Rule dynamic configuration method and device

Abstract

This application provides a method and device for dynamic rule configuration. The method includes: generating a first screening rule according to a first rule configuration instruction; screening initial data to be screened using the first screening rule according to an operation instruction for the first screening rule to obtain a first screening result; generating a second screening rule according to a second rule configuration instruction; and screening the screening result of the previous screening rule again using the second screening rule according to the operation instruction for the second screening rule to obtain a second screening result. This method can provide technical support for improving the accuracy and reliability of early warnings for unknown scenarios or new events.

Description

Technical Field

[0001] This application relates to the field of security technology, and in particular to a method and device for dynamic rule configuration. Background Technology

[0002] With the deep integration of artificial intelligence (AI) and Internet of Things (IoT) technologies, the security field is gradually evolving from passive monitoring to proactive intelligent sensing. The system analyzes real-time data to achieve early detection and proactive warnings of events.

[0003] However, current proactive intelligent sensing solutions rely heavily on learning and training from historical event data. When faced with unknown scenarios or novel events not fully covered in the training data, the system's recognition accuracy and reliability often drop significantly. Summary of the Invention

[0004] In view of this, this application provides a method and device for dynamic rule configuration.

[0005] According to a first aspect of the embodiments of this application, a method for dynamic rule configuration is provided, including: Generate the first screening rule based on the first rule configuration instruction; Based on the execution instructions for the first screening rule, the initial data to be screened is screened using the first screening rule to obtain the first screening result; Generate a second screening rule based on the second rule configuration instructions; Based on the execution instructions for the second screening rule, the screening results after the previous screening rule is executed are screened again using the second screening rule to obtain the second screening result.

[0006] According to a second aspect of the embodiments of this application, an electronic device is provided, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor, when executing a program stored in memory, implements the method provided in the first aspect.

[0007] According to a third aspect of the embodiments of this application, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, it implements the method provided in the first aspect.

[0008] According to a fourth aspect of the embodiments of this application, a computer program is provided, which is stored in a computer-readable storage medium, and when a processor executes the computer program, causes the processor to perform the method provided in the first aspect.

[0009] The rule dynamic configuration method of this application embodiment generates a first screening rule according to a first rule configuration instruction, and uses the first screening rule to screen initial data to be screened according to a running instruction for a second screening rule to obtain a first screening result. Then, it generates a second screening rule according to the second rule configuration instruction, and uses the second screening rule to screen the screening result after the previous screening rule is run again according to the running instruction for the second screening rule to obtain a second screening result. By dynamically configuring screening rules in real time and running the configured screening rules, the screening result of the configured screening rules can be obtained. This screening result can be used to verify the effectiveness of the configured screening rules. It can achieve the goal of obtaining screening rules that meet the requirements without relying on historical event data for training. During the rule dynamic configuration process, the current screening result can be screened again by configuring new screening rules based on the screening result of the configured screening rules. This provides technical support for improving the accuracy and reliability of early warning for unknown scenarios or new events. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating a method for generating early warning rules provided in an embodiment of this application; Figure 2A This is a schematic diagram of a screening rule configuration interface provided in an embodiment of this application; Figure 2B This is a schematic diagram of another screening rule configuration interface provided in an embodiment of this application; Figure 2C This is a schematic diagram of a rule parameter configuration interface provided in an embodiment of this application; Figure 3A This is a schematic diagram of a configuration interface for text-based image search rules provided in an embodiment of this application; Figure 3B This is a schematic diagram of another configuration interface for text-based image search rules provided in an embodiment of this application; Figure 4A This is a schematic diagram of the face attribute analysis rule configuration interface of an aggregation strategy provided in an embodiment of this application; Figure 4B This is a schematic diagram of the face attribute analysis rule configuration interface for an identity verification strategy provided in an embodiment of this application; Figure 4C This is a schematic diagram of the face attribute analysis rule configuration interface for a people-based strategy provided in an embodiment of this application; Figure 4D This is a schematic diagram of the face attribute analysis rule configuration interface of a feature search strategy provided in an embodiment of this application; Figure 5AThis is a schematic diagram of a human attribute analysis rule configuration interface provided in an embodiment of this application, which has a human number strategy and a detection area covering the entire image. Figure 5B This is a schematic diagram of a human attribute analysis rule configuration interface provided in an embodiment of this application, where the detection area is a target box, based on a people-counting strategy. Figure 5C This is a schematic diagram of the configuration interface for human attribute analysis rules of an aggregation strategy provided in an embodiment of this application; Figure 5D This is a schematic diagram of the configuration interface for human attribute analysis rules, which is a feature search strategy provided in an embodiment of this application and whose detection area is the entire image. Figure 5E This is a schematic diagram of the configuration interface for human attribute analysis rules, where the detection area is a target bounding box, according to a feature search strategy provided in this application embodiment. Figure 6A This is a schematic diagram of a vehicle attribute analysis rule configuration interface provided in an embodiment of this application, where the detection area is the entire map and a quantity strategy is employed. Figure 6B This is a schematic diagram of a vehicle attribute analysis rule configuration interface provided in an embodiment of this application, where the detection area is a target bounding box and a quantity strategy is used. Figure 6C This is a schematic diagram of the vehicle attribute analysis rule configuration interface for a feature search strategy provided in an embodiment of this application; Figure 6D This is a schematic diagram of the vehicle attribute analysis rule configuration interface for a vehicle orientation strategy provided in an embodiment of this application; Figure 7A This is a schematic diagram of a non-motorized vehicle attribute analysis rule configuration interface provided in an embodiment of this application, where the detection area is the entire map; Figure 7B This is a schematic diagram of a non-motor vehicle attribute analysis rule configuration interface provided in an embodiment of this application, where the detection area is a target box; Figure 7C This is a schematic diagram of the configuration interface for non-motorized vehicle attribute analysis rules, which is a feature search strategy provided in an embodiment of this application and has a detection area of ​​the entire map. Figure 7D This is a schematic diagram of the configuration interface for non-motor vehicle attribute analysis rules, which is a feature search strategy and a detection area is a target box, provided in an embodiment of this application. Figure 8 This is a schematic diagram of the AIOP algorithm rule configuration interface for an OCR algorithm strategy provided in an embodiment of this application; Figure 9 This is a schematic diagram of an early warning rule distribution interface provided in an embodiment of this application; Figure 10This is a schematic diagram of a result card provided in an embodiment of this application; Figure 11 This is a schematic diagram of a list of monitoring points to be removed, provided in an embodiment of this application; Figure 12 This is a flowchart illustrating an early warning method provided in an embodiment of this application; Figure 13 This is a schematic diagram of an alarm card provided in an embodiment of this application; Figure 14 This is a schematic diagram of the structure of a rule dynamic configuration device provided in an embodiment of this application; Figure 15 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] To enable those skilled in the art to better understand the technical solutions provided in the embodiments of this application, and to make the above-mentioned objectives, features and advantages of the embodiments of this application more apparent and understandable, the technical solutions in the embodiments of this application will be further described in detail below with reference to the accompanying drawings.

[0012] It should be noted that the sequence number of each step in the embodiments of this application does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0013] Please see Figure 1 This is a flowchart illustrating a rule dynamic configuration method provided in an embodiment of this application, as shown below. Figure 1 As shown, the dynamic configuration method for this rule may include the following steps: Step S110: Generate the first screening rule according to the first rule configuration instruction.

[0014] Step S120: According to the running instructions for the first screening rule, the initial data to be screened is screened using the first screening rule to obtain the first screening result.

[0015] Step S130: Generate the second screening rule according to the second rule configuration instruction.

[0016] Step S140: Based on the running instructions for the second screening rule, the screening results after the previous screening rule is run are screened again using the second screening rule to obtain the second screening result.

[0017] In this embodiment of the application, in order to avoid the active warning scheme from relying on training data, effective screening rules can be generated by dynamically configuring screening rules in real time and verifying the configured screening rules in real time.

[0018] For example, for any configured screening rule, the screening result of the screening rule can be obtained by running the screening rule, so as to verify the effect of the screening rule based on the screening result, thereby determining the effect of the configured screening rule in real time.

[0019] For example, if the effect of the screening rule is determined to be unsatisfactory, the configured screening rule can be adjusted, the screening result of the adjusted screening rule can be obtained by running the adjusted screening rule, and the effect of the adjusted screening rule can be verified again.

[0020] If the screening rule is found to be effective and meets the requirements, the currently configured screening rule can be determined as a valid screening rule.

