Video safety supervision information query method and system based on large language model

Abstract

The invention provides a video safety supervision information query method and system based on a large language model. The video safety supervision information query method based on the large language model comprises the following steps: receiving input information of a user from a front-end web interface in real time, and transmitting the input information of the user to a rear-end reasoning module; the method comprises the following steps: defining a tool required by a camera and alarm information query correspondingly, after receiving input information of a user, judging whether a tool corresponding to initial analysis demand information needs to be called or not according to the initial analysis demand information corresponding to the input information of the user, and sending a calling result to a back-end reasoning module; and after the calling result is received, reasoning is performed according to the calling result, and a reasoning result is fed back to the application module. The system comprises modules corresponding to the steps of the method.

Description

Technical Field

[0001] The invention proposes a video safety monitoring information query method and system based on a large language model, belonging to the technical field of video monitoring query. Background Art

[0002] Existing video security monitoring platform systems usually rely on users to operate through a graphical user interface (GUI) to control monitoring equipment and query alarm information. Users need to be familiar with the system's interface, buttons, and operating logic to effectively retrieve alarm information or issue control commands. Existing video security monitoring platform systems have the following problems:

[0003] 1. The user threshold is high, and it takes a certain amount of training to be able to use the video security monitoring platform to control cameras and query alarm information.

[0004] 2. Low query efficiency: When users perform camera control and alarm information query operations, they usually need to perform multiple steps on the platform to locate and retrieve information.

[0005] Single interaction method: The interaction between users and the platform mainly relies on clicking and issuing commands on the platform's user interface with the mouse and keyboard, lacking the flexibility of natural language interaction. Summary of the invention

[0006] The present invention provides a video safety monitoring information query method and system based on a large language model to solve the above technical problems existing in the prior art. The technical solutions adopted are as follows:

[0007] A video safety and monitoring information query system based on a large language model, the video safety and monitoring information query system based on a large language model comprises:

[0008] The streamlit application module is used to receive user input information from the front-end web interface in real time and pass the user input information to the back-end reasoning module;

[0009] The tool registration module is used to define the tools required for the corresponding camera and alarm information query. After receiving the user's input information, it determines whether it is necessary to call the tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information, and sends the call result observation to the back-end reasoning module;

[0010] The backend inference module is used to perform inference based on the call result observation after receiving the call result observation, and feed back the inference result to the streamlit application module.

[0011] Furthermore, the streamlit application module includes:

[0012] An input information monitoring module, used to monitor input information on the front-end web interface;

[0013] An input information preprocessing module is used to extract the user's input information after receiving the user's input information, and perform data cleaning on the user's input information to obtain the input information after data cleaning;

[0014] The information transmission module is used to transmit the data information after the data cleaning to the back-end reasoning module.

[0015] Furthermore, the tool registration module includes:

[0016] A definition module is used to define the tools required for the corresponding query of camera and alarm information;

[0017] An initial parsing requirement information acquisition module is used to initialize a large language model (LLM) Agent after receiving the user's input information, perform natural language understanding and parsing, and obtain the initial parsing requirement information corresponding to the user's input information;

[0018] A tool calling module, used for judging whether it is necessary to call a tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information;

[0019] A calling result generating module, for calling a tool corresponding to the user input information according to the registration information of the tool registration module when it is determined that the tool corresponding to the initial parsing requirement information needs to be called, and generating a calling result observation, wherein the calling result includes but is not limited to alarm information and algorithm information, etc.;

[0020] The result sending module is used to send the call result observation to the backend reasoning module.

[0021] Furthermore, the backend reasoning module includes:

[0022] The demand information extraction module is used to extract the user's input information and its corresponding initial parsing demand information;

[0023] A result parsing module is used to parse the call result observation and obtain the data information corresponding to the call result observation; for example, if the call result observation is alarm information, the data information corresponding to the call result observation includes key information such as alarm type, time, and location;

[0024] The final analysis requirement information acquisition module is used to obtain the user's final analysis requirement information based on the initial analysis requirement information corresponding to the user's input information and the data information corresponding to the call result observation;

[0025] The query result feedback module is used to retrieve the query result according to the final analysis requirement information, and feed back the query result to the user's front-end web interface for display.

[0026] Furthermore, the query result feedback module includes:

[0027] A real-time monitoring module is used to monitor the display parameters of the front-end web interface in real time, wherein the display parameters include loading time, frame rate, content completeness rate and page redraw rate;

[0028] A primary evaluation coefficient acquisition module, used to acquire the primary evaluation coefficient using the loading time and frame rate;

[0029] The primary evaluation coefficient is obtained by the following formula:

[0030]

[0031] Where J represents the primary evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; F ri represents the frame rate corresponding to the i-th page loading; F c Indicates the preset frame rate reference value; T zi represents the loading time corresponding to the i-th page loading; F rb represents the standard deviation of the frame rate corresponding to n page loads; T zb Indicates the standard deviation of loading time corresponding to n page loads;

[0032] A primary coefficient comparison module, used for comparing the primary evaluation coefficient with a preset primary evaluation coefficient threshold;

[0033] A no-abnormality determination module, used to determine that there is no abnormality in the front-end web interface display when the primary evaluation coefficient is lower than a preset primary evaluation coefficient threshold;

[0034] A data retrieval module, configured to retrieve the content completeness rate and the page redraw rate included in the display parameters when the primary evaluation coefficient is not lower than a preset primary evaluation coefficient threshold;

[0035] A comprehensive evaluation coefficient acquisition module, used to obtain a comprehensive evaluation coefficient by combining the primary evaluation coefficient with the content completeness rate and the page redraw rate;

[0036] The comprehensive evaluation coefficient is obtained by the following formula:

[0037]

[0038] Among them, Y represents the comprehensive evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; P ci represents the page redraw rate corresponding to the i-th page load; P wi represents the content completeness rate corresponding to the i-th page load; P cb Indicates the page redraw rate corresponding to n page loads; P wb Indicates the content completeness rate corresponding to the nth page load; J i represents the primary evaluation coefficient corresponding to the i-th page load; J b represents the primary evaluation coefficient corresponding to n page loads; J max and J min Indicates the maximum and minimum values ​​of the primary evaluation coefficient corresponding to n page loads;

[0039] An evaluation coefficient comparison module, used for comparing the comprehensive evaluation coefficient with a preset comprehensive evaluation coefficient threshold;

[0040] The abnormality determination and early warning module is used to determine that there is an abnormality in the display of the front-end web interface and issue an abnormality early warning when the comprehensive evaluation coefficient exceeds a preset comprehensive evaluation coefficient threshold.

[0041] A video safety and monitoring information query method based on a large language model, the video safety and monitoring information query method based on a large language model comprising:

[0042] Receive user input information from the front-end web interface in real time and pass the user input information to the back-end reasoning module;

[0043] Define the tools required for the corresponding camera and alarm information query. After receiving the user's input information, determine whether to call the tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information, and send the call result observation to the backend reasoning module;

[0044] After receiving the call result observation, reasoning is performed based on the call result observation, and the reasoning result is fed back to the streamlit application module.

