Alarm filtering method and system based on visual anomaly detection algorithm
By configuring the filter chain in the visual anomaly detection algorithm to filter alarm information in the time, type and frequency dimensions, the frequent alarm problem is solved, the user experience and system efficiency are improved, and the key alarms are not missed.
Patent Information
- Application Number
- CN202410951264.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-12
AI Technical Summary
The existing visual anomaly detection algorithm cannot effectively merge multiple identical alarms when an exception is detected, resulting in users receiving a large number of repeated alarm information, causing interference, and it is difficult to calculate and deduplicate based on multiple exception dimensions.
Using an alarm filtering method and system based on visual anomaly detection algorithm, filtering the corresponding time dimension information, alarm type dimension information or alarm frequency dimension by configuring a filter chain, including a time window filter, an alarm type filter and a sliding window filter Information, merge the same type of alarm information, give priority to sending alarms with high risk factors, and evaluate the filter chain effect based on the alarm label and response results for optimization.
Effectively reduce frequent and unnecessary alarms, avoid missing key alarm information, increase users' attention to key alarms, improve the efficiency and user experience of the overall visual anomaly detection system, and enhance the adaptability and scalability of the system.
Smart Images

Figure CN118942222A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of alarm filtering, and more particularly to an alarm filtering method and system based on a visual anomaly detection algorithm. Background Art
[0002] Visual anomaly detection is a method that uses computer vision technology to detect abnormal behaviors or events in images or videos. It can help people quickly identify potential security threats, equipment failures and other problems. It is of great significance for video stream monitoring systems, intelligent security systems, industrial production monitoring systems, etc. Especially in the visual anomaly detection algorithm based on video streams, notifying users of the alarm information of abnormal results will effectively prompt users.
[0003] At present, when the existing algorithm model detects an anomaly, it will directly notify the user. It is unable to perform calculations based on multiple anomalies and other dimensions to reduce the interference of alarms to users. Specifically, the existing algorithm model service needs to detect anomalies for each frame of the video. Usually, there are more than 24 frames of pictures in the video stream. If the anomaly in the video appears for one second or longer, the anomaly detection algorithm service will continuously detect 24 anomalies or more. If the alarm information of the abnormal result is notified to the user directly, multiple alarm information will be sent to the user for one abnormal result. Such excessive alarms will interfere with the user.
[0004] Therefore, how to improve the accuracy of the visual anomaly detection algorithm while merging multiple identical alarms to solve the frequent alarm problem in the visual anomaly detection algorithm and reduce unnecessary interference to users is an issue that technical personnel in this field urgently need to solve. Summary of the invention
[0005] In view of the above problems, the present invention provides an alarm filtering method and system based on a visual anomaly detection algorithm to at least solve some of the technical problems mentioned in the above background technology.
[0006] In order to achieve the above object, the present invention adopts the following technical solution:
[0007] On the one hand, the present invention provides an alarm filtering method based on a visual anomaly detection algorithm, comprising the following steps:
[0008] Detect the video data in the target usage scenario based on the visual anomaly detection algorithm; input the currently detected first alarm information to the starting end of the configured filter chain, and output the filtered second alarm information and the corresponding alarm label;
[0009] The filter chain configuration steps include:
[0010] (1) configuring a filter type, a SPELS expression, and an alarm type and an alarm condition in the SPELS expression; and forming a filter point based on the SPELS expression;
[0011] (2) According to the configured filter type, a corresponding filter is selected to form a filter chain based on the obtained filter points; each filter is responsible for filtering the corresponding time dimension information, alarm type dimension information or sliding window dimension information.
[0012] Furthermore, the corresponding time window dimension information is filtered, including:
[0013] The number of alarm messages generated within a preset time period under the target usage scenario is set to n, the number of times the same type of alarm messages are allowed to be sent within the preset time period is i, and the ni-th alarm message of the same type generated within the preset time period is filtered.
[0014] Furthermore, the corresponding alarm type dimension information is filtered, including:
[0015] Filter out the alarm information corresponding to the alarm type whose risk factor is less than the threshold;
[0016] Alternatively, according to the risk factors of different types of alarm types, the alarm type information with a high risk factor is sent to the client first;
[0017] Alternatively, on the basis of filtering out the alarm information corresponding to the alarm type with a risk factor less than the threshold, the alarm type information with a high risk factor is preferentially sent to the client.