[0021] For example, a diagram of the screening rule configuration interface can be found here. Figure 2A ,like Figure 2A As shown, the screening rule configuration interface can include a rule parameter configuration interface and a screening result display interface. The rule parameter configuration interface is used to configure the rule parameters for screening rules, generating corresponding screening rules based on the configured parameters. For the configured screening rules, data screening can be performed by running the configured rules, and the screening results are displayed on the screening result display interface. Users can determine the effectiveness of the currently configured screening rules through the screening results displayed on the screening result display interface.

[0022] For example, if the false alarm rate of the screening results displayed in the visualization interface meets the requirements (such as the false alarm rate being lower than the preset false alarm rate threshold), the user can determine that the current screening result meets the requirements.

[0023] In this embodiment of the application, in order to optimize the data screening effect, new screening rules can be added on the basis of the configured screening rules, and the new screening rules can be used to further screen the screening results of the configured screening rules.

[0024] For example, for any application scenario, when running the first screening rule configured for that application scenario (which can be called the first screening rule), the initial data to be screened can be screened using the first screening rule to obtain the corresponding screening result (which can be called the first screening result).

[0025] When running a non-first screening rule (which can be called the second screening rule) configured for this application scenario, the second screening rule can be used to screen the screening results after the previous screening rule is run again to obtain the corresponding screening results (which can be called the second screening results).

[0026] For example, for any application scenario, the data to be screened can be image data captured by monitoring points in that application scenario.

[0027] For example, image data captured by monitoring points can be modeled to obtain image modeling vectors, which are then stored in a vector database. During the dynamic configuration of rules, the generated screening rules can be used to match the image modeling vectors in the vector database.

[0028] For example, during the dynamic configuration of rules, the generated screening rules can be used to screen historical image modeling vectors; during the early warning process, the generated early warning rules can be used to process real-time image modeling vectors.

[0029] For example, the previous screening rule refers to the configured and determined previous screening rule.

[0030] For example, suppose that for a certain application scenario, the first screening rule generated according to the rule configuration command is screening rule A. When screening rule A is run, it can be used to screen the initial data to be screened, obtaining the first screening result. After obtaining the first screening result, a new screening rule, such as screening rule B, can be configured. When screening rule B is run, it can be used to screen the first screening result again, obtaining the second screening result.

[0031] It should be noted that, in the embodiments of this application, the non-first screening rule mentioned above (such as the second screening rule mentioned above) refers to a newly configured screening rule when there is at least one configured and determined to be effective screening rule (such as the first screening rule mentioned above).

[0032] For example, for any screening rule, the rule parameters can be configured and run once or multiple times. For instance, for the first screening rule mentioned above, if it is determined that the first screening result does not meet the requirements after running the first screening rule and obtaining the first screening result, the rule parameters of the first screening rule can be adjusted, and then the rule can be run again after the adjustment is completed.

[0033] In one example, for any configured screening rule, if it is determined that the screening result after running the screening rule meets the requirements (i.e., the screening rule is determined to be valid), the screening rule can be added to the list of rules to be issued.

[0034] Accordingly, for any screening rule, after obtaining the screening result after running the screening rule, if a rule addition instruction for the screening rule is detected, the screening rule is added to the list of rules to be issued.

[0035] For example, the various screening rules in the list of rules to be issued have an execution order and logical relationship.

[0036] For example, such as Figure 2B As shown, the screening rule configuration interface also includes a list of rules to be issued, which displays the list of rules to be issued, including configured and confirmed valid screening rules.

[0037] For example, the screening rule configuration interface may include an "Add Rule" button. If the user is certain that the currently configured screening rule is valid (e.g., the screening results after running the currently configured screening rule meet the requirements), the user can add the currently configured screening rule to the list of rules to be issued by triggering (e.g., clicking) the "Add Rule" button.

[0038] When the system receives an instruction to add a currently configured screening rule (such as a trigger instruction for the "Add Rule" function button), it can add the currently configured screening rule to the list of rules to be issued.

[0039] Accordingly, the first screening rule mentioned above (such as the first screening rule) can refer to the screening rule configured when the list of rules to be issued is empty; the non-first screening rule mentioned above (such as the second screening rule) can refer to the screening rule configured when there is at least one screening rule in the list of rules to be issued.

[0040] In this embodiment of the application, if the obtained screening result meets the requirements, an early warning rule can be generated by combining the configured first screening rule and second screening rule. The early warning rule can be used for early warning processing in the corresponding application scenario, such as intelligent detection of early warning events.

[0041] For example, users can click the "Save" button or the "Rule Issuance" button in the screening result configuration interface to trigger the system to generate early warning rules based on the configured screening rules.

[0042] For example, during the rule distribution process, information such as the validity period, application location, and effective time period of the warning rule can be configured according to actual needs. A diagram can be shown as follows: Figure 9 As shown.

[0043] For example, the validity period may include one day, one week, one month, indefinitely, or other custom time.

[0044] A valid time period can include one or more time periods.

[0045] For example, in the process of performing early warning analysis based on the early warning rules generated using the technical solutions provided in the embodiments of this application, the multiple screening rules included in the early warning rules can be used to screen the data in a serial and / or parallel manner to obtain the final screening result.

[0046] For example, to perform data screening using each screening rule in a sequential manner, the data can be screened according to the generation order of each screening rule in the early warning rules (such as the configuration order of the screening rules).

[0047] Each screening rule can be used to screen the screening results of the previous screening rule (for the first screening rule, the previous screening rule does not exist, and the screening result of the previous screening rule is the initial data of the subject to be screened), and the screening result of the last screening rule is the final screening result.

[0048] It should be noted that when data screening is performed in parallel using multiple configured screening rules (which may include some screening rules (including multiple screening rules) or all screening rules in the early warning rules), the multiple screening rules can be obtained separately for data screening, and the data that meets the conditions corresponding to each screening rule can be obtained separately. The intersection of the data that meets the conditions corresponding to the multiple screening rules is used to determine the final data that meets the conditions corresponding to the multiple screening rules.

[0049] It can be seen that, in Figure 1 In the illustrated method flow, a first screening rule is generated based on the first rule configuration instruction. Then, based on the execution instruction for the second screening rule, the initial data to be screened is screened using the first screening rule to obtain the first screening result. Next, a second screening rule is generated based on the second rule configuration instruction. Finally, based on the execution instruction for the second screening rule, the screening result after the previous screening rule is run is screened again using the second screening rule to obtain the second screening result. By dynamically configuring screening rules in real time and running the configured rules, the screening result of the configured screening rule is obtained. This screening result can be used to verify the effectiveness of the configured screening rule. It can achieve the desired screening rule effect without relying on historical event data for training. During the dynamic rule configuration process, based on the screening result of the already configured screening rule, new screening rules can be configured to re-screen the current screening result, providing technical support for improving the accuracy and reliability of early warnings for unknown scenarios or new events.

[0050] For example, during the dynamic configuration of rules, it is possible to support the configuration of screening rules of various different rule algorithm types, so as to perform data screening based on screening rules of various different rule algorithm types.

[0051] For example, during the screening rule configuration process, the screening rule configuration interface supports the rule algorithm type selection function. For instance, the screening rule configuration interface may include selection options corresponding to different rule algorithm types. Users can initiate the configuration of screening rules for the corresponding rule algorithm type by selecting the corresponding selection option.

[0052] For example, screening rules of different rule algorithm types can correspond to different screening rule configuration interfaces (such as...). Figure 2A (The rule parameter configuration interfaces are different in different regions), and different screening rule configuration interfaces can include different configurable items.

[0053] For example, rule algorithm types may include one or more of the following: text search image type, visual question answering type, target attribute analysis type, and AIOP (AI Open Platform) algorithm type.

[0054] For example, a diagram of the rule parameter configuration interface can be found here. Figure 2C ,like Figure 2C As shown, the rule parameter configuration interface can include tabs corresponding to text-based image search types, tabs corresponding to visual question-and-answer types, and tabs corresponding to target attribute analysis types. Figure 2C Taking the targets as examples, including faces, bodies, vehicles (motor vehicles) and non-motor vehicles, different targets correspond to different tabs) and tabs corresponding to the AIOP algorithm types. Users can control the system to enter the rule configuration interface of the corresponding rule algorithm type (such as the text search rule configuration) by selecting the corresponding tab, such as clicking the "Text Search Image" tab.

[0055] For example, users can configure the screening rules for the selected rule algorithm type by configuring the configurable items included in the rule configuration interface.

[0056] In one example, for text-to-image search rules (i.e., text-to-image search type screening rules, the same below), configurable items include prompt words (which can be called image search text prompt words).

[0057] For example, in order to improve the efficiency of prompt word configuration, prompt word recommendation is supported during the text-to-image search rule configuration process.