[0045] Furthermore, the receiving of user input information from the front-end web interface in real time and transmitting the user input information to the back-end reasoning module includes:

[0046] Monitoring input information on the front-end web interface;

[0047] After receiving the user's input information, extract the user's input information, and perform data cleaning on the user's input information to obtain the cleaned input information;

[0048] The cleaned data is transmitted to the back-end reasoning module.

[0049] Furthermore, the tool required for the corresponding query of the camera and the alarm information is defined, and after receiving the user's input information, it is determined whether it is necessary to call the tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information, and the call result observation is sent to the back-end reasoning module, including:

[0050] Define the tools needed to map cameras and alarm information queries;

[0051] After receiving the user's input information, initialize the Large Language Model (LLM) Agent to perform natural language understanding and parsing, and obtain initial parsing requirement information corresponding to the user's input information;

[0052] Determining whether it is necessary to call a tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information;

[0053] When it is determined that a tool corresponding to the initial parsing requirement information needs to be called, the tool corresponding to the user input information is called according to the registration information of the tool registration module, and a calling result observation is generated, wherein the calling result includes but is not limited to alarm information and algorithm information, etc.;

[0054] The result sending module is used to send the call result observation to the backend reasoning module.

[0055] Furthermore, after receiving the call result observation, reasoning is performed according to the call result observation, and the reasoning result is fed back to the streamlit application module, including:

[0056] Extract user input information and its corresponding initial parsing requirement information;

[0057] Parse the call result observation to obtain data information corresponding to the call result observation; for example, if the call result observation is alarm information, the data information corresponding to the call result observation includes key information such as alarm type, time, and location;

[0058] Obtain the user's final analysis requirement information based on the initial analysis requirement information corresponding to the user's input information and the data information corresponding to the call result observation;

[0059] The query results are retrieved according to the final analysis requirement information, and the query results are fed back to the user's front-end web interface for display.

[0060] Further, query results are retrieved according to the final analysis requirement information, and the query results are fed back to the user's front-end web interface for display, including:

[0061] Real-time monitoring of display parameters during the display of the front-end web interface, wherein the display parameters include loading time, frame rate, content completeness rate and page redraw rate;

[0062] Obtaining a primary evaluation coefficient using the loading time and the frame rate;

[0063] The primary evaluation coefficient is obtained by the following formula:

[0064]

[0065] Where J represents the primary evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; F ri represents the frame rate corresponding to the i-th page loading; F c Indicates the preset frame rate reference value; T zi represents the loading time corresponding to the i-th page loading; F rb represents the standard deviation of the frame rate corresponding to n page loads; T zb Indicates the standard deviation of loading time corresponding to n page loads;

[0066] comparing the primary evaluation coefficient with a preset primary evaluation coefficient threshold;

[0067] When the primary evaluation coefficient is lower than the preset primary evaluation coefficient threshold, it is determined that there is no abnormality in the front-end web interface display;

[0068] When the primary evaluation coefficient is not lower than the preset primary evaluation coefficient threshold, the content completeness rate and the page redrawing rate included in the display parameters are retrieved;

[0069] The primary evaluation coefficient is combined with the content completeness rate and the page redraw rate to obtain a comprehensive evaluation coefficient;

[0070] The comprehensive evaluation coefficient is obtained by the following formula:

[0071]

[0072] Among them, Y represents the comprehensive evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; P ci represents the page redraw rate corresponding to the i-th page load; P wi represents the content completeness rate corresponding to the i-th page load; P cb Indicates the page redraw rate corresponding to n page loads; P wb Indicates the content completeness rate corresponding to the nth page load; J i represents the primary evaluation coefficient corresponding to the i-th page load; J b represents the primary evaluation coefficient corresponding to n page loads; J max and J min Indicates the maximum and minimum values ​​of the primary evaluation coefficient corresponding to n page loads;

[0073] Comparing the comprehensive evaluation coefficient with a preset comprehensive evaluation coefficient threshold;

[0074] When the comprehensive evaluation coefficient exceeds the preset comprehensive evaluation coefficient threshold, it is determined that there is an abnormality in the front-end web interface display, and an abnormality warning is issued.

[0075] Beneficial effects of the present invention:

[0076] The video safety monitoring information query method and system proposed in the present invention based on a large language model have the following advantages over the prior art:

[0077] 1. User-friendly operation mode: Through natural language interaction, the threshold for users to use the video security monitoring platform system is lowered.

[0078] 2. Efficient query: Direct use of natural language query reduces user operation steps and improves retrieval efficiency.

[0079] High intelligence level. It can automatically execute control instructions and analyze and give suggestions according to user intentions, which improves the intelligence level of the video security monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 is a system block diagram of the system of the present invention;

[0081] Figure 2 The present invention is a flowchart of the method. DETAILED DESCRIPTION

[0082] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0083] The embodiment of the present invention proposes a video safety information query system based on a large language model, such as Figure 1 As shown, the video safety information query system based on the large language model includes:

[0084] The streamlit application module is used to receive user input information from the front-end web interface in real time and pass the user input information to the back-end reasoning module;

[0085] The tool registration module is used to define the tools required for the corresponding camera and alarm information query. After receiving the user's input information, it determines whether it is necessary to call the tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information, and sends the call result observation to the back-end reasoning module;

[0086] The backend inference module is used to perform inference based on the call result observation after receiving the call result observation, and feed back the inference result to the streamlit application module.

[0087] The working principle of the above technical solution is as follows: This embodiment designs a video safety monitoring platform information query and control system based on the Large Language Model (LLM) agent technology. LLM agent is an advanced artificial intelligence technology that uses LLM's powerful natural language understanding and generation capabilities to create an intelligent agent that can provide feedback and perform actions based on the environment and user input. The system consists of a streamlit application module, a tool registration module, and a backend reasoning module. The streamlit module is used to accept user input from the front-end web interface, call the back-end reasoning module for reasoning, and return the results to the user. The tool registration module defines the functional modules required to implement camera control and alarm information query, and tells LLM when to call these functions through registration information. The back-end reasoning module chooses whether to call the tool in the tool registration module based on user input. If called, the tool will return a call result, called observation (including alarm information, algorithm information, etc.), and the reasoning module returns the reasoning result based on user input and observation.

[0088] The effect of the above technical solution is: interacting with users through the front-end Web interface simplifies the query process and enables users to obtain the required information more quickly. The back-end reasoning module calls the corresponding tools for querying according to the user's input information, which improves the accuracy and efficiency of the query. The Streamlit application module displays the reasoning results to the user in a visual form, making the information more intuitive and easy to understand. Users can obtain the required video safety information with simple clicks and inputs, which improves the user experience. The system predefines a variety of tools that can be flexibly called according to different query requirements. The back-end reasoning module can process various types of input information and generate corresponding reasoning results, which improves the flexibility and scalability of the system. As a lightweight, multi-modal friendly application framework, Streamlit can support the seamless integration of multi-modal data such as text, images, and audio. This means that in the process of video safety information query, the system can receive and process various types of information input, further improving the practicality and flexibility of the system. The technical solution can receive and process user query requests in real time and quickly generate reasoning results. This helps operation and maintenance personnel to understand the operating status of the video safety system in a timely manner and perform corresponding optimization and management.

