and store monitoring video segments corresponding to the video surveillance task requirements from the monitoring videos collected by the lower-level branch-level terminal equipment; and to organize the monitoring video segments into a video file list and transmit the video file list to the head office terminal equipment. The production network is configured to interact with the head office terminal equipment, receive the head office terminal equipment's request to acquire target monitoring video segments; and to interact with the security network, acquire target monitoring video segments from the security network, and send the target monitoring video segments to the head office terminal equipment.

[0076] It should be noted that the functions configured in the security network correspond to those configured in the branch terminal device 103 mentioned above, and will not be elaborated further here.

[0077] Optionally, to ensure the secure, accurate, and compliant transfer of monitoring video data between branch-level and head office-level devices, the production network within the branch-level devices can be configured to enable cross-network data relay and interaction with the head office-level devices. For example, the bidirectional interaction logic between the production network, the head office-level devices, and the security network is as follows:

[0078] On the one hand, the production network establishes a compliant communication link with the head office terminal equipment. Through this communication link, it receives in real time the target monitoring video segment acquisition requests issued by the head office terminal equipment based on the video file list. For example, the acquisition request can be represented as an on-demand acquisition instruction for the target monitoring video segment.

[0079] On the other hand, the production network and the security network interact securely. Based on the target monitoring video segment acquisition requirements received from the head office terminal equipment, the target monitoring video segment corresponding to the acquisition requirements is extracted from the security network, and then the target monitoring video segment is sent to the head office terminal equipment to complete the cross-network compliant flow of video data.

[0080] The security network enables professional processing and secure storage of monitoring videos, while the production network facilitates standardized data interaction with the head office equipment. Through the coordinated operation of these two networks, the timely and accurate acquisition of video data by the head office equipment is ensured, providing core network support for the safe and efficient operation of the remote video surveillance system.

[0081] Optionally, the security network and the production network can communicate based on an encryption protocol, which may include the Financial Secure Transfer Protocol (FSTP).

[0082] To further ensure the security and compliance of cross-network data interaction within branch terminal equipment, a secure communication link is established between the security network and the production network based on a dedicated encryption protocol to achieve encrypted transmission and interaction of monitoring video-related data.

[0083] For example, the encryption protocol can include a secure file transfer protocol specific to the financial industry, such as FSTP. FSTP, based on general secure file transfer protocols, strengthens the compliance and traceability design of data transmission for financial security scenarios. It can accurately adapt to the transmission requirements of video data between security networks and production networks, ensuring that cross-network data exchanges are traceable, verifiable, and auditable.

[0084] This encrypted communication design enables functional collaboration and data interoperability between the security network and the production network, while also adhering to the security management standards for financial security data. This ensures that cross-network data transfer within the branch's equipment is both efficient and secure, thus building a solid security transmission defense line within the branch's equipment for the subsequent transmission of target monitoring video clips to the head office.

[0085] Furthermore, when the security network extracts and stores monitoring video segments corresponding to the video supervision task requirements based on the video supervision task requirements and applying edge computing algorithms from the monitoring videos collected by the lower-level branch terminal devices, the specific configuration is as follows: Based on the video supervision task requirements and the mapping relationship between the video supervision task requirements and the monitoring probes in the lower-level branch terminal devices, the target monitoring probe in the lower-level branch terminal devices corresponding to the video supervision task requirements is determined. The edge computing algorithm is then applied to extract and store the monitoring video segments corresponding to the video supervision task requirements from the monitoring videos collected by the target monitoring probe.

[0086] For example, after acquiring a video surveillance task, the security network analyzes the video surveillance task requirements corresponding to the task, clarifies the core instructions such as the scope of the branch outlets to be supervised, the monitoring scenario, the time interval, and the behavior recognition standards, and, based on the mapping relationship between the video surveillance task requirements and the monitoring probes in the lower-level branch terminal equipment, completes precise matching and screening according to this mapping relationship to determine the target monitoring probes in the lower-level branch terminal equipment that match the video surveillance task requirements.

[0087] Optionally, this mapping relationship is a standardized association rule pre-configured within the bank, which accurately binds different supervision scenarios and supervision objects to the probe numbers, probe coverage areas, and probe acquisition attributes of the corresponding monitoring areas at the branch end. The security network can determine the target monitoring probes based on this rule, lock the precise range of video acquisition from the source, and reduce interference from probe video data with no supervision value.