[0058] For example, a prompt word recommendation scheme may include: 1) Contextualized guidance.

[0059] Based on the user's selected "topic type" (such as security, transportation, retail), keywords for common search needs are recommended.

[0060] For example, high-frequency scenario templates can be preset to directly provide common search combinations. For example, for the traffic monitoring scenario (finding or tracking a specific vehicle with significant features), the search combination can be red vehicle + daytime + crossroads; for the personnel search scenario (finding a specific target in a crowd based on known features), the search combination can be male with mask + backpack + walking fast.

[0061] After the user clicks on the template, it is automatically filled into the search box. Among them, for what is filled into the search box (prompt word input box), the user can adjust it according to needs.

[0062] 2) Hierarchical structuring.

[0063] The prompt words are divided into three levels: basic attributes (such as objects / actions), detailed modifiers (such as colors / scenes), and exclusion conditions (such as "non-motor vehicle - without helmet"), reducing the user's input burden.

[0064] 3) Real-time feedback.

[0065] Dynamically recommend supplementary entries based on the content already input by the user, forming an interactive experience of "the more you input, the more accurate it becomes".

[0066] Exemplarily, the prompt word recommendation can be carried out in the way of dynamic association recommendation.

[0067] 3.1) Recommendation upon input: When the user inputs "car", the dropdown list recommends attribute words such as "vehicle color - brand - model" (such as "white sedan" or "SUV", etc.).

[0068] 3.2) Semantic expansion: When inputting "old man's joy", recommend related words such as "elderly tricycle" or "elderly scooter", etc., covering synonymous expressions.

[0069] 4) Historical and hot topic prompts.

[0070] 4.1) Personal history record (which can be called personal prompt word history record): Display the prompt words of the user's past successful searches, supporting quick reuse.

[0071] 4.2) Global hot word library: Based on the statistics of high-frequency prompt words on the platform, recommend popular prompt words for the current period (such as "festival activities" or "special weather", etc.).

[0072] 5) Multi-modal prompt word generation.

[0073] 5.1) Generate with the aid of uploaded pictures: After the user uploads a reference picture (which can be called a prompt word reference picture), the system automatically analyzes and generates text prompt words (such as "personnel in blue work clothes in the warehouse"), and the user can modify it based on this.

[0074] 5.2) Voice input to prompts: Supports users to verbally express their needs (i.e., voice prompts input via voice), which are automatically converted into structured text prompts.

[0075] For example, for text-to-image search rules, configurable items may also include at least one of target type, time range, and similarity threshold.

[0076] For example, the configuration interface for text-to-image search rules can be found in [link to configuration]. Figure 3A ,like Figure 3A As shown, for text-to-image search rules, configurable items include prompts (i.e., search text prompts), target type (such as images of people, vehicles, or non-motorized vehicles), time range (such as today, the last 3 days, the last 7 days, or other custom time ranges), and similarity thresholds (such as...). Figure 3A (Similarity in the text).

[0077] For example, for time range configuration, time period configuration can be supported, and its illustration can be shown as follows. Figure 3B As shown.

[0078] For example, the search text prompts include positive prompts (also known as positive corpus) and / or negative prompts (also known as negative corpus or reverse corpus).

[0079] For example, for text-based image search rules, the system can call a multimodal large model to perform data screening based on the configured text-based image search rules (including at least prompt words).

[0080] For example, if the text-to-image search rule is the first screening rule configured for the current application scenario, then when the text-to-image search rule is run, the initial data to be screened can be screened using the text-to-image search rule; if the text-to-image search rule is not the first screening rule configured for the current application scenario, then when the text-to-image search rule is run, the screening results after running the previous screening rule can be screened again using the text-to-image search rule.

[0081] For example, during the data screening process based on text-to-image search rules, if the prompt word is a positive prompt word, the data that matches the text-to-image search rule is retained, and the rest of the data is filtered (i.e. deleted from the current screening results); if the prompt word is a negative prompt word, the data that matches the text-to-image search rule is filtered, and the rest of the data is retained.

[0082] In one example, for visual question-answering rules, configurable options include cue words (which may be called visual question-answering cue words); visual question-answering cue words include positive cue words and / or negative cue words.

[0083] For example, for visual question answering rules, the system can call a multimodal large model to perform data screening based on the configured visual question answering rules.

[0084] For example, the results of the visual question-answering algorithm in processing the data to be screened (such as "yes" or "no") can be used to determine whether the data to be screened meets the requirements (whether it needs to be retained or filtered).

[0085] For example, when the visual question-answering prompt is a positive prompt, for any data to be screened, if the visual question-answering algorithm confirms "yes", the data to be screened is retained; if the visual question-answering algorithm confirms "no", the data to be screened is filtered.

[0086] When the visual question-answering prompt is a negative prompt, for any data to be screened, if the visual question-answering algorithm confirms "yes", the data to be screened is filtered; if the visual question-answering algorithm confirms "no", the data to be screened is retained.

[0087] For example, taking "handheld drone" as an example, positive prompts could be as follows: Please analyze this image carefully to determine if a drone is present. A drone could be in one of two states: Handheld state: being held in someone's hand (e.g., preparing for takeoff, landing, or carrying).

[0088] Core judgment logic and false alarm resistance requirements: The primary basis for identification is morphological characteristics: Regardless of the state, the key structural features of the drone must be clearly identified, such as multiple rotors, arms, landing gear, and a typical square or cross-shaped fuselage. This is fundamental to distinguishing it from square handheld items such as mobile phones, umbrellas, and books.

[0089] Validation in context: If the drone is handheld, the contact between the hand and the drone body must be observable, and the drone's structure must be fully visible. Do not mistake other objects without rotors in the hand for a drone.

[0090] If the presence of a drone is confirmed (i.e., "yes"), please provide feedback in the following format: Status: [Holding] Location: (e.g., "in the hand of the character in the bottom left corner") Quantity: [specific quantity] Identifiable types: (e.g., multi-rotor / fixed-wing / racing drone, etc.; if not identifiable, this can be omitted) "Valid drone target identified."

[0091] If there is no drone in the picture, or the object cannot be identified as a drone due to blurriness or obstruction (e.g., only a square object is visible without rotor features), please answer directly: "No valid drone target detected." (i.e., "No").

[0092] Based on the aforementioned visual question-answering prompts, a multimodal large model can be invoked to process the data to be screened using visual question-answering algorithms, and the processing results corresponding to each piece of data to be screened can be determined. If the processing result is "yes" (confirmation of the existence of drones), the data to be screened is retained; if the processing result is "no" (no valid drone target is identified), the data to be screened is filtered.

[0093] For example, for visual question answering rules, configurable options may also include detection regions, such as target boxes or the entire image.

[0094] For example, in order to improve the efficiency of prompt word configuration, prompt word expansion is supported during the configuration process of visual question answering rules.

[0095] For example, a prompt word expansion scheme may include: 1. Multi-granularity question-answering support At the macro level: Supports asking questions about the overall scenario (such as "Did any abnormal events occur in the diagram?").

[0096] At the detailed level: it supports in-depth analysis of specific goals, attributes, and relationships (such as "What is the person in red doing?").

[0097] Timing level: Supports dynamic questions across frames (such as "Is the vehicle accelerating?").

[0098] 2. Natural language generalization ability It is compatible with synonyms (such as "riding a bike" and "cycling") and colloquial descriptions (such as "Is anyone not wearing a mask?").

[0099] 3. Domain Adaptation It includes pre-set templates for commonly used questions in fields such as security, transportation, and retail, reducing the learning cost for users.

[0100] The specific functions are as follows.

[0101] 1) Structured question template.

[0102] Attribute query template: "Target [Location]: In which area is the [object / person] located?"; "Target [Quantity]: How many cars are there in the picture?".

[0103] Behavior analysis template: "Action description: What is [person / vehicle] doing?"; "Anomaly detection: Is there [driving in the wrong direction / gathering] behavior?".

[0104] Relational reasoning template: "Spatial relationship: Is [A] to the left of [B]?"; "Logical relationship: Are the two people talking?".

[0105] 2) Dynamic Question Generation Assistant.

[0106] Intelligent autocomplete: When a user enters "this person", the system automatically completes the input as "this person's [gender / age / behavior]?".

[0107] Question recommendations: Automatically generate questionable dimensions based on image content (e.g., if a vehicle is detected, recommend questions related to "color / model / license plate").

[0108] Multi-turn question-and-answer memory: Supports context-dependent questioning (e.g., first ask "How many people are there?", then ask "What percentage of them are wearing masks?").

[0109] 3) Multimodal problem input.