[0089] To sum up, the video safety information query method based on the large language model realizes the real-time processing of user input information, tool calling and feedback of reasoning results in terms of working principle; in terms of technical effect, it improves query efficiency, enhances user experience, improves system flexibility, supports multimodal data interaction and optimizes operation and maintenance management.

[0090] In one embodiment of the present invention, the streamlit application module includes:

[0091] An input information monitoring module, used to monitor input information on the front-end web interface;

[0092] An input information preprocessing module is used to extract the user's input information after receiving the user's input information, and perform data cleaning on the user's input information to obtain the input information after data cleaning;

[0093] The information transmission module is used to transmit the data information after the data cleaning to the back-end reasoning module.

[0094] The working principle of the above technical solution is as follows: the front-end web interface continuously monitors the user's input behavior through JavaScript or other front-end technologies. This includes the user's operations on interface elements such as text boxes, drop-down menus, buttons, and data entered by the user through devices such as keyboards and mice. After receiving the user's input information, the front-end web interface will immediately extract this information. The extracted information may include multiple data types such as text, numbers, dates, and times. Next, the input information is cleaned. The data cleaning process includes removing invalid characters (such as spaces, line breaks, special symbols, etc.), checking whether the data type meets expectations, and verifying the integrity and consistency of the data. The purpose of data cleaning is to ensure that the information passed to the back-end reasoning module is accurate, valid, and meets expectations. After data cleaning, the input information is encapsulated into a suitable data format (such as JSON, XML, etc.). Then, this information is sent to the back-end reasoning module through HTTP requests or other communication protocols (such as WebSocket, Ajax, etc.). After receiving this information, the back-end reasoning module will perform further processing and analysis to generate the query results required by the user or perform corresponding operations.

[0095] The effect of the above technical solution is: through the data cleaning step, invalid characters and erroneous data in the user input can be removed to ensure that the information passed to the back-end reasoning module is accurate and valid. This helps to reduce errors and anomalies in the back-end processing process and improve the stability and reliability of the system. Real-time monitoring of user input behavior and timely extraction and cleaning of data can ensure that users can quickly get feedback or query results after entering information. This helps to improve user satisfaction and loyalty and enhance the user experience of the system. The technical solution adopts a general data cleaning and delivery mechanism that can support user input information of various types and formats. This enables the system to easily cope with new needs or changes that may arise in the future, and enhances the scalability and flexibility of the system. By monitoring and cleaning the input information at the front end, the impact of malicious input or attack behavior on the back-end system can be prevented to a certain extent. This helps to improve the security of the system and protect the security of user data and system resources.

[0096] To sum up, in the process of receiving user input information on the front-end Web interface and transmitting it to the back-end reasoning module, the technical solution achieves improved data accuracy, optimized user experience, enhanced system scalability and improved system security through steps such as real-time monitoring, data cleaning and information transmission.

[0097] In one embodiment of the present invention, the tool registration module includes:

[0098] A definition module is used to define the tools required for the corresponding query of camera and alarm information;

[0099] An initial parsing requirement information acquisition module is used to initialize a large language model (LLM) Agent after receiving the user's input information, perform natural language understanding and parsing, and obtain the initial parsing requirement information corresponding to the user's input information;

[0100] A tool calling module, used for judging whether it is necessary to call a tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information;

[0101] A calling result generating module, for calling a tool corresponding to the user input information according to the registration information of the tool registration module when it is determined that the tool corresponding to the initial parsing requirement information needs to be called, and generating a calling result observation, wherein the calling result includes but is not limited to alarm information and algorithm information, etc.;

[0102] The result sending module is used to send the call result observation to the backend reasoning module.

[0103] The working principle of the above technical solution is as follows: the system first defines various tools required for camera and alarm information query. These tools may include database query statements, API interfaces, algorithm models, etc., which are used to implement specific query functions. These tools are registered in the system so that they can be correctly called when needed. When the system receives the user's input information, it initializes the large language model (LLM) agent. The LLM agent understands and parses the natural language input by the user and extracts the specific content that the user wants to query, that is, the initial parsing demand information. This step realizes the conversion from natural language to query requirements that the system can understand. The system determines whether it is necessary to call the tools corresponding to these demand information based on the initial parsing demand information. This judgment process may involve comprehensive consideration of factors such as the type, scope, and complexity of the query requirements. When it is determined that the tool corresponding to the initial parsing demand information needs to be called, the system finds and calls the corresponding tool according to the registration information of the tool registration module. The tool executes the query or processing task and generates the call result observation. These results may include alarm information, algorithm information, query results, etc. The call result observation is the basis for subsequent reasoning and decision-making. The result sending module sends the call result observation to the back-end reasoning module. After receiving this information, the backend reasoning module will perform further processing and analysis to generate the final query results required by the user or perform corresponding operations.

[0104] The effect of the above technical solution is: by defining and registering the tools required for the query, the system can quickly find and call the corresponding tools for query. This avoids the tedious process of manually writing query statements or calling interfaces in traditional methods, and improves the query efficiency. LLM Agent can understand and parse the user's natural language input and convert it into query requirements that the system can understand. This enables the system to support multiple query methods and query conditions, and enhances the flexibility of the query. Through the tool call judgment step, the system can ensure that only tools corresponding to the query requirements are called for query. This avoids the interference of irrelevant tools and the generation of erroneous results, and improves the accuracy of the query. Users can input query requirements in natural language without understanding the internal structure and query syntax of the system. This reduces the user's usage threshold and learning cost, and optimizes the user experience. This technical solution is not only applicable to camera and alarm information query scenarios, but can also be extended to other similar query requirements. This makes the system have a wider application prospect and stronger market competitiveness.

[0105] In summary, the technical solution achieves efficient, flexible, accurate, easy-to-use and scalable technical effects in terms of tool definition, input information parsing, tool calling and sending of calling results for camera and alarm information query.

[0106] In one embodiment of the present invention, the backend reasoning module includes:

[0107] The demand information extraction module is used to extract the user's input information and its corresponding initial parsing demand information;

[0108] A result parsing module is used to parse the call result observation and obtain the data information corresponding to the call result observation; for example, if the call result observation is alarm information, the data information corresponding to the call result observation includes key information such as alarm type, time, and location;

[0109] The final analysis requirement information acquisition module is used to obtain the user's final analysis requirement information based on the initial analysis requirement information corresponding to the user's input information and the data information corresponding to the call result observation;

[0110] The query result feedback module is used to retrieve the query result according to the final analysis requirement information, and feed back the query result to the user's front-end web interface for display.