[0018] Furthermore, the corresponding sliding window dimension information is filtered, specifically including:
[0019] The same alarm information generated within the second preset time period is merged.
[0020] Furthermore, it also includes: when the warning label meets the preset warning level, sending the second warning information corresponding to the warning label to the user end.
[0021] Furthermore, it also includes: according to the second alarm information, sending corresponding response measures to the user end.
[0022] Furthermore, it also includes: evaluating the filtering effect of the filter chain according to the second alarm information and the corresponding response result, and optimizing the filter chain according to the evaluation result.
[0023] On the other hand, the present invention further provides an alarm filtering system based on a visual anomaly detection algorithm, characterized in that, by applying any of the above-mentioned methods, the system comprises a configuration module and an alarm filtering module;
[0024] The configuration module is used to configure the filter type, the SPEEL expression, and the alarm type and alarm condition in the SPEEL expression; form a filter point based on the SPEEL expression; select the corresponding filter according to the configured filter type and form a filter chain according to the obtained filter points; each filter is responsible for filtering the corresponding time window, alarm type and / or alarm frequency;
[0025] The alarm filtering module is used to input the first alarm information currently detected based on the visual anomaly detection algorithm to the starting end of the configured filter chain, and output the second alarm information after filtering and the corresponding alarm label.
[0026] Furthermore, it also includes an alarm module;
[0027] The alarm module is used to send the second alarm information to the user end when the alarm tag meets the preset alarm level; and to send corresponding response measures to the user end according to the second alarm information.
[0028] Further, an optimization module is included;
[0029] The optimization module is used to evaluate the filtering effect of the filter chain according to the second alarm information and the corresponding response result, and optimize the filter chain according to the evaluation result.
[0030] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses an alarm filtering method and system based on a visual anomaly detection algorithm, which has the following beneficial effects:
[0031] The present invention filters the corresponding time window dimension information, alarm type dimension information or alarm frequency dimension information respectively through a time window filter, an alarm type filter and a sliding window filter, thereby reducing frequent alarms and unnecessary alarms while avoiding missing key alarm information; thereby increasing the user's attention to key alarm information, while also reducing the user's burden and improving the overall performance of the visual anomaly detection system and user experience.
[0032] The present invention introduces a filter chain, and the configuration of the filter chain can be flexibly adjusted according to different monitoring scenarios and requirements, so that the system can adapt to various environments and conditions, thereby enhancing the adaptability and scalability of the system.
[0033] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention.
[0034] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0036] Figure 1 A schematic flow chart of an alarm filtering method based on a visual anomaly detection algorithm provided in an embodiment of the present invention.
[0037] Figure 2 A schematic diagram of the framework of an alarm filtering system based on a visual anomaly detection algorithm provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0039] Embodiment 1:
[0040] See also Figure 1 As shown, the embodiment of the present invention discloses an alarm filtering method based on a visual anomaly detection algorithm, comprising the following steps:
[0041] Input the first alarm information currently detected by the visual anomaly detection algorithm to the starting end of the configured filter chain, and output the filtered second alarm information and the corresponding alarm label;
[0042] The filter chain configuration steps include:
[0043] (1) Configure the filter type, Spel expression, and parameters such as the alarm type and alarm condition in the Spel expression. Then, filter points are formed based on the Spel expression. The Spel expression can be expressed as:
[0044] {"tag":"Filter","editType":"expressionParam","expression":"#contain('555,666,777,888',#root.getInt('anomaly_code'))","enableStatus":1,"visibleStatus":1,"expressionParam":[]}
[0045] For example, if the filter is a Boolean type filter, the filter will get the spel expression (#contain('555,666,777,888',#root.getInt('anomaly_code'))) of the current Boolean value and the alarm type latitude information in the alarm data;
[0046] (2) According to the configured filter type, select the corresponding filter to form a filter chain based on the obtained filter points;
[0047] Each filter is responsible for processing the corresponding alarm dimension information, such as filtering the corresponding time dimension information, alarm type dimension information or alarm frequency dimension information; in the specific filtering process, a spel expression executor can be created, and the current spel expression is executed through the spel expression executor, and whether the filter condition is met is obtained according to the alarm type and expression in the alarm data. If the filter condition is met, the Boolean filter point returns true, and the current alarm content is filtered; among them,
[0048] (1) Configure the time window filter to filter the corresponding time window dimension information, including:
[0049] Set the number of alarm messages generated within a preset time period in the target usage scenario to n, the number of times the same type of alarm messages are allowed to be sent within the preset time period to i, and filter the ni-th alarm message of the same type generated within the preset time period. For example, in the traffic management scenario, assume that each channel ID represents a camera; the shooting range of a preset number of cameras is used as a scene group, and each scene group is set to send only one alarm message within 1 minute. The time filter will limit only one alarm to be passed in 1 minute based on the time dimension obtained through the spel expression; if 5 alarm messages of the same type are generated within this 1 minute, the first 4 alarms will be filtered out. Based on this, the first 4 alarms with low preparation can be avoided, reducing the system false alarm rate.