[0110] Select and ask questions: After the user selects an area of ​​the image, related questions are automatically generated (e.g., select a vehicle → "What color is this car?").

[0111] Voice questioning: Supports voice input questions, automatically converts them to text and optimizes the expression (e.g., the spoken "What is that person dressed like?" is converted to "What are the characteristics of this person's clothing?").

[0112] 4) Question-based clarification mechanism.

[0113] When the question is vague, the system will ask a follow-up question to narrow down the scope (e.g., if a user asks "What are people doing?", the system will ask "Please specify a specific person or area").

[0114] For example, optional answer buttons could be provided (e.g., asking "Vehicle color?" → displaying "Red / Yellow / Blue" for quick selection).

[0115] In one example, the target attribute analysis rules include screening conditions for attributes of one type or multiple different types of targets.

[0116] For example, the target of one or more different types mentioned above may include one or more of the following: face, human body, vehicle (motor vehicle) and non-motor vehicle.

[0117] As an example, for faces, various different strategies can be configured, such as aggregation strategy, identity verification strategy, people counting strategy, or feature search strategy.

[0118] For example, an aggregation strategy: using a face model, people whose faces meet the similarity requirements are grouped together as a set based on a specified similarity.

[0119] Identity verification strategy: For detected faces, a 1:N comparison is performed using the face list database to determine whether the detected face has a matching face in the face list database.

[0120] People counting strategy: Used to determine whether the number of faces in a specified detection region (which can be called the face detection region in this example) meets the requirements. For example, the detection region may include a bounding box (such as a face detection box) or the entire image.

[0121] Feature search strategy: Used to determine whether a face matching the features is detected based on the configured features (such as whether the person is wearing glasses or a mask).

[0122] It should be noted that, in the embodiments of this application, during the dynamic configuration of rules, multiple target attribute analysis rules for the same type of target (such as face, human body or vehicle, etc.) can be configured, and these multiple target attribute analysis rules for the same type of target can be configured with the same strategy or different strategies.

[0123] For example, during the dynamic configuration of rules, two face attribute analysis rules can be configured. One face attribute analysis rule can be an aggregation strategy, and the other face attribute analysis rule can be an identity verification strategy.

[0124] For example, when the strategy is configured as an aggregation strategy, the facial attribute analysis rule configuration can support configurations such as similarity threshold, time interval, frequency of occurrence, and monitoring point filtering. A diagram illustrating this can be seen as follows: Figure 4A As shown.

[0125] The similarity threshold is used to identify a specified face (if the similarity between the identified face and the specified face exceeds the similarity threshold, the specified face is identified); the time interval is used to define the recognition result whose occurrence count needs to be counted. Assuming the time interval is 5 minutes, the same face identified within 5 minutes is counted as 1 occurrence. If the time interval between two identifications of the same face exceeds 5 minutes, the face occurrence count is updated; the occurrence count is used to define the occurrence count of the same face that satisfies the aggregation strategy. For example, if the occurrence count is greater than or equal to 2, then the same face must have an occurrence count greater than or equal to 2; the monitoring point is used to define the monitoring point from which the video / image data for recognition is sourced.

[0126] For example, when the strategy is configured as an identity verification strategy, the facial attribute analysis rule configuration can support the configuration of a list database and a similarity threshold, as illustrated in the diagram below. Figure 4B As shown.

[0127] The list database is used to define the list database used to verify the identity of the identified faces.

[0128] For example, assuming the list is configured as list A and the similarity threshold is configured as 88%, if the similarity between the identified face and any face in list A exceeds 88%, then the face is determined to meet the face attribute analysis rules.

[0129] For example, when the strategy is configured as a people-based strategy, the face attribute analysis rule configuration can support the configuration of the face count range and the detection area (i.e., the count detection area), as illustrated in the diagram below. Figure 4C As shown.

[0130] For example, the detection area may include the entire image or the target bounding box.

[0131] For example, when the strategy is configured as a feature search strategy, the facial attribute analysis rule configuration can support configurations for gender, age group, glasses (whether glasses are worn), masks (whether masks are worn), and detection areas, as illustrated in the diagram below. Figure 4D As shown.

[0132] As an example, for the human body, various different strategies can be configured, such as the number of people strategy, aggregation strategy, or feature search strategy.

[0133] For example, when the strategy is configured as a people-based strategy, the human attribute analysis rule configuration can support the configuration of the people range and the detection area, as illustrated in the diagram below. Figure 5A and Figure 5B As shown.

[0134] in, Figure 5A and Figure 5B These are schematic diagrams of the human attribute analysis rule configuration interface, showing the cases where the detection area is the entire image and the target bounding box, respectively. For the case where the detection area is the target bounding box, the quantity detection area can be determined by expanding the human detection box (target bounding box) outwards, and then it can be determined whether the number of people within the quantity detection area meets the requirements.

[0135] For example, when detecting a cyclist, the detected human body is usually the person in the front row. The bounding box of this person can be expanded to determine the number of people in the expanded bounding box, thereby determining whether there is a "cyclist carrying a passenger" behavior.

[0136] For example, when the strategy is configured as an aggregation strategy, the configuration of human attribute analysis rules can support similarity thresholds, time intervals, frequency of occurrence, and monitoring point configurations, as illustrated in the diagram below. Figure 5C As shown.

[0137] For example, when the strategy is configured as a feature search strategy, the human attribute analysis rule configuration can support configurations such as detection area, gender, upper garment, lower garment, backpack (whether a backpack is worn), carrying items (whether items are carried), hat (whether a hat is worn), and cycling (whether cycling is done). A diagram illustrating this can be seen as follows: Figure 5D and Figure 5E As shown.

[0138] As an example, for vehicles, various different strategies can be configured, such as quantity strategies, feature search strategies, or vehicle orientation strategies.

[0139] For example, when the strategy is configured as a quantity strategy, the vehicle attribute analysis rule configuration can support the configuration of vehicle quantity range and detection area, as illustrated in the diagram below. Figure 6A and 6B As shown.

[0140] For example, when the strategy is configured as a feature search strategy, the vehicle attribute analysis rule configuration can support configurations such as license plate number, vehicle color, vehicle brand, detection area, vehicle model, pendant (whether there is a pendant), cardboard box (whether there is a cardboard box), passenger seat, phone call, sun visor (whether the sun visor is open), and ornament (whether there is an ornament). A diagram can be shown as follows: Figure 6C As shown.

[0141] For example, when the strategy is configured as a vehicle orientation strategy, the vehicle attribute analysis rule configuration can support vehicle orientation configuration, as illustrated in the diagram below. Figure 6D As shown.

[0142] Among them, the vehicle configuration options can include the front, rear, or unknown (i.e., not limited to the front or rear).

[0143] As an example, for non-motorized vehicles, various different strategies can be configured, such as quantity strategies or feature search strategies.

[0144] For example, when the strategy is configured as a quantity strategy, the non-motorized vehicle attribute analysis rule configuration can support the configuration of quantity range and detection area, as illustrated in the diagram below. Figure 7A and Figure 7B As shown.

[0145] For example, when the strategy is configured as a feature search strategy, the non-motorized vehicle attribute analysis rule configuration can support vehicle color, detection area, vehicle type (such as two-wheeled or three-wheeled vehicle), vehicle category (such as bicycle or motorcycle (electric vehicle)), number of riders, front frame (whether there is a front frame), helmet (whether a helmet is worn), etc., as illustrated in the diagram. Figure 7C and Figure 7D As shown.

[0146] In one example, the AIOP algorithm rules can support the configuration of various strategies, such as OCR algorithm strategy, abnormal behavior detection strategy, or regional statistics strategy.

[0147] Taking OCR algorithm strategies as an example, the AIOP algorithm rule configuration can support the configuration of detection regions and sensitive words, as illustrated in the diagram below. Figure 8 As shown.

[0148] For example, such as Figure 8 As shown, the sensitive word configuration can include warning sensitive word configuration and removal sensitive word configuration. Warning sensitive words are sensitive words that need to trigger warnings (also known as positive sensitive words), and removal sensitive words are sensitive words that need to have their warnings canceled (also known as negative sensitive words).

[0149] It should be noted that, in the embodiments of this application, the rule algorithm type is not limited to the text-to-image search type, visual question answering type, target attribute analysis type and AIOP algorithm type described in the above embodiments, but may also include other algorithm types, such as voice analysis type.

[0150] In one example, for speech analysis rules (i.e., screening rules for speech analysis types), it is possible to support the configuration of a variety of different strategies, such as speech keyword strategies.