[0111] The working principle of the above technical solution is as follows: the system first extracts the user's input information, which usually includes the form filled in by the user on the front-end web interface, the options selected, or the natural language text entered. At the same time, the system also extracts the initial parsing requirement information corresponding to the input information. This information is obtained by the large language model (LLM) agent in the previous step after natural language understanding and parsing of the user input. After receiving the call result observation, the system parses it to obtain the corresponding data information. The call result observation may contain multiple types of information, such as alarm information, algorithm analysis results, etc. For each type of information, the system extracts key fields, such as alarm type, time, location, etc. The system combines the initial parsing requirement information corresponding to the user's input information with the data information corresponding to the call result observation to obtain the user's final parsing requirement information. This step may involve further refinement or correction of the initial parsing requirement information, as well as supplementation or adjustment based on the new information provided by the call result observation. According to the final parsing requirement information, the system retrieves the corresponding query results. These query results may come from databases, file systems, external APIs, etc. The system will feed back the query results in an appropriate form (such as tables, charts, text, etc.) to the user's front-end web interface for display.

[0112] The effect of the above technical solution is: by parsing the call result observation, the system can make full use of the data information therein to provide users with more detailed and accurate query results. This avoids the problem of information waste or omission that may exist in traditional methods. The system can infer based on the user's input information and the call result observation to obtain the user's final parsing demand information. This enables the system to understand the user's query intention more accurately and provide users with more personalized services. The system feeds back the query results to the user's front-end Web interface in an intuitive and easy-to-understand form for display. This reduces the user's usage threshold and learning cost, and improves user satisfaction and loyalty. Improve system response speed: Through efficient call result parsing and reasoning process, the system can quickly generate query results and feedback to the user. This improves the system's response speed and reduces the user's waiting time. This technical solution is not only applicable to alarm information query scenarios, but can also be extended to other similar query requirements. This gives the system a wider application prospect and stronger market competitiveness.

[0113] In summary, the technical solution achieves efficient, accurate, easy-to-use and scalable technical effects in terms of parsing and reasoning of call results and feedback of query results, etc. These effects together improve the overall performance of the system and user experience.

[0114] In one embodiment of the present invention, the query result feedback module includes:

[0115] A real-time monitoring module is used to monitor the display parameters of the front-end web interface in real time, wherein the display parameters include loading time, frame rate, content completeness rate and page redraw rate;

[0116] A primary evaluation coefficient acquisition module, used to acquire the primary evaluation coefficient using the loading time and frame rate;

[0117] The primary evaluation coefficient is obtained by the following formula:

[0118]

[0119] Where J represents the primary evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; F ri represents the frame rate corresponding to the i-th page loading; F c Indicates the preset frame rate reference value; T zi represents the loading time corresponding to the i-th page loading; F rb represents the standard deviation of the frame rate corresponding to n page loads; T zb Indicates the standard deviation of loading time corresponding to n page loads;

[0120] A primary coefficient comparison module, used for comparing the primary evaluation coefficient with a preset primary evaluation coefficient threshold;

[0121] A no-abnormality determination module, used to determine that there is no abnormality in the front-end web interface display when the primary evaluation coefficient is lower than a preset primary evaluation coefficient threshold;

[0122] A data retrieval module, configured to retrieve the content completeness rate and the page redraw rate included in the display parameters when the primary evaluation coefficient is not lower than a preset primary evaluation coefficient threshold;

[0123] A comprehensive evaluation coefficient acquisition module, used to obtain a comprehensive evaluation coefficient by combining the primary evaluation coefficient with the content completeness rate and the page redraw rate;

[0124] The comprehensive evaluation coefficient is obtained by the following formula:

[0125]

[0126] Among them, Y represents the comprehensive evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; P ci represents the page redraw rate corresponding to the i-th page load; P wi represents the content completeness rate corresponding to the i-th page load; P cbIndicates the page redraw rate corresponding to n page loads; P wb Indicates the content completeness rate corresponding to the nth page load; J i represents the primary evaluation coefficient corresponding to the i-th page load; J b represents the primary evaluation coefficient corresponding to n page loads; J max and J min Indicates the maximum and minimum values ​​of the primary evaluation coefficient corresponding to n page loads;

[0127] An evaluation coefficient comparison module, used for comparing the comprehensive evaluation coefficient with a preset comprehensive evaluation coefficient threshold;

[0128] The abnormality determination and early warning module is used to determine that there is an abnormality in the display of the front-end web interface and issue an abnormality early warning when the comprehensive evaluation coefficient exceeds a preset comprehensive evaluation coefficient threshold.

[0129] The working principle of the above technical solution is: the system monitors the key display parameters in the display process of the front-end web interface in real time, including loading time, frame rate, content integrity rate and page redraw rate. These parameters can fully reflect the display performance and user experience of the front-end web interface. The system uses the loading time and frame rate to calculate the primary evaluation coefficient J. The primary evaluation coefficient J comprehensively considers factors such as the number of page loads, the difference between the frame rate of each load and the preset frame rate reference value, the frame rate standard deviation and the loading time standard deviation. Through this step, the system can make a preliminary evaluation of the basic display performance of the front-end web interface. The system compares the primary evaluation coefficient J with the preset primary evaluation coefficient threshold. This step is used to determine whether the display performance of the front-end web interface meets the basic requirements. When the primary evaluation coefficient J is not lower than the preset threshold, the system believes that there may be problems with the display performance of the front-end web interface, so it is necessary to further retrieve the content integrity rate and page redraw rate for comprehensive evaluation. The system combines the primary evaluation coefficient J, the content integrity rate Pwi and the page redraw rate Pci to calculate the comprehensive evaluation coefficient Y. The comprehensive evaluation coefficient Y not only considers the basic display performance (loading time and frame rate), but also adds the evaluation of content integrity and page rendering efficiency. Through this step, the system can comprehensively and accurately evaluate the display performance of the front-end web interface. The system compares the comprehensive evaluation coefficient Y with the preset comprehensive evaluation coefficient threshold. When the comprehensive evaluation coefficient Y exceeds the preset threshold, the system determines that there is an abnormality in the display of the front-end web interface and issues an abnormality warning. This step can detect problems in a timely manner and remind relevant personnel to handle them, ensuring user experience and system stability.

[0130] The effect of the above technical solution is: by real-time monitoring of key display parameters (loading time, frame rate, content completeness rate and page redraw rate), the system can comprehensively and accurately reflect the display performance of the front-end web interface. The system uses primary evaluation coefficients and comprehensive evaluation coefficients to evaluate the display performance of the front-end web interface at multiple levels and dimensions. This evaluation method not only takes into account the basic display performance, but also adds the evaluation of content integrity and page rendering efficiency, which improves the accuracy and comprehensiveness of the evaluation. When the comprehensive evaluation coefficient exceeds the preset threshold, the system can issue an abnormal warning in time. This helps relevant personnel to discover and deal with problems in a timely manner, avoiding adverse effects on user experience and system stability. Through real-time monitoring and abnormal warning mechanisms, the system can promptly discover and optimize the display performance problems of the front-end web interface. This helps to improve user experience and satisfaction and enhance the market competitiveness of the system.