[0050] (2) Configure the alarm type filter to filter the corresponding alarm type dimension information, including:
[0051] Filter out the alarm information corresponding to the alarm type with a risk factor less than the threshold; for example, the preset alarm types in the target usage scenario include: fall abnormality, absence from post abnormality, personnel crowding abnormality, etc., and their corresponding risk factors are 5, 5, and 3 respectively; set the threshold to 4, so as to filter out the alarm information corresponding to the excessive personnel abnormality; based on this, unnecessary alarm type dimension information can be filtered out according to the actual situation of the user;
[0052] Alternatively, according to the risk factors of different types of alarms, the alarm types with high risk factors are sent to the client first; for example, the risk factors of intrusion anomaly, absence from post anomaly and crowding anomaly are 6, 5 and 3 respectively, that is, the alarm information of intrusion anomaly is sent first, which is actually the alarm information of absence from post anomaly, and the alarm information of crowding anomaly is sent last;
[0053] Alternatively, on the basis of filtering out the alarm information corresponding to the alarm type with a risk factor less than the threshold, the alarm type information with a high risk factor is preferentially sent to the client.
[0054] (3) Configure the sliding window filter to filter the corresponding sliding window dimension information, including:
[0055] The same alarm information generated within the second preset time period is merged to reduce the occurrence of continuous repeated alarms.
[0056] The above-mentioned time window filter, alarm type filter and sliding window filter can be used together to reduce frequent alarms and unnecessary alarms while avoiding missing key alarm information; thereby increasing the user's attention to key alarm information, while also reducing the user's burden and improving the overall performance of the visual anomaly detection system and user experience.
[0057] Next, the relevant contents of the above-mentioned alarm filtering method based on the visual anomaly detection algorithm are described in detail.
[0058] 1. Target usage scenarios
[0059] In the embodiments of the present invention, the above-mentioned target usage scenarios can be public place video surveillance scenarios, traffic management scenarios, industrial production control scenarios, environmental monitoring scenarios, artificial intelligence-assisted diagnosis scenarios, etc.; specifically: for public place video surveillance scenarios, it can help detect abnormal situations such as crowding, falls, disputes, intrusions, etc., and assist in public security management; for traffic management scenarios, it can help detect abnormal situations such as stranded personnel, vehicle congestion, traffic accidents, traffic violations, etc., and assist traffic management departments to deal with them in a timely and effective manner; for industrial production control scenarios, it can help detect abnormal situations such as equipment failures, production line blockages, and process parameter abnormalities, and assist in factory safety production and operation; for environmental monitoring scenarios, embodiments of the present invention can help detect abnormal situations such as air quality monitoring and water quality monitoring, and help environmental regulatory departments take corresponding measures to protect the environment and public health; for artificial intelligence-assisted diagnosis scenarios, such as artificial intelligence-assisted medical testing, embodiments of the present invention can help detect abnormal situations in medical images and assist medical staff in their work.