[0151] For example, when the strategy is configured as a voice keyword strategy, the voice analysis rules support the configuration of keywords (such as "danger", "help", etc.), configuration modes (such as exact match or fuzzy match, etc.), and keyword priorities (such as high / medium / low, etc.).

[0152] Keywords can be entered by the user or imported; keyword priority can be used to determine the triggering order of linkage rules. Based on the text matched by the keywords, the linkage order can be determined according to the configured priority.

[0153] For example, it supports real-time voice stream monitoring, dynamically displays the speech-to-text results of the current audio stream, and highlights the hit keywords.

[0154] For example, the audio stream can be captured by a camera that supports audio capture, and the audio captured by the camera can be converted into text information, which can then be stored as text modeling.

[0155] For example, the voice analysis rule configuration can also provide testing functionality, allowing users to input simulated audio to verify the validity of the rules.

[0156] As can be seen, in the embodiments of this application, the data screening method of combining large models (such as multimodal large models) with small models (such as object detection models) can improve the accuracy and reliability of data screening.

[0157] For example, users can choose to use large model rules (such as text-to-image search rules, visual question answering rules) and / or small model rules (such as target attribute analysis rules) for data screening according to their actual needs.

[0158] It should be noted that during the dynamic configuration of rules, a large rule generation model can be used to generate rules. This large rule generation model can automatically analyze and determine which large-model rules and / or small-model rules to use based on user needs (such as language input).

[0159] For example, for the configuration of screening rules of any rule algorithm type, the corresponding screening rules can be generated based on the configuration parameters of the screening rules of that rule algorithm type (set by the user through the above configuration interface).

[0160] In one example, the above-mentioned configuration parameters for generating corresponding screening rules based on the selected rule algorithm type can include: When the selected rule algorithm type includes multiple rule algorithm types, the corresponding screening rules are generated based on the configuration parameters of the screening rules for these multiple rule algorithm types.

[0161] For example, during the configuration of a screening rule, one is not limited to selecting a single rule algorithm type. Instead, multiple rule algorithm types can be selected according to requirements, and configuration parameters for each of these multiple rule algorithm types can be set to generate a screening rule that combines configuration parameters for multiple rule algorithm types.

[0162] For example, during the configuration of a screening rule, you can select a facial attribute analysis algorithm type and a human attribute analysis algorithm type, and set the configuration parameters (such as facial features) for the facial attribute analysis algorithm type and the configuration parameters (such as human features) for the human attribute analysis algorithm type, respectively. This generates a screening rule that combines the configuration parameters of the facial attribute analysis algorithm type and the human attribute analysis algorithm type. When running this screening rule, hitting the target requires meeting both the facial features and the human features configured in the screening rule.

[0163] In one example, the rule algorithm type includes the target attribute analysis type; The configurable items for screening rules of the target attribute analysis type can include the target number, which indicates the number of targets detected in the screening results that match the screening rule.

[0164] For example, taking the rule algorithm type as the target attribute analysis type, during the process of generating screening rules, configurable items may include the target quantity, such as the number of faces, the number of people, or the number of vehicles.

[0165] For example, the target quantity can be set to a single value or a range of quantities.

[0166] As an example, the configurable options for screening rules of the target attribute analysis type can also include a quantity detection region configuration, which can include a target bounding box or a full map.

[0167] In some embodiments, the rule dynamic configuration scheme provided in this application may further include: The list of rules to be issued is displayed through a visual interface; When the list of rules to be issued includes multiple screening rules, the visualization interface displays the execution order of these multiple screening rules, as well as the logical relationship between them.

[0168] For example, to allow relevant personnel to more intuitively view the screening results included in the list of rules to be issued, the system can display the list of rules to be issued through a visual interface, for example, through... Figure 2B The interface showing the list of rules to be issued displays the list of rules to be issued.

[0169] The interface displaying the list of rules to be issued can show the screening rules included in the list, and, if multiple screening rules are included, the execution order of these rules and the logical relationship between them. A diagram illustrating this can be shown below. Figure 9 or Figure 10 As shown.

[0170] For example, arrows can be used to indicate the execution order of screening rules.

[0171] For example, the logical relationship between screening rules may include a first type of logical relationship or a second type of logical relationship.

[0172] For example, when the logical relationship between the current screening rule and the previous screening rule is a first type of logical relationship (the first type of logical relationship can be called "addition"), the screening result after running the current screening rule is the screening result that satisfies the current screening rule among the screening results after running the previous screening rule. That is, running the current screening rule can further determine the screening result that satisfies the current screening rule from the screening results after running the previous screening rule, and the determined screening result is retained.

[0173] When the logical relationship between the current screening rule and the previous screening rule is the second type of logical relationship, the screening result after running the current screening rule is the screening result that does not meet the current screening rule among the screening results after running the previous screening rule. That is, running the current screening rule can further determine the screening results that meet the current screening rule from the screening results after running the previous screening rule, and filter the determined screening results.

[0174] In one example, the rule dynamic configuration scheme provided in this application embodiment may further include: When the list of rules to be issued includes multiple screening rules, the execution order of each screening rule in the list of rules to be issued is adjusted according to the operation instructions for the screening rules in the list of rules to be issued, and / or the logical relationship between the multiple screening rules. The screening results are updated based on the execution order of the adjusted screening rules and / or the logical relationship between the multiple screening rules.

[0175] For example, users can adjust the execution order of screening rules in the list of rules to be issued displayed in the visual interface by dragging and dropping them.

[0176] In addition, users can adjust the logical relationships between screening rules in the list of rules to be issued. For example, they can change the first type of logical relationship to the second type of logical relationship, or change the second type of logical relationship to the first type of logical relationship.

[0177] In one example, the rule dynamic configuration scheme provided in this application embodiment may further include: Based on the deletion command for any screening rule in the list of rules to be issued, delete the screening rule from the list of rules to be issued; If there are multiple remaining screening rules in the list of rules to be issued, update the execution order of these multiple screening rules, as well as the logical relationship between these multiple screening rules.

[0178] For example, users can delete screening rules added to the list of rules to be issued according to their actual needs.

[0179] Accordingly, if a deletion instruction is detected for any screening rule in the list of rules to be issued, the screening rule can be deleted from the list of rules to be issued. If there are multiple remaining screening rules, the execution order of the remaining multiple screening rules and the logical relationship between the multiple screening rules can be updated.

[0180] For example, suppose the list of rules to be issued initially includes screening rules 1 to 3, and the execution order of each screening rule is screening rule 1, screening rule 2, and screening rule 3. The logical relationship between screening rule 1 and screening rule 2 is a first-type logical relationship, and the logical relationship between screening rule 2 and screening rule 3 is a second-type logical relationship. If a deletion instruction for screening rule 2 is detected, screening rule 2 can be deleted from the list of rules to be issued. For the remaining screening rules (screening rule 1 and screening rule 3), the execution order is to execute screening rule 1 first and then screen rule 3. The logical relationship between screening rule 3 and screening rule 1 is a second-type logical relationship.

[0181] In one example, the rule dynamic configuration scheme provided in this application embodiment may further include: The logical type of the screening rule includes a first logical type or a second logical type; the screening rule of the first logical type is used to screen and retain the screening results that match the screening rule; the screening rule of the second logical type is used to screen and filter the screening results that match the screening rule. For non-first screening rules added to the list of rules to be issued, the logical relationship between the screening rule and the previous screening rule is determined based on the logical type of the screening rule.

[0182] For example, for configurable items including prompt words in the screening rules (such as the text-to-image search rules or visual question-and-answer rules mentioned above), when the prompt word is a positive prompt word, the logical type of the screening rule is the first logical type, and when the prompt word is a negative prompt word, the logical type of the screening rule is the second logical type.

[0183] For example, when a screening rule is added to the list of rules to be issued, for a screening rule that is not the first one in the list, the logical relationship between the screening rule and the previous screening rule can be determined based on the logical type of the screening rule.

[0184] For example, if the logical type of the screening rule is the first logical type, the logical relationship between the screening rule and the previous screening rule can be a first type logical relationship; if the logical type of the screening rule is the second logical type, the logical relationship between the screening rule and the previous screening rule can be a second type logical relationship.

[0185] In one example, for any screening rule in the list of rules to be issued displayed through a visual interface, the number of screening results after running that screening rule is shown in the visual interface. A diagram illustrating this can be shown as follows: Figure 9 or Figure 10 As shown.

[0186] In one example, for any screening rule in the list of rules to be issued displayed through a visual interface, the configuration parameters of that screening rule are displayed through the visual interface, as illustrated in the diagram below. Figure 10 As shown.

[0187] In some embodiments, the rule dynamic configuration scheme provided in this application may further include: For any given screening rule, the screening results after running the rule are displayed through a visual interface.