[0131] On the other hand, the technical solution can fully reflect the overall performance of the front-end web interface by real-time monitoring of multiple key performance indicators such as loading time, frame rate, content integrity rate and page redraw rate during the display process of the front-end web interface. Through the calculation of the primary evaluation coefficient and the comprehensive evaluation coefficient, the technical solution realizes a multi-level evaluation of the performance of the front-end web interface, taking into account both the basic display performance (such as loading time and frame rate) and the evaluation of content integrity and page rendering efficiency. The technical solution calculates the primary evaluation coefficient and the comprehensive evaluation coefficient through specific mathematical formulas, realizes the quantitative evaluation of the performance of the front-end web interface, and avoids the ambiguity and uncertainty of subjective evaluation. The calculation of the primary evaluation coefficient and the comprehensive evaluation coefficient both take into account the situation of multiple page loadings, and dynamically adjusts according to the actual situation, thereby improving the accuracy and adaptability of the evaluation. Through real-time monitoring and abnormal early warning mechanism, the technical solution can timely discover performance problems that may occur in the display process of the front-end web interface, such as too long loading time, too low frame rate, etc. Timely discovery and optimization of performance problems help improve user experience and satisfaction, and avoid user loss and negative word of mouth caused by performance problems. Through performance monitoring and evaluation, this technical solution can prevent potential system failures and performance bottlenecks, and improve system stability and reliability. Timely discovery and resolution of performance issues can help reduce system maintenance costs and avoid serious consequences such as system crashes and data loss caused by performance issues.

[0132] In summary, the technical solution achieves efficient, accurate and timely technical effects in performance monitoring, evaluation and abnormal warning of the front-end web interface display process. These effects jointly improve the overall performance of the system and user experience.

[0133] The video safety monitoring information query method based on the large language model proposed in the embodiment of the present invention is as follows: Figure 2As shown, the video safety information query method based on the large language model includes:

[0134] S1, receiving user input information from the front-end web interface in real time, and passing the user input information to the back-end reasoning module;

[0135] S2, define the tools required for the corresponding query of the camera and alarm information, and after receiving the user's input information, determine whether it is necessary to call the tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information, and send the call result observation to the back-end reasoning module;

[0136] S3. After receiving the call result observation, reasoning is performed based on the call result observation, and the reasoning result is fed back to the streamlit application module.

[0137] The working principle of the above technical solution is as follows: the user inputs query information through the front-end Web interface, which may be about the status and location of the camera, or the details of the alarm information. The front-end Web interface receives the user's input information in real time and passes it to the back-end reasoning module. The system predefines various tools required for camera query and alarm information query, which may include database query statements, API interfaces, etc. After receiving the user's input information, the system determines whether it is necessary to call the tools corresponding to these demand information based on the initial parsing demand information corresponding to the input information. The judgment result is sent to the back-end reasoning module in the form of a call result observation. After receiving the call result observation, the back-end reasoning module calls the corresponding tool for query or processing based on the information therein. The reasoning process may involve further analysis of the input information, access to the database or API interface, and preliminary processing of the query results. After the reasoning is completed, the reasoning result is fed back to the streamlit application module. The Streamlit application module receives the reasoning result returned by the back-end reasoning module. The reasoning result is displayed to the user in a visual form, which may include the location and status information of the camera, details of the alarm information, etc.

[0138] The effect of the above technical solution is: interacting with users through the front-end Web interface simplifies the query process and enables users to obtain the required information more quickly. The back-end reasoning module calls the corresponding tools for querying according to the user's input information, which improves the accuracy and efficiency of the query. The Streamlit application module displays the reasoning results to the user in a visual form, making the information more intuitive and easy to understand. Users can obtain the required video safety information with simple clicks and inputs, which improves the user experience. The system predefines a variety of tools that can be flexibly called according to different query requirements. The back-end reasoning module can process various types of input information and generate corresponding reasoning results, which improves the flexibility and scalability of the system. As a lightweight, multi-modal friendly application framework, Streamlit can support the seamless integration of multi-modal data such as text, images, and audio. This means that in the process of video safety information query, the system can receive and process various types of information input, further improving the practicality and flexibility of the system. The technical solution can receive and process user query requests in real time and quickly generate reasoning results. This helps operation and maintenance personnel to understand the operating status of the video safety system in a timely manner and perform corresponding optimization and management.

[0139] To sum up, the video safety information query method based on the large language model realizes the real-time processing of user input information, tool calling and feedback of reasoning results in terms of working principle; in terms of technical effect, it improves query efficiency, enhances user experience, improves system flexibility, supports multimodal data interaction and optimizes operation and maintenance management.

[0140] In one embodiment of the present invention, the step of receiving user input information from a front-end web interface in real time and transmitting the user input information to a back-end reasoning module includes:

[0141] S101, monitoring input information on the front-end web interface;

[0142] S102, after receiving the user's input information, extract the user's input information, and perform data cleaning on the user's input information to obtain the cleaned input information;

[0143] S103: passing the cleaned data information to a backend reasoning module.

[0144] The working principle of the above technical solution is as follows: the front-end web interface continuously monitors the user's input behavior through JavaScript or other front-end technologies. This includes the user's operations on interface elements such as text boxes, drop-down menus, buttons, and data entered by the user through devices such as keyboards and mice. After receiving the user's input information, the front-end web interface will immediately extract this information. The extracted information may include multiple data types such as text, numbers, dates, and times. Next, the input information is cleaned. The data cleaning process includes removing invalid characters (such as spaces, line breaks, special symbols, etc.), checking whether the data type meets expectations, and verifying the integrity and consistency of the data. The purpose of data cleaning is to ensure that the information passed to the back-end reasoning module is accurate, valid, and meets expectations. After data cleaning, the input information is encapsulated into a suitable data format (such as JSON, XML, etc.). Then, this information is sent to the back-end reasoning module through HTTP requests or other communication protocols (such as WebSocket, Ajax, etc.). After receiving this information, the back-end reasoning module will perform further processing and analysis to generate the query results required by the user or perform corresponding operations.

[0145] The effect of the above technical solution is: through the data cleaning step, invalid characters and erroneous data in the user input can be removed to ensure that the information passed to the back-end reasoning module is accurate and valid. This helps to reduce errors and anomalies in the back-end processing process and improve the stability and reliability of the system. Real-time monitoring of user input behavior and timely extraction and cleaning of data can ensure that users can quickly get feedback or query results after entering information. This helps to improve user satisfaction and loyalty and enhance the user experience of the system. The technical solution adopts a general data cleaning and delivery mechanism that can support user input information of various types and formats. This enables the system to easily cope with new needs or changes that may arise in the future, and enhances the scalability and flexibility of the system. By monitoring and cleaning the input information at the front end, the impact of malicious input or attack behavior on the back-end system can be prevented to a certain extent. This helps to improve the security of the system and protect the security of user data and system resources.

[0146] To sum up, in the process of receiving user input information on the front-end Web interface and transmitting it to the back-end reasoning module, the technical solution achieves improved data accuracy, optimized user experience, enhanced system scalability and improved system security through steps such as real-time monitoring, data cleaning and information transmission.