[0060] 2. Middleware involved
[0061] In the process of filtering alarm information, the middleware involved is the spel expression; the SpEL (Spring Expression Language) is a powerful expression language, usually used with the Spring framework, for dynamically evaluating expressions at runtime. In an embodiment of the present invention, the SpEL expression can be used to configure the filter chain to achieve flexible filtering and management of abnormal alarms. The following is a possible usage of the SpEL expression in the present invention:
[0062] (1) Time filter conditions: SpEL expressions can be used to define time filter conditions, such as notifying users only of abnormalities detected within a specific time period. For example, you can use SpEL expressions to write conditions like time>='08:00:00'&&time<='18:00:00', which means that users will only be notified of abnormalities detected during working hours;
[0063] (2) Exception type filtering conditions: SpEL expressions can also be used to define exception type filtering conditions, such as notifying users of only certain types of exceptions. For example, you can use SpEL expressions to write conditions like type == 'intrusion behavior', which means that users will only be notified if intrusion behavior anomalies are detected.
[0064] (3) Abnormal duration condition: SpEL expressions can also be used to define abnormal duration conditions, such as notifying the user only of abnormal situations whose duration exceeds a certain threshold. For example, you can use SpEL expressions to write a condition similar to duration>=5, which means that only abnormal situations that last more than 5 seconds will be notified to the user.
[0065] (4) Comprehensive condition filtering: SpEL expressions can also be used to combine multiple filtering conditions to implement more complex filtering logic. For example, you can use SpEL expressions to write conditions like type=='fire'&&(time>='00:00:00'&&time<='06:00:00'), which means that only fire anomalies detected in the early morning hours will be notified to the user.
[0066] (5) Dynamic configuration: Since SpEL expressions are parsed dynamically at runtime, it is possible to dynamically configure filter conditions without modifying the source code. This allows system administrators to adjust the filtering strategy for abnormal alarms at any time according to actual needs, improving the flexibility and maintainability of the system.
[0067] In summary, the SpEL expression can be used in the present invention to define various filtering conditions, realize flexible filtering and management of abnormal alarms, thereby improving system efficiency and user experience.
[0068] 3. Related content involved in the formation process of the filter chain:
[0069] (1) Filtering point:
[0070] The filter point is the location where the filter is set. The processing of alarm information needs to be carried out in stages. Since the information contained in each stage is different, filter points can be set at different stages of alarm processing for different information; for example, after the anomaly detection model notifies the abnormal alarm to the server, it will first perform a series of processes such as message conversion, legitimacy verification, delay verification, system internal information matching, abnormal repeatability verification, message label classification, sending abnormal notifications, and alarm classification storage. Based on this, suitable filter points can be set at stages that require frequent changes and flexible configuration. The functions of each filter point can be different or the same. For example, filter points with filter functions and filter points with label classification can classify and grade different alarms. There can be filter points for alarm labels for secondary filtering. Filter points can also form filter point chains layer by layer in order.
[0071] (2) Filter configuration:
[0072] Filter configuration (FilterConfig) plays a key role in abnormal alarm processing. As the name implies, filter configuration is to configure the filter applicability. Configuration items include filter type, spel expression, time configuration, number of times allowed to pass, interruption time, etc. Different filter configurations may be required for different filters. The storage data structure of this part adopts json format. By flexibly configuring various filtering conditions and behaviors, accurate filtering and management of abnormal alarms can be achieved. For example, in a video surveillance system, the filter configuration can be used to set the time window size, and abnormalities are only notified within a specific time period. At the same time, further filtering is performed according to conditions such as abnormality type and duration to ensure that users only receive key alarm information, reduce unnecessary interference, and improve the accuracy and practicality of the system.
[0073] (3) Filter:
[0074] The filter is a key component in the present invention. The final filtering effect is the result of the combined action of multiple filters. From the perspective of classification, it can be divided into filters and label filters. The filter mainly plays a filtering role and can filter out most of the duplicate data. The label filter can be classified according to its importance after filtering out most of the data. It plays a grading role. Generally speaking, the filter is responsible for the multi-dimensional screening and processing of abnormal alarms. It includes various parts such as filter configuration, input format, output results, etc., and can flexibly execute specific filter algorithms according to the configuration. Through precise configuration, the filter can carefully process the alarm before the abnormal detection model service passes the abnormal alarm to the subsequent processing flow to improve accuracy, reduce user interference, and enhance user attention. Through the filter, the system can effectively filter, optimize and streamline the alarm information to ensure that the user receives the most concerned and accurate abnormal events, thereby improving the overall system performance and user experience.