[0088] Furthermore, the security network invokes locally deployed edge algorithms to perform targeted and intelligent analysis and processing only on the monitoring videos collected in real time by the identified target monitoring probes, extracting monitoring video segments that match the requirements of the video supervision task from the monitoring videos of the target monitoring probes.

[0089] Optionally, the captured monitoring video clips can be standardized to unify the video format, encoding method, and naming rules, and the standardized monitoring video clips can be stored locally on the security network.

[0090] By identifying target monitoring probes and extracting monitoring video segments that correspond to the requirements of video supervision tasks, a structured, high-fidelity, secure, and traceable video data source is provided for subsequent video segment organization and production network retrieval. This achieves accuracy and efficiency in video extraction and reduces the processing and storage costs of redundant video data.

[0091] In some embodiments, when the security network applies edge detection algorithms to extract monitoring video segments corresponding to the requirements of the video supervision task from the monitoring videos collected by the target monitoring probes, the specific configuration is as follows: Initial extraction rules are determined based on preset scene features, time ranges, and action rules in the edge detection algorithm. The initial extraction rules are optimized based on the real-time dynamics of personnel density in the monitoring videos collected by the target monitoring probes, and / or based on the environmental perception information of the target monitoring probes. Based on the optimized extraction rules, the monitoring videos collected by the target monitoring probes are filtered, and monitoring video segments corresponding to the requirements of the video supervision task are extracted.

[0092] For example, by applying edge computing algorithms and based on the core judgment criteria such as scene features, time ranges, and action rules preset in the edge computing algorithms, initial capture rules matching the requirements of the video supervision task are determined. Specifically, the scene feature rules clarify the visual boundaries, core monitoring areas, and scene-specific recognition features of each supervision scene; the time range rules lock in the supervision period required by the task, key security time nodes, and regular monitoring time intervals; and the action rules define the triggering judgment conditions such as compliant operations, violations, abnormal equipment status, and abnormal personnel actions that need to be captured. It is important to note that all of the above rules are set according to the requirements of the video supervision task.

[0093] Optionally, the core execution parameters in the initial cropping rules may include the basic cropping frequency, single cropping duration, trigger-based cropping threshold, and video segment integration specifications.

[0094] Furthermore, during the continuous acquisition of monitoring video by the target monitoring probes, the security network uses edge computing algorithms to analyze the video footage in real time, identifying and extracting dynamic personnel density within the scene, including key indicators such as the number of people per unit frame, personnel flow rate, degree of personnel concentration in key areas, and duration of personnel stay. Simultaneously, it can acquire environmental perception information corresponding to the target monitoring probes, including environmental dimensions such as light intensity, temperature and humidity changes, sound decibels, and equipment operating status in the monitored area.

[0095] Optionally, a comprehensive analysis and gradient determination of personnel density dynamics and / or environmental perception information can be performed, and the core execution parameters of the initial capture rule can be dynamically optimized based on the determination results. For example, when high-density personnel gathering or a sudden drop in light intensity affects image recognition, the capture frequency can be automatically increased, the trigger capture sensitivity can be improved, and the single capture duration can be shortened to ensure the capture of critical images. When personnel density is low and stable and the environmental conditions are normal, the capture frequency can be appropriately reduced, the trigger threshold can be relaxed, and redundant capture segments can be integrated to reduce the generation of invalid data.

[0096] Furthermore, the monitoring videos collected by the target monitoring probes are screened according to the optimized extraction rules. This includes: completing routine video segment extraction at the frequency set by the optimized extraction rules; and simultaneously, based on the adjusted trigger thresholds in the rules, triggering the extraction of key behaviors, equipment anomalies, scene anomalies, and other content that meet the requirements of the video supervision task in real time. This ensures the accuracy and effectiveness of the screening process and extracts monitoring video segments that highly match the requirements of the video supervision task.

[0097] The initial cropping rules are determined by pre-defined rules in the edge algorithm, ensuring that the cropping behavior conforms to the requirements of the video supervision task from the source. The initial cropping rules are optimized by real-time scene data such as personnel density and environmental perception, which can realize the dynamic adaptation of the cropping rules to the real-time monitoring scene and improve the efficiency and accuracy of video cropping.