[0147] In one embodiment of the present invention, the tool required for the corresponding query of the camera and the alarm information is defined, and after receiving the user's input information, it is determined whether it is necessary to call the tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information, and the calling result observation is sent to the back-end reasoning module, including:

[0148] S201, defining the tools required for the correspondence between the camera and the alarm information query;

[0149] S202: after receiving the user's input information, initialize the large language model (LLM) Agent, perform natural language understanding and parsing, and obtain initial parsing requirement information corresponding to the user's input information;

[0150] S203, judging whether it is necessary to call a tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information;

[0151] S204, when it is determined that a tool corresponding to the initial parsing requirement information needs to be called, the tool corresponding to the user input information is called according to the registration information of the tool registration module, and a calling result observation is generated, wherein the calling result includes but is not limited to alarm information and algorithm information, etc.;

[0152] S205: Send the call result observation to the backend reasoning module.

[0153] The working principle of the above technical solution is as follows: the system first defines various tools required for camera and alarm information query. These tools may include database query statements, API interfaces, algorithm models, etc., which are used to implement specific query functions. These tools are registered in the system so that they can be correctly called when needed. When the system receives the user's input information, it initializes the large language model (LLM) agent. The LLM agent understands and parses the natural language input by the user and extracts the specific content that the user wants to query, that is, the initial parsing demand information. This step realizes the conversion from natural language to query requirements that the system can understand. The system determines whether it is necessary to call the tools corresponding to these demand information based on the initial parsing demand information. This judgment process may involve comprehensive consideration of factors such as the type, scope, and complexity of the query requirements. When it is determined that the tool corresponding to the initial parsing demand information needs to be called, the system finds and calls the corresponding tool according to the registration information of the tool registration module. The tool executes the query or processing task and generates the call result observation. These results may include alarm information, algorithm information, query results, etc. The call result observation is the basis for subsequent reasoning and decision-making. The result sending module sends the call result observation to the back-end reasoning module. After receiving this information, the backend reasoning module will perform further processing and analysis to generate the final query results required by the user or perform corresponding operations.

[0154] The effect of the above technical solution is: by defining and registering the tools required for the query, the system can quickly find and call the corresponding tools for query. This avoids the tedious process of manually writing query statements or calling interfaces in traditional methods, and improves the query efficiency. LLM Agent can understand and parse the user's natural language input and convert it into query requirements that the system can understand. This enables the system to support multiple query methods and query conditions, and enhances the flexibility of the query. Through the tool call judgment step, the system can ensure that only tools corresponding to the query requirements are called for query. This avoids the interference of irrelevant tools and the generation of erroneous results, and improves the accuracy of the query. Users can input query requirements in natural language without understanding the internal structure and query syntax of the system. This reduces the user's usage threshold and learning cost, and optimizes the user experience. This technical solution is not only applicable to camera and alarm information query scenarios, but can also be extended to other similar query requirements. This makes the system have a wider application prospect and stronger market competitiveness.

[0155] In summary, the technical solution achieves efficient, flexible, accurate, easy-to-use and scalable technical effects in terms of tool definition, input information parsing, tool calling and sending of calling results for camera and alarm information query.

[0156] In one embodiment of the present invention, after receiving the call result observation, reasoning is performed according to the call result observation, and the reasoning result is fed back to the streamlit application module, including:

[0157] S301, extracting user input information and its corresponding initial parsing requirement information;

[0158] S302, parsing the call result observation, and obtaining data information corresponding to the call result observation; for example, if the call result observation is alarm information, the data information corresponding to the call result observation includes key information such as alarm type, time, and location;

[0159] S303, obtaining the user's final analysis requirement information according to the initial analysis requirement information corresponding to the user's input information and the data information corresponding to the call result observation;

[0160] S304: retrieve query results according to the final analysis requirement information, and feed back the query results to the user's front-end web interface for display.

[0161] The working principle of the above technical solution is as follows: the system first extracts the user's input information, which usually includes the form filled in by the user on the front-end web interface, the options selected, or the natural language text entered. At the same time, the system also extracts the initial parsing requirement information corresponding to the input information. This information is obtained by the large language model (LLM) agent in the previous step after natural language understanding and parsing of the user input. After receiving the call result observation, the system parses it to obtain the corresponding data information. The call result observation may contain multiple types of information, such as alarm information, algorithm analysis results, etc. For each type of information, the system extracts key fields, such as alarm type, time, location, etc. The system combines the initial parsing requirement information corresponding to the user's input information with the data information corresponding to the call result observation to obtain the user's final parsing requirement information. This step may involve further refinement or correction of the initial parsing requirement information, as well as supplementation or adjustment based on the new information provided by the call result observation. According to the final parsing requirement information, the system retrieves the corresponding query results. These query results may come from databases, file systems, external APIs, etc. The system will feed back the query results in an appropriate form (such as tables, charts, text, etc.) to the user's front-end web interface for display.

[0162] The effect of the above technical solution is: by parsing the call result observation, the system can make full use of the data information therein to provide users with more detailed and accurate query results. This avoids the problem of information waste or omission that may exist in traditional methods. The system can infer based on the user's input information and the call result observation to obtain the user's final parsing demand information. This enables the system to understand the user's query intention more accurately and provide users with more personalized services. The system feeds back the query results to the user's front-end Web interface in an intuitive and easy-to-understand form for display. This reduces the user's usage threshold and learning cost, and improves user satisfaction and loyalty. Improve system response speed: Through efficient call result parsing and reasoning process, the system can quickly generate query results and feedback to the user. This improves the system's response speed and reduces the user's waiting time. This technical solution is not only applicable to alarm information query scenarios, but can also be extended to other similar query requirements. This gives the system a wider application prospect and stronger market competitiveness.

[0163] In summary, the technical solution achieves efficient, accurate, easy-to-use and scalable technical effects in terms of parsing and reasoning of call results and feedback of query results, etc. These effects together improve the overall performance of the system and user experience.

[0164] In one embodiment of the present invention, query results are retrieved according to the final parsing requirement information, and the query results are fed back to the user's front-end web interface for display, including:

[0165] S3041. Real-time monitoring of display parameters during the display of the front-end web interface, wherein the display parameters include loading time, frame rate, content completeness rate, and page redraw rate;

[0166] S3042, obtaining a primary evaluation coefficient using the loading time and frame rate;

[0167] The primary evaluation coefficient is obtained by the following formula:

[0168]

[0169] Where J represents the primary evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; F ri represents the frame rate corresponding to the i-th page loading; F c Indicates the preset frame rate reference value; T zi represents the loading time corresponding to the i-th page loading; F rb represents the standard deviation of the frame rate corresponding to n page loads; T zb Indicates the standard deviation of loading time corresponding to n page loads;

[0170] S3043, comparing the primary evaluation coefficient with a preset primary evaluation coefficient threshold;

[0171] S3044, when the primary evaluation coefficient is lower than the preset primary evaluation coefficient threshold, it is determined that there is no abnormality in the front-end web interface display;

[0172] S3045: when the primary evaluation coefficient is not lower than a preset primary evaluation coefficient threshold, the content completeness rate and the page redrawing rate included in the display parameters are retrieved;

[0173] S3046, using the primary evaluation coefficient in combination with the content completeness rate and the page redraw rate to obtain a comprehensive evaluation coefficient;

[0174] The comprehensive evaluation coefficient is obtained by the following formula:

[0175]

[0176] Among them, Y represents the comprehensive evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; P ci represents the page redraw rate corresponding to the i-th page load; P wi represents the content completeness rate corresponding to the i-th page load; P cb Indicates the page redraw rate corresponding to n page loads; P wb Indicates the content completeness rate corresponding to the nth page load; J i represents the primary evaluation coefficient corresponding to the i-th page load; J b represents the primary evaluation coefficient corresponding to n page loads; J max and J min Indicates the maximum and minimum values ​​of the primary evaluation coefficient corresponding to n page loads;

[0177] S3047, comparing the comprehensive evaluation coefficient with a preset comprehensive evaluation coefficient threshold;

[0178] S3048. When the comprehensive evaluation coefficient exceeds the preset comprehensive evaluation coefficient threshold, it is determined that there is an abnormality in the front-end web interface display, and an abnormality warning is issued.