[0075] (4)Filter target:
[0076] FilterTarget is a key concept in the filter system. It refers to the specific alarm data object targeted by the filter during processing. This data is usually expressed in JSON format and contains alarm type, alarm time, accuracy, and other complex information that may be involved. The setting of filter target provides the filter with specific basis and goals for judging and processing alarm information;
[0077] The functions of filter targets include: 1) Providing judgment basis: The detailed information contained in the filter target provides the filter with the basis for judging whether the message needs to be passed or filtered, ensuring that the filter can effectively perform tasks according to the preset rules. 2) Supporting complex decisions: The complex information in the filter target (such as accuracy) can help the filter make more complex decisions, such as deciding whether the alarm process should be triggered based on the credibility of the alarm. 3) Deduplication of messages: By analyzing the alarm content in the filter target, the filter can identify and avoid duplicate alarm information and reduce unnecessary interference. 4) Time dimension screening: The filter can match the alarm time with the preset time dimension requirements, and only pass the alarms that meet the time conditions, thereby improving the relevance and timeliness of the alarm. 5) Data normalization: The filter target allows the filter to normalize multiple alarm data to ensure the consistency and comparability of the data, which is convenient for subsequent analysis and processing. 6) Optimizing the alarm process: The clear setting of the filter target helps to optimize the entire alarm process, making the alarm system more efficient and accurate, and improving the overall alarm quality. 7) Enhance system adaptability: The flexibility of filter targets enables the system to adapt to a variety of different alarm scenarios and requirements. By adjusting the configuration of filter targets, it can quickly adapt to new alarm types or changed business requirements.
[0078] (5) Filter processing results:
[0079] The filter processing result (FilterResult) is the output generated by the filter system after processing the filter target. It records the analysis results and decisions of the filter on the alarm data. This result not only includes the judgment of whether the alarm triggers the filter, but also includes the label information related to the alarm. This information provides key basis for the subsequent alarm processing process.
[0080] The functions of the filtering results include: 1) Alarm status record: The filtering results record whether the alarm data meets the triggering conditions of the filter. This status information is crucial for judging the severity and urgency of the alarm. 2) Filter trigger identification: By recording which filters are triggered, the filtering results provide a basis for classifying and prioritizing alarm data, which helps the system to effectively manage alarms. 3) Alarm tag information: The alarm tag information in the filtering results contains key attributes such as the type and level of the alarm. This information is crucial for deciding whether to notify the user and how to respond to the alarm. 4) User notification decision: Based on the tag level in the filtering results, the system can decide whether to send a notification to the user, as well as the content and method of the notification, to ensure that the user only receives key and relevant alarm information. 5) Alarm data analysis: The filtering results provide a basis for further analysis of the alarm data, so that data analysis and mining can be more accurately targeted at actual problems. 6) System performance optimization: By analyzing the filtering results, system administrators can understand the performance and efficiency of the filters, and then adjust and optimize the filters to improve the overall performance of the system.
[0081] (6)Filter chain:
[0082] Filter Chain is a processing mechanism composed of multiple filter components connected in series according to a predetermined order. Each filter processes the data passing through according to its configured rules until the data meets the final filtering criteria or reaches the predetermined processing goal.
[0083] The functions of the filter chain include: 1) Multi-dimensional screening: The filter chain can screen abnormal alarms from different angles through the collaborative work of multiple filters, ensuring that only alarm information that meets all set conditions will be passed to the next processing link. 2) Fine-grained management: Each filter can be configured for different abnormal types, durations, severity and other factors, so that the filter chain can fine-tune the management of alarm information and improve the pertinence and effectiveness of processing. 3) Improve accuracy: Through the step-by-step screening of the filter chain, false alarms and duplicate alarms can be effectively reduced, thereby improving the accuracy of alarm information and reducing unnecessary interference to users. 4) Optimize performance: The filter chain pre-processes the alarm information before it is passed to the end user, reducing the burden of subsequent processing processes and optimizing the overall performance of the system. 5) Enhance user experience: The filter chain ensures that users receive carefully screened and classified alarm information, which is more critical and accurate, thereby improving the user experience and satisfaction. 6) Flexibility and scalability: The design of the filter chain allows flexible addition, removal or reordering of filter components according to actual needs, so that the entire system can adapt to changing business needs and technical environments.