[0098] As one possible implementation, the head office-side equipment includes a head office management and control system, a head office imaging platform, and a head office data department. The head office management and control system deploys a video monitoring module, which is configured to: issue video monitoring tasks to branch office-side equipment; the video monitoring module, configured to retrieve corresponding target monitoring video segments from branch office-side equipment as needed based on a list of video files; the head office imaging platform, configured to interact with the video monitoring module, retrieve target monitoring video segments from the module, assign video segment identifiers to the target monitoring video segments, and store the target monitoring video segments based on the video segment identifiers; and the head office data department, configured to interact with the head office imaging platform, retrieve target monitoring video segments based on video segment identifiers, and apply a large-scale visual understanding model to perform compliance identification based on multimodal semantic analysis of the target monitoring video segments, obtaining and outputting the compliance identification results corresponding to the target monitoring video segments.

[0099] For example, the head office control system issues video surveillance tasks to branch office devices via a secure communication link. The video surveillance module deployed within the head office control system interacts with the branch office devices, sending a request to the branch office devices to acquire the target video segments based on the list of video files transmitted by the branch office devices. Based on this request, the module retrieves the corresponding target video segments from the branch office devices. It can be understood that the video surveillance module refers to an independently developed functional module / subsystem within the head office control system specifically designed to handle the entire video surveillance process.

[0100] Optionally, the video monitoring module interacts with the head office's imaging platform via a dedicated data transmission interface. The head office's imaging platform obtains target monitoring video clips from the video monitoring module, assigns a video clip identifier (imageID) to each received target monitoring video clip, and stores the target monitoring video clips based on the video clip identifier. It can be understood that the head office's imaging platform refers to the bank's unified, compliant, and secure centralized storage platform for unstructured data (videos / images).

[0101] In some embodiments, the head office data department interacts with the head office image platform to retrieve target monitoring video segments based on video segment identifiers, and applies a large visual understanding model to perform full-dimensional analysis of the visual features, temporal actions, and scene environment of the target monitoring video segments. This can accurately identify key subjects, core behaviors, scene features, and equipment status in the video, achieving structured visual understanding of the video content.

[0102] Furthermore, the visual recognition results are semantically matched with compliance rules in the video supervision task requirements (such as "personnel are prohibited from entering the vault during off-duty hours"). Multimodal fusion technology is used to transform visual features into semantic information that can be interfaced with compliance rules, determining whether the behavior and state in the monitored video clips comply with the video supervision task requirements, clearly identifying compliant and non-compliant scenarios, and labeling non-compliant scenarios with types such as unauthorized personnel entry, improper operation, and abnormal equipment status. The compliance determination results are then standardized and structured to generate compliance identification results containing core content such as non-compliance type, time of non-compliance, corresponding monitoring scenario, identification confidence level, and video clip association information.

[0103] Through the specialized division of labor and efficient collaboration among the head office control system, video supervision module, head office imaging platform, and head office data department, the head office terminal equipment achieves closed-loop management of the entire process, including video supervision task distribution, on-demand acquisition of target videos, standardized storage, and large-scale multimodal semantic compliance recognition. This not only improves the refinement, standardization, and intelligence of video data processing, but also ensures accurate traceability and efficient acquisition of video data through unique identifier binding, further enhancing the quality inspection efficiency and compliance identification accuracy of remote video supervision.

[0104] In one possible implementation, the head office data department is further configured to: after outputting the compliance identification results corresponding to the target monitoring video clips, obtain the results of manual review. Based on the results of manual review, optimize and adjust the prompt words and sub-models of the visual understanding big model to generate an optimized visual understanding big model.

[0105] For example, after the Head Office Data Department outputs the compliance identification results corresponding to the target monitoring video clips, it can feed these results back to the video supervision module for video supervisors to review. The video supervisors, drawing on their professional supervisory and review experience, manually review the compliance identification results, obtaining a manual review result. The Head Office Data Department can obtain this manual review result by interacting with the video supervision module. The video supervisors are dedicated personnel responsible for frontline video supervision work.

[0106] Optionally, the manual review results include core information such as the correctness of the compliance identification results, the specific scenarios and reasons for misjudgment / omission, the accurate labeling of violations, and suggestions for correcting the identification rules, while also associating the identifiers and image features of the corresponding target monitoring video segments.

[0107] Furthermore, the head office's data department systematically decomposed and analyzed the results of manual review, extracted the deviations and optimization directions of the model recognition, and carried out targeted dual optimization and adjustment of the prompt words and sub-models of the visual understanding big model to generate an optimized visual understanding big model.