[0179] The working principle of the above technical solution is: the system monitors the key display parameters in the display process of the front-end web interface in real time, including loading time, frame rate, content integrity rate and page redraw rate. These parameters can fully reflect the display performance and user experience of the front-end web interface. The system uses the loading time and frame rate to calculate the primary evaluation coefficient J. The primary evaluation coefficient J comprehensively considers factors such as the number of page loads, the difference between the frame rate of each load and the preset frame rate reference value, the frame rate standard deviation and the loading time standard deviation. Through this step, the system can make a preliminary evaluation of the basic display performance of the front-end web interface. The system compares the primary evaluation coefficient J with the preset primary evaluation coefficient threshold. This step is used to determine whether the display performance of the front-end web interface meets the basic requirements. When the primary evaluation coefficient J is not lower than the preset threshold, the system believes that there may be problems with the display performance of the front-end web interface, so it is necessary to further retrieve the content integrity rate and page redraw rate for comprehensive evaluation. The system combines the primary evaluation coefficient J, the content integrity rate Pwi and the page redraw rate Pci to calculate the comprehensive evaluation coefficient Y. The comprehensive evaluation coefficient Y not only considers the basic display performance (loading time and frame rate), but also adds the evaluation of content integrity and page rendering efficiency. Through this step, the system can comprehensively and accurately evaluate the display performance of the front-end web interface. The system compares the comprehensive evaluation coefficient Y with the preset comprehensive evaluation coefficient threshold. When the comprehensive evaluation coefficient Y exceeds the preset threshold, the system determines that there is an abnormality in the display of the front-end web interface and issues an abnormality warning. This step can detect problems in a timely manner and remind relevant personnel to handle them, ensuring user experience and system stability.

[0180] The effect of the above technical solution is: by real-time monitoring of key display parameters (loading time, frame rate, content completeness rate and page redraw rate), the system can comprehensively and accurately reflect the display performance of the front-end web interface. The system uses primary evaluation coefficients and comprehensive evaluation coefficients to evaluate the display performance of the front-end web interface at multiple levels and dimensions. This evaluation method not only takes into account the basic display performance, but also adds the evaluation of content integrity and page rendering efficiency, which improves the accuracy and comprehensiveness of the evaluation. When the comprehensive evaluation coefficient exceeds the preset threshold, the system can issue an abnormal warning in time. This helps relevant personnel to discover and deal with problems in a timely manner, avoiding adverse effects on user experience and system stability. Through real-time monitoring and abnormal warning mechanisms, the system can promptly discover and optimize the display performance problems of the front-end web interface. This helps to improve user experience and satisfaction and enhance the market competitiveness of the system.

[0181] On the other hand, the technical solution can fully reflect the overall performance of the front-end web interface by real-time monitoring of multiple key performance indicators such as loading time, frame rate, content integrity rate and page redraw rate during the display process of the front-end web interface. Through the calculation of the primary evaluation coefficient and the comprehensive evaluation coefficient, the technical solution realizes a multi-level evaluation of the performance of the front-end web interface, taking into account both the basic display performance (such as loading time and frame rate) and the evaluation of content integrity and page rendering efficiency. The technical solution calculates the primary evaluation coefficient and the comprehensive evaluation coefficient through specific mathematical formulas, realizes the quantitative evaluation of the performance of the front-end web interface, and avoids the ambiguity and uncertainty of subjective evaluation. The calculation of the primary evaluation coefficient and the comprehensive evaluation coefficient both take into account the situation of multiple page loadings, and dynamically adjusts according to the actual situation, thereby improving the accuracy and adaptability of the evaluation. Through real-time monitoring and abnormal early warning mechanism, the technical solution can timely discover performance problems that may occur in the display process of the front-end web interface, such as too long loading time, too low frame rate, etc. Timely discovery and optimization of performance problems help improve user experience and satisfaction, and avoid user loss and negative word of mouth caused by performance problems. Through performance monitoring and evaluation, this technical solution can prevent potential system failures and performance bottlenecks, and improve system stability and reliability. Timely discovery and resolution of performance issues can help reduce system maintenance costs and avoid serious consequences such as system crashes and data loss caused by performance issues.

[0182] In summary, the technical solution achieves efficient, accurate and timely technical effects in performance monitoring, evaluation and abnormal warning of the front-end web interface display process. These effects jointly improve the overall performance of the system and user experience.

[0183] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A video safety information query system based on a large language model, characterized in that: The video safety information query system based on the large language model includes: The streamlit application module is used to receive user input information from the front-end web interface in real time and pass the user input information to the back-end reasoning module; The tool registration module is used to define the tools required for the corresponding camera and alarm information query. After receiving the user's input information, it determines whether it is necessary to call the tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information, and sends the call result observation to the back-end reasoning module; The backend inference module is used to perform inference based on the call result observation after receiving the call result observation, and feed back the inference result to the streamlit application module.

2. According to the video safety information query system based on the large language model of claim 1, it is characterized in that: The streamlit application module includes: An input information monitoring module, used to monitor input information on the front-end web interface; An input information preprocessing module is used to extract the user's input information after receiving the user's input information, and perform data cleaning on the user's input information to obtain the input information after data cleaning; The information transmission module is used to transmit the data information after the data cleaning to the back-end reasoning module.

3. According to the video safety information query system based on the large language model of claim 1, it is characterized in that: The tool registration module includes: A definition module is used to define the tools required for the corresponding query of camera and alarm information; An initial parsing requirement information acquisition module is used to initialize the large language model Agent after receiving the user's input information, perform natural language understanding and parsing, and obtain the initial parsing requirement information corresponding to the user's input information; A tool calling module, used for judging whether it is necessary to call a tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information; A calling result generating module, for calling a tool corresponding to the user input information according to the registration information of the tool registration module when it is determined that the tool corresponding to the initial parsing requirement information needs to be called, and generating a calling result observation; The result sending module is used to send the call result observation to the backend reasoning module.