[0084] In summary, the embodiments of the present invention are applicable to various scenarios that require visual anomaly detection, can improve the security, efficiency and accuracy of the system, reduce interference to users, and enhance users' attention to abnormal situations, and have broad application prospects and market potential.
[0085] Embodiment 2:
[0086] On the basis of the above-mentioned embodiment 1, it also includes: judging the alarm label output by the filter chain, and when the alarm label meets the preset alarm level, sending the second alarm information corresponding to the alarm label to the user end; and according to the second alarm information, sending the corresponding response measures to the user end.
[0087] Embodiment three:
[0088] On the basis of the above-mentioned embodiment 2, it further includes: evaluating the filtering effect of the filter chain according to the second alarm information and the corresponding response result, and optimizing the filter chain according to the evaluation result.
[0089] Next, the above three embodiments are specifically described through a specific example:
[0090] (1) System deployment and initialization:
[0091] Install and configure surveillance cameras according to the target usage scenario to ensure that the target usage area can be covered;
[0092] Deploy algorithm model services and filter chain systems, including window filters, alarm type filters, and sliding window filters;
[0093] According to the characteristics and security requirements of the monitoring site, the configuration of the filter chain is initialized, such as setting the first preset time, the risk factor threshold and the second preset time.
[0094] (2) Video stream capture and processing:
[0095] The camera captures the video stream of the monitored place in real time and transmits it to the algorithm model service;
[0096] Perform necessary preprocessing on the video stream, such as compression and denoising, to improve the efficiency and accuracy of subsequent processing.
[0097] (3) Anomaly detection and alarm generation:
[0098] The algorithm model service analyzes each frame of video image to detect whether there are any abnormal situations, such as stranded personnel, vehicle obstruction, intrusion, etc.
[0099] When an anomaly is detected, a preliminary alarm message is generated and relevant metadata such as timestamp, location information, etc. are attached.
[0100] (4) Alarm information filtering and processing:
[0101] The generated alarm information is input into the filter chain for multi-dimensional and multi-scenario intelligent filtering processing; specifically, the time window filter is responsible for controlling the frequency and time interval of alarms; the alarm type filter classifies and prioritizes alarm types to ensure that critical alarms are given priority; the sliding window filter merges multiple identical alarms to reduce the occurrence of continuous repeated alarms.
[0102] (5) Alarm decision and user notification:
[0103] Based on the filtering results output by the filter chain, the system decides whether it is necessary to notify the monitoring personnel or automatically take corresponding measures; if notification is required, the system will send an alarm message to the preset recipient through an appropriate method (such as the alarm system of the monitoring center, mobile application push, etc.), and can also send a response strategy for the alarm message.
[0104] (6) Response and Disposal:
[0105] After receiving the alarm information and response strategy, the preset recipient will take effective response measures, such as on-site investigation, dispatching security personnel, etc., based on the severity and type of the alarm and referring to the sent response strategy; if the system supports automatic disposal, such as integration with traffic control systems, it can automatically adjust traffic lights or guide vehicle diversion.
[0106] (7) System optimization and maintenance:
[0107] Regularly analyze filtered alarm information and response results, evaluate the effectiveness of the filter chain, and optimize the configuration of the filter chain; update algorithm models and filter rules according to changes in monitoring locations and new security requirements to ensure the effectiveness and adaptability of the system.
[0108] Through these specific implementation steps, the present invention can effectively reduce the problem of frequent alarms in the video surveillance system, improve the accuracy and relevance of the alarms, ensure that monitoring personnel can focus on handling truly important events, and thus improve the security and efficiency of the entire monitoring system.
[0109] Embodiment 4:
[0110] See also Figure 2 As shown, an embodiment of the present invention discloses an alarm filtering system based on a visual anomaly detection algorithm, and applies the above-mentioned alarm filtering method based on a visual anomaly detection algorithm. The system includes a configuration module and an alarm filtering module;
[0111] Configuration module, used to configure filter type, Spel expression, and alarm type and alarm condition in Spel expression; form filter points based on Spel expression; select corresponding filters according to the configured filter type and form filter chain according to the obtained filter points; each filter is responsible for filtering the corresponding time dimension information, alarm type dimension information or sliding window dimension information;
[0112] The alarm filtering module is used to input the first alarm information currently detected based on the visual anomaly detection algorithm to the starting end of the configured filter chain, and output the second alarm information after filtering and the corresponding alarm label.