[0108] Specifically, at the prompt word level, to address the misjudgment / missed judgment issues caused by vague prompt word descriptions, missing scenario definitions, and inaccurate behavior descriptions discovered during manual review, the core information of the prompt words is supplemented and improved, and the expression logic is optimized, enabling the model to more accurately understand the business requirements for compliance judgment. For example, in the "seal management" scenario, which is a key area of ​​supervision by banks, if manual review finds that the model failed to identify the single-person operation violation in the seal loading and unloading process due to the prompt words not clearly defining key operational requirements, the specific behavior description of "two-person loading and unloading" will be added to the prompt words corresponding to that scenario. At the same time, the compliance judgment standard for this operation will be clarified, enabling the model to accurately capture the core compliance actions in this scenario and reduce the recognition bias caused by missing prompt word information.

[0109] At the sub-model level, recognizing the significant differences in business characteristics, compliance requirements, and identification priorities across various scenarios in remote video supervision of banks, the head office data department will optimize and upgrade the models based on manual review results, breaking away from generic identification logic. Independent, dedicated sub-models will be constructed for different supervision scenarios such as warehouse management, employee behavior, vault security, seal management, and counter operations. Furthermore, the action recognition logic of each sub-model will be individually optimized to address identification issues identified during manual review. For example, for the "warehouse management" scenario, the focus will be on optimizing the timing analysis and feature recognition logic for actions such as material storage and retrieval, vault door opening and closing, and personnel entry and exit, adapting to the operational procedures and compliance requirements of the warehouse scenario.

[0110] Optionally, by iteratively optimizing the prompt words and selectively tuning the sub-models, an optimized visual understanding model is generated and then reapplied to the head office data department for compliance identification.

[0111] By manually reviewing the results and optimizing the prompts and sub-models of the visual understanding model, we can accurately compensate for model recognition errors, enable the model to accurately understand the business requirements of compliance judgment, improve the recognition accuracy in various subdivided supervision scenarios, and further improve the recognition accuracy in complex scenarios.

[0112] It is important to note that in the remote video supervision system provided in this application embodiment, while leveraging the advantages of the large-scale visual understanding model for efficient batch processing, it is necessary to control the recognition accuracy and supervision quality through manual review. In other words, the system should prioritize intelligence and supplement it with manual review, maintaining a balance between efficiency and accuracy in remote video supervision. For example, the direct output results from the large-scale visual understanding model may account for 70%, while the results from manual review may account for 30%.

[0113] Figure 3 This is a schematic diagram illustrating the application of the remote video monitoring system provided in an embodiment of this application. For example... Figure 3 As shown, the branch security monitoring / server terminal in the branch (corresponding to the branch-end device) obtains the locally stored monitoring video from the multimedia terminal of the sub-branch (corresponding to the sub-branch-end device) through the security video server, obtains the corresponding video stream, and transmits the video stream and the video stream obtained from the sub-branch vault camera together to the server / edge box (corresponding to the security network) deployed in the branch. Among them, the locally stored monitoring video in the sub-branch multimedia terminal is obtained by the sub-branch camera.

[0114] Within the server / edge box, data transmission application software (for transmitting video file lists) is developed, and edge algorithms (for extracting video clips) are deployed. A built-in mapping relationship between video supervision tasks and monitoring probes is also included (for matching target monitoring probes). The server / edge box interacts with the production network (corresponding to the production network) through an address mapping device. The server / edge box transmits the video file list in message form via the address mapping device, while the production network retrieves videos from the server / edge box via the address mapping device. The head office (corresponding to the head office terminal device) selects and queries the file list through the head office management system backend (corresponding to the video supervision module), obtains the corresponding videos through the production network, and stores the videos in the head office imaging system backend (corresponding to the head office imaging platform). The head office imaging system backend obtains and outputs the compliance identification results corresponding to the target monitoring video clips through the big data and big model server (corresponding to the head office data department), and feeds these compliance identification results back to the head office management system backend for manual review.

[0115] Table 1 illustrates the transformation ideas, construction difficulties, and solutions of the video surveillance system provided in the embodiments of this application at different business stages and for different business situations.

[0116] Table 1

[0117]

[0118]

[0119] It is important to note that the review roles of the visual understanding big data model and manual review differ depending on the video supervision scenario. For example, the video supervision scenarios supported by the visual understanding big data model provided in this application include: warehouse management, seal management, cash handling machine cash replenishment, performance supervision, and employee daily behavior.