4. According to the video safety information query system based on the large language model of claim 1, it is characterized in that: The back-end reasoning module includes: The demand information extraction module is used to extract the user's input information and its corresponding initial parsing demand information; A result parsing module is used to parse the call result observation and obtain the data information corresponding to the call result observation; The final analysis requirement information acquisition module is used to obtain the user's final analysis requirement information based on the initial analysis requirement information corresponding to the user's input information and the data information corresponding to the call result observation; The query result feedback module is used to retrieve the query result according to the final analysis requirement information, and feed back the query result to the user's front-end web interface for display.

5. According to the video safety information query system based on the large language model of claim 4, it is characterized in that: The query result feedback module includes: A real-time monitoring module is used to monitor the display parameters of the front-end web interface in real time, wherein the display parameters include loading time, frame rate, content completeness rate and page redraw rate; A primary evaluation coefficient acquisition module, used to acquire the primary evaluation coefficient using the loading time and frame rate; The primary evaluation coefficient is obtained by the following formula: Where J represents the primary evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; F ri represents the frame rate corresponding to the i-th page loading; F c Indicates the preset frame rate reference value; T zi represents the loading time corresponding to the i-th page loading; F rb represents the standard deviation of the frame rate corresponding to n page loads; T zb Indicates the standard deviation of loading time corresponding to n page loads; A primary coefficient comparison module, used for comparing the primary evaluation coefficient with a preset primary evaluation coefficient threshold; A no-abnormality determination module, used to determine that there is no abnormality in the front-end web interface display when the primary evaluation coefficient is lower than a preset primary evaluation coefficient threshold; A data retrieval module, configured to retrieve the content completeness rate and the page redraw rate included in the display parameters when the primary evaluation coefficient is not lower than a preset primary evaluation coefficient threshold; A comprehensive evaluation coefficient acquisition module, used to obtain a comprehensive evaluation coefficient by combining the primary evaluation coefficient with the content completeness rate and the page redraw rate; The comprehensive evaluation coefficient is obtained by the following formula: Among them, Y represents the comprehensive evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; P ci represents the page redraw rate corresponding to the i-th page load; P wi represents the content completeness rate corresponding to the i-th page load; P cb Indicates the page redraw rate corresponding to n page loads; P wb Indicates the content completeness rate corresponding to the nth page load; J i represents the primary evaluation coefficient corresponding to the i-th page load; J b represents the primary evaluation coefficient corresponding to n page loads; J max and J min Indicates the maximum and minimum values ​​of the primary evaluation coefficient corresponding to n page loads; An evaluation coefficient comparison module, used for comparing the comprehensive evaluation coefficient with a preset comprehensive evaluation coefficient threshold; The abnormality determination and early warning module is used to determine that there is an abnormality in the display of the front-end web interface and issue an abnormality early warning when the comprehensive evaluation coefficient exceeds a preset comprehensive evaluation coefficient threshold.

6. A video safety monitoring information query method based on a large language model, characterized in that: The video safety monitoring information query method based on the large language model includes: Receive user input information from the front-end web interface in real time and pass the user input information to the back-end reasoning module; Define the tools required for the corresponding camera and alarm information query. After receiving the user's input information, determine whether to call the tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information, and send the call result observation to the back-end reasoning module; After receiving the call result observation, reasoning is performed based on the call result observation, and the reasoning result is fed back to the streamlit application module.

7. The video safety monitoring information query method based on a large language model according to claim 6 is characterized in that: The receiving of user input information from the front-end web interface in real time and transmitting the user input information to the back-end reasoning module includes: Monitoring input information on the front-end web interface; After receiving the user's input information, extract the user's input information, and perform data cleaning on the user's input information to obtain the cleaned input information; The cleaned data is transmitted to the back-end reasoning module.

8. The video safety monitoring information query method based on a large language model according to claim 6 is characterized in that: The tool required for the corresponding query of the camera and alarm information is defined, and after receiving the user's input information, it is determined whether it is necessary to call the tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information, and the call result observation is sent to the back-end reasoning module, including: Define the tools needed to map cameras and alarm information queries; After receiving the user's input information, initialize the large language model Agent, perform natural language understanding and parsing, and obtain initial parsing requirement information corresponding to the user's input information; Determining whether it is necessary to call a tool corresponding to the initial parsing requirement information according to the initial parsing requirement information corresponding to the user's input information; When it is determined that a tool corresponding to the initial parsing requirement information needs to be called, the tool corresponding to the user input information is called according to the registration information of the tool registration module, and a calling result observation is generated; The result sending module is used to send the call result observation to the backend reasoning module.

9. The video safety monitoring information query method based on a large language model according to claim 6 is characterized in that: After receiving the call result observation, reasoning is performed based on the call result observation, and the reasoning result is fed back to the streamlit application module, including: Extract user input information and its corresponding initial parsing requirement information; Parse the call result observation to obtain the data information corresponding to the call result observation; Obtain the user's final analysis requirement information based on the initial analysis requirement information corresponding to the user's input information and the data information corresponding to the call result observation; The query results are retrieved according to the final analysis requirement information, and the query results are fed back to the user's front-end web interface for display.

10. The video safety monitoring information query method based on a large language model according to claim 9 is characterized in that: Retrieving query results according to the final parsing requirement information, and feeding back the query results to the user's front-end web interface for display, including: Real-time monitoring of display parameters during the display of the front-end web interface, wherein the display parameters include loading time, frame rate, content completeness rate and page redraw rate; Obtaining a primary evaluation coefficient using the loading time and the frame rate; The primary evaluation coefficient is obtained by the following formula: Where J represents the primary evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; F ri represents the frame rate corresponding to the i-th page loading; F c Indicates the preset frame rate reference value; T zi represents the loading time corresponding to the i-th page loading; F rb represents the standard deviation of the frame rate corresponding to n page loads; T zb Indicates the standard deviation of loading time corresponding to n page loads; comparing the primary evaluation coefficient with a preset primary evaluation coefficient threshold; When the primary evaluation coefficient is lower than the preset primary evaluation coefficient threshold, it is determined that there is no abnormality in the front-end web interface display; When the primary evaluation coefficient is not lower than the preset primary evaluation coefficient threshold, the content completeness rate and the page redrawing rate included in the display parameters are retrieved; The primary evaluation coefficient is combined with the content completeness rate and the page redraw rate to obtain a comprehensive evaluation coefficient; The comprehensive evaluation coefficient is obtained by the following formula: Among them, Y represents the comprehensive evaluation coefficient; n represents the number of page loads experienced during the display of the front-end web interface; P ci represents the page redraw rate corresponding to the i-th page load; P wi represents the content completeness rate corresponding to the i-th page load; P cb Indicates the page redraw rate corresponding to n page loads; P wb Indicates the content completeness rate corresponding to the nth page load; J i represents the primary evaluation coefficient corresponding to the i-th page load; J b represents the primary evaluation coefficient corresponding to n page loads; J max and J min Indicates the maximum and minimum values ​​of the primary evaluation coefficient corresponding to n page loads; Comparing the comprehensive evaluation coefficient with a preset comprehensive evaluation coefficient threshold; When the comprehensive evaluation coefficient exceeds the preset comprehensive evaluation coefficient threshold, it is determined that there is an abnormality in the front-end web interface display, and an abnormality warning is issued.

Images