[0113] It also includes an alarm module; the alarm module is used to send a second alarm message to the user end when the alarm tag meets the preset alarm level; and is used to send corresponding response measures to the user end according to the second alarm message.
[0114] It also includes an optimization module; the optimization module is used to evaluate the filtering effect of the filter chain according to the second alarm information and the corresponding response result, and optimize the filter chain according to the evaluation result.
[0115] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0116] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. An alarm filtering method based on a visual anomaly detection algorithm, characterized in that: The steps include: Detect video data in the target usage scenario based on a visual anomaly detection algorithm; input the currently detected first alarm information to the starting end of the configured filter chain, and output the filtered second alarm information and the corresponding alarm label; The filter chain configuration steps include: (1) Configuring the filter type, SPELS expression, and the alarm type and alarm condition in the SPELS expression; forming a filter point based on the SPELS expression; (2) According to the configured filter type, the corresponding filter is selected to form a filter chain based on the obtained filter points; each filter is responsible for filtering the corresponding time dimension information, alarm type dimension information or sliding window dimension information.
2. The alarm filtering method based on the visual anomaly detection algorithm according to claim 1 is characterized in that: Filter the corresponding time window dimension information, including: The number of alarm messages generated within a preset time period under the target usage scenario is set to n, the number of times the same type of alarm messages are allowed to be sent within the preset time period is set to i, and the ni-th alarm message of the same type generated within the preset time period is filtered.
3. The alarm filtering method based on the visual anomaly detection algorithm according to claim 1 is characterized in that: Filter the corresponding alarm type dimension information, including: Filter out the alarm information corresponding to the alarm type with a risk factor less than the threshold; Alternatively, according to the risk factors of different types of alarms, the alarm type information with a high risk factor is sent to the client first; Alternatively, on the basis of filtering out the alarm information corresponding to the alarm type with a risk factor less than a threshold, the alarm type information with a high risk factor is preferentially sent to the client.
4. The alarm filtering method based on the visual anomaly detection algorithm according to claim 1 is characterized in that: Filter the corresponding sliding window dimension information, including: The same alarm information generated within the second preset time period is merged.
5. The alarm filtering method based on the visual anomaly detection algorithm according to claim 1 is characterized in that: Also includes: When the alarm tag meets the preset alarm level, the second alarm information corresponding to the alarm tag is sent to the user terminal.
6. The alarm filtering method based on the visual anomaly detection algorithm according to claim 3 is characterized in that: Also includes: According to the second alarm information, corresponding response measures are sent to the user end.
7. The alarm filtering method based on a visual anomaly detection algorithm according to claim 4 is characterized in that: Also includes: The filtering effect of the filter chain is evaluated according to the second alarm information and the corresponding response result, and the filter chain is optimized according to the evaluation result.
8. An alarm filtering system based on a visual anomaly detection algorithm, characterized in that: Applying the method described in any one of claims 1 to 7, the system comprises a configuration module and an alarm filtering module; The configuration module is used to configure the filter type, SPEIL expression, and the alarm type and alarm condition in the SPEIL expression; Forming a filter point based on the Spel expression; selecting a corresponding filter according to the configured filter type and forming a filter chain based on the obtained filter point; Each filter is responsible for filtering the corresponding time dimension information, alarm type dimension information or sliding window dimension information; The alarm filtering module is used to input the first alarm information currently detected based on the visual anomaly detection algorithm into the starting end of the configured filter chain, and output the second alarm information after filtering and the corresponding alarm label.
9. The alarm filtering system based on the visual anomaly detection algorithm according to claim 8, characterized in that: It also includes an alarm module; The alarm module is configured to send the second alarm information to the user terminal when the alarm tag meets a preset alarm level; and to send corresponding response measures to the user terminal according to the second alarm information.
10. The alarm filtering system based on visual anomaly detection algorithm according to claim 8, characterized in that: Also includes optimization modules; The optimization module is configured to evaluate the filtering effect of the filter chain according to the second alarm information and the corresponding response result, and optimize the filter chain according to the evaluation result.