[0120] Table 2

[0121]

[0122] Table 2 shows the supervision subcategories, supervision scenarios, and intelligent supervision points corresponding to the video supervision scenarios for warehouse management.

[0123] Table 3

[0124]

[0125] For warehouse management scenarios, the overall testing results are as follows: most warehouse management scenarios involve people recognition, while some scenarios (handover and inspection) include action recognition. The large model test results can generally provide the time points of people's appearance, actions, and departure. However, due to situations such as actions being too fast, small in amplitude, or occlusion in some scenarios, the large model's action recognition effect is not good and cannot completely replace manual review. The model can be used for preliminary recognition first to assist manual judgment.

[0126] Table 4

[0127]

[0128] Table 3 shows the supervision subcategories, supervision scenarios, and intelligent supervision points corresponding to the video supervision scenarios related to seal management. For seal management scenarios, the overall testing results are as follows: most seal management scenarios involve personnel action recognition at workstations. Some workstation scenarios are more complex with many redundant elements, making it difficult for the large model to accurately identify the use of seals and the loading and unloading of seals by business personnel in complex scenarios. Furthermore, the performance in detecting the collection and storage of desktop paper documents and computer lock screens is also poor. With adjustments to the descriptions of objects such as the seal control machine and seals in the prompts, the accuracy of the large model's output results has significantly improved, with improvements in accuracy at time points and in the precision of personnel action recognition. The large model can be used for initial review.

[0129] Table 5

[0130]

[0131] Table 4 shows the supervision subcategories, supervision scenarios, and intelligent supervision points corresponding to video supervision scenarios related to cash handling machine clearing and cash replenishment. For cash handling machine clearing and cash replenishment, the overall scenario testing results are as follows: These mainly include the entire clearing and cash replenishment process, door / corridor opening and closing scenarios, and cash sorting scenarios, with a generally high recognition accuracy. Specifically, during the entire clearing and cash replenishment process, the cash jamming scenario requires description of the personnel's actions during the jamming to optimize the prompts; the door opening and closing scenario includes key turning and card / facial recognition actions, which are relatively small in amplitude and partially obstructed in the monitoring video, making them difficult to recognize; the large model's recognition rate for the entire cash counting process is high, close to the results of manual analysis.

[0132] Table 5 shows the supervision subcategories, scenarios, and intelligent supervision points corresponding to video supervision of performance of duties. For performance supervision scenarios, most involve the recognition of human actions, such as the actions of a person inspecting items, and the actions of those accompanying them. The large-scale model's recognition results are relatively accurate, and most scenarios are operationally usable. However, considering that some performance supervision scenarios require verification of personnel identity, but the large-scale model cannot determine identity based on personnel actions, in such scenarios, it is advisable to have the large-scale model generate an auxiliary review report including human actions, which can then be submitted manually as an attachment, thus shortening the time required to review the monitoring videos.

[0133] Table 6 shows the supervision subcategories, supervision scenarios, and intelligent supervision points corresponding to video supervision scenarios of employee daily behavior. For the employee daily behavior category, the overall test results are as follows: The employee daily behavior category mainly involves recognizing the actions of staff (actions to verify the identity of escort personnel, actions to check whether the tail box is locked). The actions of the staff need to be described in detail in the prompt words. The model can first perform preliminary recognition and then use it to assist manual judgment.

[0134] Table 6

[0135]

[0136] In summary, the remote video monitoring system provided in this application has at least the following advantages:

[0137] First, video surveillance tasks are distributed from the head office equipment to the branch equipment, and the branch equipment automatically collects monitoring videos from the branch outlets. The security network within the branch equipment, based on the requirements of the video surveillance task, applies edge computing algorithms to determine the target monitoring probes in the lower-level branch equipment that correspond to the task requirements. This reduces the processing and storage costs of redundant video data, ensuring the accuracy and efficiency of video capture and guaranteeing quality inspection accuracy from the data source level. The initial capture rules for the corresponding video surveillance task requirements are optimized using real-time scene data such as personnel density and environmental perception, achieving dynamic adaptation between the capture rules and the real-time monitoring scene. Based on the optimized capture rules, corresponding monitoring video segments are selected and captured to improve the efficiency and accuracy of video capture.