[0188] For example, it can be done through Figure 2A The screening results display interface shown displays the screening results after running the screening rules.

[0189] In one example, displaying the screening results after running the screening rule through a visual interface could include: The visual interface displays the data that meets the criteria after running the screening rule, and / or the visual interface displays the data that has been filtered after running the screening rule.

[0190] For example, the display of screening results may include the display of data that meets the criteria, and / or the display of filtered data.

[0191] For example, such as Figure 10 As shown, the screening results display interface can include a data display interface for data meeting the criteria and a data display interface for filtered data. Users can choose to switch the screening results display interface to the data display interface for data meeting the criteria to view the screening results retained after running the current screening rule, or switch to the data display interface for filtered data to view the screening results filtered after running the current screening rule, according to their actual needs.

[0192] In one example, the screening results displayed in the visualization interface include monitoring point information; After displaying the screening results after running the screening rule through a visual interface, it may also include: When a removal command is detected, the monitoring point to be removed is removed; the removal command carries the identifier of the monitoring point to be removed; the removal command is used to instruct the removal of the monitoring point corresponding to the identifier of the monitoring point to be removed.

[0193] For example, users can select monitoring points that are irrelevant to the current application scenario based on the screening results displayed in the visual interface, and trigger a removal command for the selected monitoring points (which can be called a removal command for monitoring points).

[0194] For example, the instruction to remove a monitoring point may carry the identifier of the monitoring point to be removed (i.e., the identifier information of the monitoring point to be removed, such as the monitoring point name or ID).

[0195] When the system detects a command to remove a monitoring point, it can remove the monitoring point according to the identifier of the monitoring point to be removed carried in the command.

[0196] For example, for a removed monitoring point, the screening result for that monitoring point will no longer be displayed on the screening result display interface (e.g., it will no longer be displayed on the screen). Figure 10 (As shown in the data display interface that meets the conditions), and the system no longer needs to perform early warning analysis on the data of that monitoring point.

[0197] For example, the system can record the removed monitoring points in a list (which can be called the removed monitoring point list). Users can view the removed monitoring points through this list and choose to restore the monitoring points according to actual needs, for example, by clicking... Figure 11 The "Restore" button on the interface shown restores the removed monitoring points.

[0198] As an example, the rule dynamic configuration scheme provided in this application embodiment may further include: When a command to restore a monitoring point is detected, the monitoring point to be restored is restored. The command to restore a monitoring point carries the monitoring point to be restored and is used to instruct the monitoring point corresponding to the identifier of the monitoring point to be restored to be restored.

[0199] For example, after a removed monitoring point is restored, the screening results of that monitoring point can be displayed again on the screening results display interface, and the system can restart the early warning analysis of the data of that monitoring point.

[0200] In some embodiments, after displaying the screening results after running the screening rule through a visual interface, the process may further include: If the operation focus is detected to be in the area corresponding to any screening result, a result card for that screening result is output; wherein, the result card includes at least one operation function option for that screening result; Based on the detected selection instruction for a first operation function option among at least one operation function option, the function processing corresponding to the first operation function option is performed based on the screening result.

[0201] For example, the location of the focus of operation may include the operation cursor (such as the mouse cursor).

[0202] For example, for screening results displayed through a visual interface, the operation focus can be moved to any screening result, such as by hovering the mouse cursor over the display area corresponding to a screening result in the visual interface, to trigger the display of a result card for that screening result. Through this result card, specific functional processing for that screening result can be initiated.

[0203] Accordingly, after displaying the screening results through a visual interface, if the focus of the operation is detected to be in the area corresponding to any screening result, a result card for that screening result can be output.

[0204] For example, the results card may include at least one operational function option for the screening result.

[0205] For example, the above-mentioned operation function options may include, but are not limited to, one or more functions such as video playback, human image search, adding to the clue database, and identity verification. For instance, its illustration may be as follows: Figure 10 As shown.

[0206] Please see Figure 12 This is a flowchart illustrating an early warning method provided in an embodiment of this application, as shown below. Figure 12 As shown, the early warning method may include the following steps: Step S1210: For any application scenario, obtain the data to be analyzed.

[0207] For example, for any application scenario, the data to be analyzed can be determined based on an analysis task created for that application scenario.

[0208] For example, analysis tasks may include, but are not limited to: video structured analysis tasks, third-party analysis tasks, timed snapshot analysis tasks, and local image analysis tasks.

[0209] For example, a video structured analysis task involves intelligently analyzing video streams or video files to extract structured metadata (timestamps, target types, attribute features, trajectory information, etc.) and generating event image data.

[0210] Third-party analysis tasks: Tasks that call external third-party algorithms or services to process and analyze media data.

[0211] Scheduled image capture and analysis task: A task type that automatically captures and analyzes images according to a preset time schedule.

[0212] Local image analysis task: A task that performs batch or single analysis on locally stored static image files.

[0213] For example, for any application scenario, the data specified in the analysis task can be identified as the data to be analyzed.

[0214] Step S1220: Analyze and process the data to be analyzed according to the early warning rules of this application scenario.

[0215] For example, the warning rules for this application scenario can be generated in the manner described in the above embodiments.

[0216] Step S1230: If it is determined that there is data in the data to be analyzed that meets the screening rules included in the early warning rules, early warning processing is performed on the data.

[0217] In some embodiments, the analysis and processing of the data to be analyzed based on the early warning rules of the application scenario may include: Determine the algorithms corresponding to the multiple screening rules included in the early warning rules for this application scenario; The data to be analyzed is processed using algorithms corresponding to multiple screening rules included in the early warning rules of this application scenario, in a serial and / or parallel manner.

[0218] For example, the algorithms corresponding to the screening rules may include, but are not limited to, target (face, human body, vehicle or non-motorized vehicle) detection algorithms, AIOP algorithms, speech-to-text algorithms or visual question answering algorithms.

[0219] For example, the algorithm to be used can be determined based on the algorithm type of the screening rule.

[0220] In some embodiments, after the above-mentioned warning processing for the data, the method further includes: This data will be displayed on the clue analysis interface; The early warning scheme provided in this application embodiment may further include: If the operation focus is detected to be in the area corresponding to any warning result, an alarm card is output for that warning result; wherein, the alarm card includes at least one operation function option for that warning result; If a selection instruction for a second operation function option among the at least one operation function option is detected, the function processing corresponding to the second operation function option is executed.

[0221] For example, image data that triggers an alert can be displayed on the clue analysis interface so that the alert can be determined as a positive or false alarm during the analysis process.

[0222] For example, for the warning results displayed on the clue analysis interface, the operation focus can be moved to any warning result. For example, by hovering the mouse cursor over the display area corresponding to a certain warning result in the clue analysis interface, the display of the alarm card (also called the warning card) for that warning result can be triggered. Through the alarm card, specific functional processing for that warning result can be initiated.

[0223] Accordingly, after displaying the warning results through a visual interface, if the focus of the operation is detected to be in the area corresponding to any warning result, an alarm card can be output for that warning result.

[0224] For example, the alarm card may include at least one operational function option for the warning result.

[0225] For example, the above-mentioned operation function options may include, but are not limited to, one or more functions such as positive warning (used to determine if the warning result is positive), false warning (used to determine if the warning result is false), and video playback. For example, its schematic diagram may be as follows: Figure 13 As shown.

[0226] It should be noted that, in this embodiment of the application, the warning results displayed through the clue analysis interface can also support batch processing of positive or false alarms, removal of monitoring points, cancellation of issuance (for warning results determined to be false alarms, cancellation must be performed first, and then a second screening must be conducted), and export of all or part of the warning results.

[0227] To enable those skilled in the art to better understand the technical solutions provided in the embodiments of this application, the technical solutions provided in the embodiments of this application are described below with reference to specific examples.

[0228] In this embodiment, for any application scenario, the complete early warning process may include: I. Special Topic Creation.

[0229] For example, the system can support users to create topics and configure topic names, scenario descriptions, alert recipients, sharing permissions, etc.

[0230] For example, a corresponding topic can be created for any application scenario.

[0231] II. Create an analysis task.

[0232] For example, once a topic has been created, analysis tasks can be created. Analysis tasks can include, but are not limited to, video structured analysis tasks, third-party analysis tasks, timed snapshot analysis tasks, and local image analysis tasks.

[0233] For example, one or more analysis tasks can be created for any given topic.

[0234] III. Rule Configuration.

[0235] For example, for any analysis task, screening rules can be configured for that analysis task.

[0236] For example, the specific implementation process of the screening rule configuration can be found in the relevant description in the above embodiments.