[0138] Secondly, the security network within the branch office equipment organizes the captured monitoring video clips into a video file list, enabling the head office equipment to quickly locate and retrieve target video clips on demand. The video file list is then transmitted to the head office equipment, reducing manual video downloading and uploading. The production network within the branch office equipment enables standardized data exchange with both the head office equipment and the security network. This dual-network collaboration ensures the timeliness and accuracy of the head office equipment's on-demand acquisition of video data. The security network and production network communicate using encrypted protocols, making cross-network data transfer within the branch office equipment both efficient and secure.

[0139] Third, the head office terminal equipment, through the specialized division of labor and efficient collaboration among the head office control system, video supervision module, head office imaging platform, and head office data department, achieves closed-loop management of the entire process, including video supervision task distribution, on-demand acquisition of target videos, standardized storage, and large-scale model multimodal semantic compliance recognition. This configuration not only enhances the refinement, standardization, and intelligence of video data processing but also ensures accurate traceability and efficient acquisition of video data through unique identifier binding, further improving the quality inspection efficiency and compliance identification accuracy of remote video supervision.

[0140] Fourth, the head office's terminal equipment utilizes a large-scale visual understanding model to perform compliance identification on target monitoring video clips based on multimodal semantic analysis. This significantly shortens the quality inspection cycle, reduces manpower, and accurately identifies compliance issues in complex scenarios. It also reduces information gaps and misjudgments caused by single-modal analysis, enabling the management and supervision of a wider range of videos. The resulting compliance identification improves the efficiency and accuracy of the remote video supervision process. By manually reviewing the results and optimizing the prompts and sub-models of the large-scale visual understanding model, model recognition errors can be accurately compensated for, enabling the model to accurately understand the business requirements of compliance judgments and improve recognition accuracy in various subdivided supervision scenarios, further enhancing the recognition accuracy in complex scenarios.

[0141] In summary, the remote video supervision system provided in this application embodiment can effectively improve the quality inspection efficiency and accuracy of the remote video supervision process.

[0142] Figure 4 This is a flowchart illustrating a remote video monitoring method provided in an embodiment of this application, applied to a branch office device in the aforementioned remote video monitoring system. For example... Figure 4 As shown, the remote video supervision method includes:

[0143] S201. Upon receiving the video supervision task issued by the head office terminal device, and based on the video supervision task requirements, apply the edge algorithm to extract and store the monitoring video segments corresponding to the video supervision task requirements from the monitoring videos collected by the lower-level branch terminal devices.

[0144] For example, after receiving a video supervision task from the head office, the task is parsed to obtain the corresponding video supervision task requirements. Based on the parsed requirements, an edge computing algorithm is applied. Leveraging the edge computing algorithm's capabilities for scene feature recognition and temporal action analysis, the monitoring videos collected by the lower-level branch's equipment are targeted and precisely analyzed. The monitoring video segments corresponding to the video supervision task requirements are then extracted and stored.

[0145] S202. Organize the monitored video clips into a video file list and transmit the video file list to the head office terminal device.

[0146] For example, the stored monitoring video clips are categorized and structured, and the completeness, validity, and compliance of each clip are verified one by one. Redundant clips, including duplicates, damaged clips, or those without supervisory value, are removed, and a standardized list of video files is generated. This list is then transmitted to the central office equipment. The video file list includes information such as the unique identifier, storage path, capture time, associated probe, and task matching degree for each monitoring video clip.

[0147] S203. Provide target monitoring video clips to the head office terminal equipment. The target monitoring video clips are obtained on demand by the head office terminal equipment based on the video file list. The head office terminal equipment is used to perform compliance identification based on multimodal semantic analysis on the target monitoring video clips, and obtain and output the compliance identification results corresponding to the target monitoring video clips.

[0148] For example, the branch-level equipment, acting as the core of the front-end data supply, specifically provides the head office-level equipment with target monitoring video clips that meet the requirements of the video supervision task. However, these target monitoring video clips are not acquired indiscriminately from all sources. Instead, the head office-level equipment initiates an on-demand acquisition request based on a list of video files already transmitted by the branch-level equipment, combined with the bank-wide video supervision review priorities, computing resource scheduling, and key supervision needs. The branch-level equipment then precisely selects and transmits these video monitoring clips from its locally stored archives.

[0149] Optionally, the head office terminal device is used to apply a large visual understanding model to perform compliance identification based on multimodal semantic analysis on target monitoring video segments acquired on demand. This includes parsing multi-dimensional semantic content in the video frame, matching and verifying the parsing results with the compliance judgment rules preset by the video supervision task, and identifying whether there are any violations in the video segments. Finally, the compliance identification results corresponding to the target monitoring video segments are obtained and output.