[0237] IV. Dynamic Early Warning.

[0238] For example, early warning rules can be generated based on the configured and verified screening rules, and the generated early warning rules can be sent to the early warning terminal, which can then detect early warning events based on the early warning rules.

[0239] The following examples illustrate this point.

[0240] Scenario Example 1: Screening personnel who distribute small cards.

[0241] 1.1 Create a topic corresponding to this scenario.

[0242] 1.2 Create an analysis task.

[0243] 1.3 Rule Configuration.

[0244] For example, positive keyword filtering: Based on natural language descriptions (such as "people with their phone flashlights on"), a multimodal large model is used to initially retrieve video frames and obtain candidate results. After the search, rules are added and displayed on the left.

[0245] Reverse exclusion optimization: Use reverse suggestions (such as "cyclist" or "woman") to exclude obviously irrelevant results, improving filtering accuracy. Add rules after the search and display them on the left.

[0246] Multi-attribute joint screening: Combining attributes such as the number of people and vehicle type for secondary filtering, forming a funnel-shaped screening pipeline. Rules are added after the search and displayed on the left.

[0247] For example, since the distribution of small cards usually involves one person, and they are not usually cyclists, the data with multiple people can be filtered by combining the human body number strategy and the human body attribute analysis rules, and the data with the vehicle attribute analysis rules can be filtered by combining the vehicle type of bicycle or electric vehicle, so as to achieve secondary filtering of the data and further improve the accuracy of screening people distributing small cards.

[0248] 1.4 Dynamic early warning.

[0249] For example, early warning rules can be generated by combining verified screening rules.

[0250] For example, valid text-to-image search rules and target attribute analysis rules can be combined to generate early warning rules and sent to the early warning terminal, such as a video stream parsing system.

[0251] The video stream analysis system can intelligently analyze the video stream based on the received warning rules, and automatically trigger a warning and push it to the security platform when a video image that meets the warning rules is found.

[0252] Scenario Example 2: Screening drones.

[0253] 2.1 Create a topic corresponding to this scenario.

[0254] 2.2 Create an analysis task.

[0255] 2.3 Rule Configuration.

[0256] For example, text-based image search rules and visual question answering rules can be combined to determine whether a drone exists.

[0257] 2.3.1 Initial screening using text-based image search.

[0258] For example, input prompts (positive prompts): The user inputs a query command in natural language, such as "drone" or "aircraft".

[0259] Preliminary results: Using a multimodal large model, a candidate set containing possible targets is generated based on the above prompt words, providing a foundation for subsequent fine screening.

[0260] For example, for screening rules that have been determined to be effective, they can be added to the list of rules to be issued through the "Add Rule" function button in the screening rule configuration interface (which can be displayed on the left side of the screening rule configuration interface).

[0261] 2.3.2 Visual question-and-answer screening.

[0262] For example, the question-answering logic design is as follows: Based on the initial screening results, the system uses the semantic understanding capabilities of a multimodal large model to perform structured question-answering verification. For instance: Example of a problem: Does the drone have a rotor structure? Does the target have a typical UAV shape (such as a quadcopter or hexacopter)? "Should we exclude distractors (such as birds, cameras, umbrellas, etc.)?" For example, for screening rules where the visual cue words are positive (such as "Does a drone rotor structure exist?" or "Does the target have a typical drone shape (such as a quadcopter or a hexacopter)?"), if the visual question answering algorithm confirms "yes", the data is retained; otherwise, the data is filtered.

[0263] For the screening rule where visual cues are negative cues, data is filtered if the visual question-answering algorithm confirms "yes"; otherwise, the data is retained.

[0264] Fine screening process: For each candidate frame, the model generates an answer based on the question (such as "yes / no" or a specific attribute judgment).

[0265] Only targets that pass all question-and-answer verifications are retained, while false positives caused by similar objects (such as kites, birds, etc.) are excluded.

[0266] Results optimization: By gradually narrowing the target range through multiple rounds of question and answer, the screening accuracy is significantly improved.

[0267] 2.4 Dynamic early warning.

[0268] For example, early warning rules can be generated by combining verified screening rules.

[0269] For example, effective text-based image search rules (such as text-based image search rules that include the positive prompt word "drone") can be combined with visual question answering rules to generate early warning rules and send them to the early warning terminal, such as a video stream parsing system.

[0270] The video stream analysis system can intelligently analyze the video stream based on the received warning rules, and automatically trigger a warning and push it to the security platform when a video image that meets the warning rules is found.

[0271] Scenario Example 3: Screening for sensitive words on vehicle body.

[0272] 3.1 Create a topic corresponding to this scenario.

[0273] 3.2 Create an analysis task.

[0274] 3.3 Rule Configuration.

[0275] For example, text-based image search rules and AIOP algorithm rules can be combined to determine whether there are sensitive words related to vehicle bodies.

[0276] 3.3.1 Initial screening using text-based image search.

[0277] For example, input prompts (positive prompts): the user inputs a natural language description of "vehicles with text on the body", and preliminary screening results are retrieved through a multimodal large model.

[0278] 3.3.2. Refined screening of reverse prompt words.

[0279] Exemplarily, input reverse prompt words such as "bus", "taxi", and "engineering vehicle", and filter the retrieval results of the previous step through a multi-modal large model.

[0280] 3.3.3. Refined screening of AIOP algorithm.

[0281] Exemplarily, integrate the OCR algorithm to recognize the vehicle body text, and combine it with the sensitive word library for final screening.

[0282] Exemplarily, the sensitive word library can include early warning sensitive words and / or exclusion sensitive words.

[0283] Exemplarily, the early warning sensitive words can include words related to conflict or aggression; the exclusion sensitive words can include normal words with words related to conflict or aggression.

[0284] For example, the word "beat" belongs to the words related to aggression, but "take a taxi", "dial", "roll donkey jelly" and the like belong to normal words.

[0285] Exemplarily, for the determined effective screening rules, through the "Add Rule" function button in the screening rule configuration interface, add them to the list of rules to be issued (which can be displayed on the left side of the screening rule configuration interface).

[0286] 3.4. Dynamic early warning.

[0287] Exemplarily, based on the verified effective screening rules, combine and generate early warning rules.

[0288] For example, combine the image search by text rule including the positive prompt word "vehicle with text on the body", the image search by text rule including reverse prompt words (such as "bus", "taxi", and "engineering vehicle", etc.), and the above verified effective AIOP algorithm rules to generate an early warning rule (i.e., an early warning strategy), and send it to the early warning end, such as a video stream analysis system.

[0289] The video stream analysis system can perform intelligent analysis on the video stream according to the received early warning rules, and automatically trigger an early warning and push it to the security platform when it determines that there is a video image that meets the early warning rules.

[0290] Scenario Example 4. Screening for protecting schools and students.

[0291] 4.1. Create a special topic corresponding to this scenario.

[0292] 4.2. Create an analysis task.

[0293] 4.3. Rule configuration.

[0294] For example, timed image capture: Input image data of a specified monitoring point within a specified time range.

[0295] If there are no people or security guards wearing reflective vests (such as green reflective vests) at the school gate, it is considered that there are no school protection personnel and that it is necessary to call the police.

[0296] Joint screening of human attributes: secondary filtering is performed by combining human quantitative attribute strategies.

[0297] For example, school safety screening mainly checks whether there are school safety personnel (people wearing reflective vests or security guards, etc.) at designated locations (such as school gates) during school hours.

[0298] Since there are usually many people at the school gate during school hours, we can combine human body number strategy and human body attribute analysis rules to filter out data with a small number of people, such as data with a number of people not within the preset human body number threshold range (e.g., 3-100), to achieve secondary filtering of the data and further improve the accuracy of school safety screening.

[0299] 4.4 Dynamic early warning.

[0300] For example, early warning rules can be generated by combining verified screening rules.

[0301] For example, valid text-to-image search rules and target attribute analysis rules can be combined to generate early warning rules and sent to the early warning terminal, such as a video stream parsing system.

[0302] The video stream analysis system can intelligently analyze the video stream based on the received warning rules, and automatically trigger a warning and push it to the security platform when a video image that meets the warning rules is found.

[0303] The method provided in this application has been described above. The apparatus provided in this application is described below: Please see Figure 14 This is a schematic diagram of the structure of a rule dynamic configuration device provided in an embodiment of this application, as shown below. Figure 14 As shown, the rule dynamic configuration device may include: The generation unit is configured to generate a first screening rule according to a first rule configuration instruction; The screening unit is configured to screen the initial data to be screened according to the first screening rule based on the running instructions for the first screening rule, and obtain the first screening result; The generation unit is also configured to generate a second screening rule based on the second rule configuration instruction; The screening unit is also configured to, based on the running instructions for the second screening rule, re-screen the screening results after the previous screening rule has been run using the second screening rule, and obtain the second screening result.

[0304] For example, the specific implementation process of dynamic rule configuration by each functional unit in the rule dynamic configuration device can be found in the relevant descriptions in the above embodiments.

[0305] Please see Figure 15 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. The electronic device may include a processor 1501, a communication interface 1502, a memory 1503, and a communication bus 15015. The processor 1501, communication interface 1502, and memory 1503 communicate with each other via the communication bus 15015. The memory 1503 stores a computer program; the processor 1501 can execute the rule dynamic generation method described above by executing the program stored in the memory 1503.

[0306] The memory 1503 mentioned in this document can be any electronic, magnetic, optical, or other physical storage device that can contain or store information such as executable instructions, data, etc. For example, the memory 1503 can be: RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, storage drive (such as hard disk drive), solid-state drive, any type of storage disk (such as optical disc, DVD, etc.), or similar storage media, or combinations thereof.

[0307] This application also provides a computer-readable storage medium storing a computer program, such as... Figure 15 The memory 1503 in the computer program can be used by Figure 15 The processor 1501 in the electronic device shown executes to implement the rule dynamic generation method described above.

[0308] This application also provides a computer program stored in a computer-readable storage medium, such as... Figure 15 The memory 1503 in the memory, and when the processor executes the computer program, it causes the processor 1501 to execute the rule dynamic generation method described above.

[0309] This application embodiment also provides a rule dynamic configuration and early warning system, which may include: a rule configuration engine, a rule and algorithm repository, an early warning analysis and execution engine, and a monitoring device; wherein: The rule configuration engine is configured to dynamically configure rules in the manner described in the above embodiments.

[0310] For example, the rule configuration engine can be deployed on an application server (or application server cluster, which can be called a rule configuration server). This application server (or application server cluster) enables human-computer interaction by providing a graphical user interface (GUI) service to user clients. Users access the screening rule configuration interface provided by this engine (as described above) through a browser or dedicated client software. Figure 2A or Figure 2B The screening rule configuration interface is shown below. This interface visually presents rule building elements (such as parameter input (or selection) boxes, algorithm selectors, etc.). Users trigger and generate structured configuration instructions through operations on the interface. These configuration instructions are sent to the backend service of the rule configuration engine via the network, thereby driving the rule configuration engine to perform relevant processing for dynamic rule configuration based on the received configuration instructions, such as screening rule generation and screening rule execution.

[0311] The rules and algorithms repository is configured to store screening rules generated by the rule configuration engine, as well as algorithm entities that implement various analysis functions (such as large model algorithms and small model algorithms (such as object detection algorithms)), and to store the mapping relationship between screening rules and algorithms (used to establish dynamic links between screening rules and the algorithms required to execute the core computational logic of the rule).

[0312] For example, the rules and algorithms repository can be deployed using a high-performance, highly reliable distributed storage cluster (i.e., a rules and algorithms storage device).

[0313] Rule storage can take the form of a relational database cluster or a document-oriented database cluster; algorithm storage can take the form of an object storage service or a distributed file system.

[0314] The early warning analysis execution engine is configured to load early warning rules (which can be generated by combining multiple screening rules dynamically configured by the rule configuration engine) and call the corresponding algorithm to process the monitoring data for early warning.

[0315] For example, the early warning analysis execution engine can load early warning rules from the rule and algorithm repository based on the scheduling strategy or event triggering, and call the corresponding algorithm (such as calling a multimodal large model) to perform early warning analysis on the monitoring data of the monitoring equipment.

[0316] For example, the early warning analysis execution engine can be deployed in the form of a distributed computing cluster or a container orchestration cluster (which can be called an early warning analysis server). For instance, it can be deployed as one or more sets of computing node servers, or it can be packaged as a Docker container and deployed on a K8s cluster.

[0317] The surveillance equipment is configured to continuously or on-demand collect video or image data.

Claims

1. A method for dynamic rule configuration, characterized in that, include: Generate the first screening rule based on the first rule configuration instruction; Based on the execution instructions for the first screening rule, the initial data to be screened is screened using the first screening rule to obtain the first screening result; Generate a second screening rule based on the second rule configuration instructions; Based on the execution instructions for the second screening rule, the screening results after the previous screening rule is executed are screened again using the second screening rule to obtain the second screening result.

2. The method according to claim 1, characterized in that, Based on the rule configuration instructions, generate screening rules, including: Select the command based on the rule algorithm type, and the rule configuration interface for the selected rule algorithm type will be displayed; Based on the configuration instructions for the configurable items in the rule configuration interface, determine the configuration parameters of the screening rule for the selected rule algorithm type; Based on the configuration parameters of the selected rule algorithm type, generate the corresponding screening rules.

3. The method according to claim 2, characterized in that, The rule algorithm types include target attribute analysis types; The configurable items for the screening rule of the target attribute analysis type include the number of targets, which indicates the number of targets detected in the screening results that match the screening rule.

4. The method according to claim 3, characterized in that, The configurable options for the screening rules of the target attribute analysis type also include quantity detection region configuration, which includes target bounding boxes or the entire image.

5. The method according to claim 1, characterized in that, For any screening rule, after obtaining the screening results after running the screening rule, the following is also included: If a rule addition instruction is detected for the screening rule, the screening rule is added to the list of rules to be issued; the screening rules in the list of rules to be issued have an execution order and logical relationship.

6. The method according to claim 5, characterized in that, The method further includes: The list of rules to be issued is displayed through a visual interface; When the list of rules to be issued includes multiple screening rules, the visualization interface displays the execution order of the multiple screening rules and the logical relationship between the multiple screening rules.

7. The method according to claim 6, characterized in that, The method further includes: When the list of rules to be issued includes multiple screening rules, the execution order of each screening rule in the list of rules to be issued and / or the logical relationship between the multiple screening rules are adjusted according to the operation instructions for the screening rules in the list of rules to be issued. The screening results are updated based on the execution order of the adjusted screening rules and / or the logical relationship between the multiple screening rules.

8. The method according to claim 5, characterized in that, The method further includes: Based on the deletion instruction for any screening rule in the list of rules to be issued, delete the screening rule from the list of rules to be issued; If there are multiple remaining screening rules in the list of rules to be issued, update the execution order of these multiple screening rules and the logical relationship between them.

9. The method according to any one of claims 5-8, characterized in that, The logical relationships between screening rules include either the first type of logical relationship or the second type of logical relationship; When the logical relationship between the current screening rule and the previous screening rule is the first type of logical relationship, the screening result after running the current screening rule is the screening result that satisfies the current screening rule among the screening results after running the previous screening rule; When the logical relationship between the current screening rule and the previous screening rule is of the second type, the screening result after running the current screening rule is the screening result that does not meet the current screening rule among the screening results after running the previous screening rule.

10. The method according to any one of claims 5-8, characterized in that, The method further includes: The logical type of the screening rule includes a first logical type or a second logical type; the screening rule of the first logical type is used to screen and retain the screening results that match the screening rule; the screening rule of the second logical type is used to screen and filter the screening results that match the screening rule. For a non-first screening rule added to the list of rules to be issued, the logical relationship between the screening rule and the previous screening rule is determined based on the logical type of the screening rule.

11. The method according to claim 6, characterized in that, For any screening rule in the list of rules to be issued displayed through the visualization interface, the visualization interface displays at least one of the following: the configuration parameters of the screening rule, the number of screening results after running the screening rule, and the number of screening results after running the screening rule.

12. The method according to claim 11, characterized in that, The process of displaying the screening results after running the screening rule through a visual interface includes: The visual interface displays the data that meets the criteria after running the screening rule, and / or the visual interface displays the data that has been filtered after running the screening rule.

13. The method according to claim 11, characterized in that, The screening results displayed in the visualization interface include monitoring point information; After displaying the screening results after running the screening rule through a visual interface, the method further includes: Upon detecting a command to remove a monitoring point, the monitoring point to be removed is removed; the command to remove a monitoring point carries the identifier of the monitoring point to be removed; the command to remove a monitoring point is used to instruct the removal of the monitoring point corresponding to the identifier of the monitoring point to be removed.

14. The method according to claim 13, characterized in that, The method further includes: Upon detecting a command to restore a monitoring point, the monitoring point to be restored is restored. The command to restore a monitoring point carries an identifier of the monitoring point to be restored, and the command is used to instruct the restoration of the monitoring point corresponding to the identifier of the monitoring point to be restored.

15. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method described in any one of claims 1-14.

Images