[0150] Figure 5 This is another flowchart illustrating the remote video monitoring method provided in this application embodiment, applied to the head office device in the aforementioned remote video monitoring system. For example... Figure 5 As shown, the remote video supervision method includes:

[0151] S301, Issue video monitoring tasks to branch terminal equipment.

[0152] For example, based on the supervision needs of different branches and different security scenarios, video supervision tasks are automatically generated and distributed to branch-level devices. These distributed video supervision tasks can include a series of clear and detailed task requirements, such as the scope of the branches to be supervised, core security monitoring scenarios, specified supervision time intervals, compliance judgment standards, key monitoring behaviors, and video capture and storage specifications.

[0153] S302. Based on the video file list provided by the branch terminal device, obtain the corresponding target monitoring video segments from the branch terminal device as needed. The video file list contains relevant information about the monitoring video segments. The monitoring video segments are extracted by the branch terminal device from the monitoring videos collected by the lower-level branch terminal device according to the video supervision task requirements corresponding to the video supervision task.

[0154] For example, based on the video file list provided by the branch terminal, a targeted video retrieval command is sent to the branch terminal device as needed, thereby retrieving the corresponding target monitoring video segment from the branch terminal device as required. The video file list contains relevant information about the monitoring video segment, such as a unique identifier, storage path, retrieval time, the branch, and the corresponding monitoring probe, to achieve accurate mapping between the segment and the specific monitoring video segment.

[0155] Optionally, the monitoring video can be obtained by the branch terminal device after receiving the video supervision task issued by the head office terminal device, and by using edge algorithms to perform targeted filtering, intelligent analysis and precise extraction of the monitoring video collected by the lower-level branch terminal device according to the video supervision task requirements corresponding to the video supervision task.

[0156] S303. Apply a large visual understanding model to perform compliance identification on target monitoring video segments based on multimodal semantic analysis, and obtain and output the compliance identification results corresponding to the target monitoring video segments.

[0157] For example, after acquiring the target monitoring video clip, a large-scale visual understanding model optimized for financial security supervision scenarios is applied to perform compliance identification based on multimodal semantic analysis on the target monitoring clip. Specifically, the model comprehensively analyzes multi-dimensional information such as visual features, temporal actions, and scene environment in the video footage to identify compliant operations and violations during the branch's operation. Simultaneously, the identification results are structured to ultimately generate and output the compliance identification result corresponding to the target monitoring video clip.

[0158] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device 40 provided in this application embodiment includes at least one processor 401 and a memory 402. Optionally, the electronic device 40 further includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus.

[0159] In a specific implementation, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to perform the above-described method.

[0160] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0161] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0162] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0163] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0164] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0165] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0166] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0167] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0168] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0169] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0170] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0171] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0172] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0173] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A remote video monitoring system, characterized in that, include: Head office equipment, branch office equipment, and sub-branch equipment, including: The head office terminal device is configured to issue video supervision tasks to the branch terminal device; The branch terminal device is configured to collect monitoring videos of the branch outlets. The branch terminal device is configured to respond to the acquisition of the video supervision task, and according to the video supervision task requirements corresponding to the video supervision task, apply an edge algorithm to extract and store monitoring video segments corresponding to the video supervision task requirements from the monitoring videos collected by the lower-level branch terminal devices; and organize the monitoring video segments into a video file list and transmit the video file list to the head office terminal device. The head office terminal device is also configured to obtain the corresponding target monitoring video segments from the branch terminal devices as needed based on the video file list; and to apply a large visual understanding model to perform compliance identification based on multimodal semantic analysis on the target monitoring video segments, and to obtain and output the compliance identification result corresponding to the target monitoring video segments.

2. The remote video monitoring system according to claim 1, characterized in that, The branch office equipment includes: a security network and a production network, wherein: The security network is configured to respond to the acquisition of the video supervision task, and according to the video supervision task requirements corresponding to the video supervision task, apply an edge algorithm to extract and store monitoring video segments corresponding to the video supervision task requirements from the monitoring videos collected by the lower-level branch terminal devices; and organize the monitoring video segments into a video file list and transmit the video file list to the head office terminal device. The production network is configured to interact with the head office terminal device to receive the head office terminal device's request to acquire the target monitoring video clip; and to interact with the security network to acquire the target monitoring video clip from the security network and send the target monitoring video clip to the head office terminal device.

3. The remote video monitoring system according to claim 2, characterized in that, The security network and the production network communicate based on an encryption protocol, which includes the Financial Security Transmission Protocol (FSTP).

4. The remote video monitoring system according to claim 2, characterized in that, When the security network, based on the video supervision task requirements corresponding to the video supervision task, applies an edge computing algorithm to extract and store monitoring video segments corresponding to the requirements of the video supervision task from the monitoring videos collected by the lower-level branch terminal devices, the specific configuration is as follows: Based on the video supervision task requirements corresponding to the video supervision task, and based on the mapping relationship between the video supervision task requirements and the monitoring probes in the lower-level branch terminal equipment, the target monitoring probes in the lower-level branch terminal equipment corresponding to the video supervision task requirements are determined. An edge detection algorithm is applied to extract monitoring video segments from the monitoring videos collected by the target monitoring probe that correspond to the requirements of the video supervision task, and these segments are then stored.

5. The remote video monitoring system according to claim 4, characterized in that, When the security network applies an edge computing algorithm to extract monitoring video segments corresponding to the requirements of the video supervision task from the monitoring videos collected by the target monitoring probes, the specific configuration is as follows: Based on the scene features, time range, and action rules preset in the edge algorithm, the initial interception rules are determined; The initial capture rules are optimized based on the dynamic personnel density in the real-time scene from the monitoring video collected from the target monitoring probe, and / or based on the environmental perception information of the target monitoring probe. Based on the optimized extraction rules, the monitoring videos collected by the target monitoring probe are filtered, and monitoring video segments corresponding to the requirements of the video supervision task are extracted.

6. The remote video supervision system according to any one of claims 1 to 5, characterized in that, The head office terminal equipment includes a head office management and control system, a head office imaging platform, and a head office data department. The head office management and control system is equipped with a video surveillance module, wherein: The head office control system is configured to issue video monitoring tasks to the branch terminal devices; The video monitoring module is configured to acquire corresponding target monitoring video segments from the branch terminal device as needed based on the video file list. The head office imaging platform is configured to interact with the video supervision module, obtain target monitoring video segments from the video supervision module, assign video segment identifiers to the target monitoring video segments, and store the target monitoring video segments based on the video segment identifiers; The head office data department is configured to interact with the head office image platform, retrieve the target monitoring video segment based on the video segment identifier, and apply a large visual understanding model to perform compliance identification based on multimodal semantic analysis on the target monitoring video segment, thereby obtaining and outputting the compliance identification result corresponding to the target monitoring video segment.

7. The remote video monitoring system according to claim 6, characterized in that, The head office data department is also configured as follows: After outputting the compliance identification result corresponding to the target monitoring video segment, obtain the manual review result; Based on the results of the manual review, the prompt words and sub-models of the visual understanding model are optimized and adjusted to generate an optimized visual understanding model.

8. A remote video supervision method, characterized in that, The remote video supervision method, applied to a branch office device in a remote video supervision system as described in any one of claims 1 to 7, comprises: Upon receiving a video monitoring task from the head office terminal device, the system applies an edge algorithm to extract and store monitoring video segments corresponding to the video monitoring task requirements from the monitoring videos collected by the lower-level branch terminal devices, based on the video monitoring task requirements. The monitored video clips are organized into a list of video files, and the list of video files is transmitted to the head office terminal device; In addition, the head office terminal device provides target monitoring video segments, which are obtained on demand by the head office terminal device based on the video file list. The head office terminal device is used to perform compliance identification based on multimodal semantic analysis on the target monitoring video segments, and obtain and output the compliance identification result corresponding to the target monitoring video segments.

9. A remote video supervision method, characterized in that, The remote video supervision method, applied to the central terminal device in any one of claims 1 to 7, comprises: Send video surveillance tasks to branch office equipment; Based on the video file list provided by the branch terminal device, the corresponding target monitoring video segment is obtained from the branch terminal device as needed. The video file list contains relevant information about the monitoring video segment. The monitoring video segment is extracted by the branch terminal device from the monitoring video collected by the lower-level branch terminal device according to the video supervision task requirements corresponding to the video supervision task. A large-scale visual understanding model is applied to perform compliance identification on the target monitoring video segment based on multimodal semantic analysis, and the compliance identification result corresponding to the target monitoring video segment is obtained and output.

10. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in claim 8 or 9.

Images