Safety monitoring method and device based on artificial intelligence, monitoring equipment and medium

Abstract

The invention relates to the technical field of safety monitoring, and provides a safety monitoring method and device based on artificial intelligence, monitoring equipment and a medium. According to the method and the device, abnormal behavior detection is performed on the monitoring video containing the monitoring object through the AI model, so that the video can be analyzed without interruption and omission for 24 hours, and the possibility of human negligence and misjudgment is reduced. When the abnormal behavior is detected, the video clip related to the abnormal behavior is extracted from the monitoring video and sent to the monitoring manager for manual verification, so that the amount of video needing to be checked by the monitoring manager is reduced, the working efficiency is improved, the monitoring manager is ensured to directly pay attention to the specific situation of the abnormal behavior, and the user experience is improved. And corresponding decisions can be made quickly. And when the verification feedback mark of the monitoring manager is received and the verification feedback mark is matched with the preset alarm triggering mark, the alarm system is triggered, so that the conditions of false alarm and missing alarm are avoided, and the effectiveness and credibility of the alarm system are improved.

Description

Technical Field

[0001] The present application relates to the field of security monitoring technology, and in particular to an artificial intelligence-based security monitoring method, device, monitoring equipment and medium. Background Art

[0002] As the care needs of the elderly and people with cognitive impairment are growing, timely detection and response to key events such as falls and wandering are crucial to ensure the safety of the elderly and people with cognitive impairment. Therefore, real-time monitoring and early warning of the behavior of the elderly and people with cognitive impairment are required to detect and deal with dangerous situations such as falls in a timely manner.

[0003] However, traditional monitoring methods mainly rely on manual observation and intervention, which has problems such as insufficient human resources and inability to achieve 24-hour uninterrupted monitoring. Summary of the invention

[0004] Based on this, it is necessary to propose a security monitoring method, device, monitoring equipment and medium based on artificial intelligence to address the technical problem that existing nursing monitoring cannot achieve 24-hour uninterrupted monitoring.

[0005] In a first aspect, a security monitoring method based on artificial intelligence is provided, the method comprising:

[0006] Detect abnormal behaviors in surveillance videos containing surveillance objects based on AI models;

[0007] When abnormal behavior is detected, obtaining a video clip related to the abnormal behavior from the surveillance video;

[0008] Sending the video clip to a monitoring manager for manual verification, and receiving a verification feedback mark from the monitoring manager;

[0009] When the verification feedback mark matches the preset alarm trigger mark, the alarm system is triggered.

[0010] In a second aspect, a security monitoring device based on artificial intelligence is provided, the device comprising:

[0011] A detection module is used to detect abnormal behaviors of surveillance videos containing surveillance objects based on AI models;

[0012] An acquisition module, used for acquiring video clips related to the abnormal behavior from the monitoring video when abnormal behavior is detected;

[0013] A feedback module, used to send the video clip to a monitoring manager for manual verification and receive a verification feedback mark from the monitoring manager;

[0014] The alarm module is used to trigger the alarm system when the verification feedback mark matches the preset alarm trigger mark.

[0015] In a third aspect, a monitoring device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned artificial intelligence-based security monitoring method when executing the computer program.

[0016] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned artificial intelligence-based security monitoring method are implemented.

[0017] This application first uses an AI model to detect abnormal behavior in surveillance videos. Compared with traditional manual monitoring, the AI ​​model can analyze videos continuously and without omission, quickly respond to and identify potential security threats, and reduce the possibility of human negligence and misjudgment. After detecting abnormal behavior, it can automatically and accurately extract video clips related to abnormal behavior from the surveillance video, which not only reduces the amount of video that the monitoring manager needs to view and improves work efficiency, but also ensures that the monitoring manager can directly focus on key information, that is, the specific context in which the abnormal behavior occurs, which helps the monitoring manager to understand the problem faster and make corresponding decisions. The extracted video clips are sent to the monitoring manager for manual verification, and the verification feedback mark of the manager is received. The introduction of manual verification links can further improve the accuracy and reliability of the system. The professional knowledge and experience of the monitoring manager can make up for the shortcomings of the AI ​​model in some complex or ambiguous scenarios and ensure the accuracy of the alarm triggering. At the same time, the manual verification feedback mechanism also helps to continuously optimize the performance of the AI ​​model and form a virtuous circle. When the verification feedback mark of the monitoring manager matches the preset alarm trigger mark, the system triggers the alarm system, ensuring that the alarm system will only be triggered when it is confirmed that there is real abnormal behavior, avoiding false alarms and missed alarms, and improving the effectiveness and credibility of the alarm system. At the same time, timely alarm notifications help to quickly respond to and handle security incidents and reduce potential losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application 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 some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] in:

[0020] Figure 1 An application environment diagram of a security monitoring method based on artificial intelligence in one embodiment;

[0021] Figure 2 A flowchart of a security monitoring method based on artificial intelligence in one embodiment;

[0022] Figure 3 is a structural block diagram of a security monitoring device based on artificial intelligence in one embodiment;

[0023] Figure 4 Artificial intelligence-based security monitoring is a structural block diagram of a monitoring device in one embodiment. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0025] The artificial intelligence-based security monitoring method provided in the embodiments of the present application can be applied in Figure 1 In the application environment of, the monitoring device communicates with the monitoring camera, the remote server and the nursing end device through the network. The monitoring camera is installed in the nursing environment or nursing facility, and is used to capture the monitoring video of the monitored object in real time, and transmit the monitoring video to the monitoring device in real time. The monitored object refers to the elderly and cognitively impaired people who need to be monitored and cared for. An AI model is deployed in the monitoring device, and abnormal behavior detection is performed on the monitoring video in real time based on the AI ​​model. When abnormal behavior is detected, the video clip related to the abnormal behavior is obtained from the monitoring video, and the video clip is transmitted to the remote server. The monitoring manager views the video clip through the display interface provided by the remote server to verify the abnormal behavior, and makes a verification feedback mark through the display interface. The monitoring device receives the verification feedback mark, and when the verification feedback mark made by the monitoring manager matches the preset alarm trigger mark, the alarm system is triggered to send a security alarm to the nursing end device, so that the nursing staff corresponding to the nursing end device can take timely measures.

[0026] The remote server may be implemented as an independent server or a server cluster consisting of multiple servers. The nursing end device may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, virtual helmet displays, smart glasses displays, and portable wearable devices. The present application is described in detail below through specific embodiments.

[0027] See also Figure 2As shown, Figure 2 A flowchart of a security monitoring method based on artificial intelligence provided in an embodiment of the present application includes the following steps:

[0028] S21, performing abnormal behavior detection on surveillance videos containing surveillance objects based on the AI ​​model.

[0029] Surveillance video of a monitored object refers to a continuous video data stream captured by surveillance cameras installed in a care environment or care facility, which is used to monitor and record a specific area in real time.

[0030] The AI ​​model is a machine learning model pre-trained based on artificial intelligence technology, which is used to detect abnormal behavior of monitored objects in surveillance videos. The AI ​​model in this application is an abnormal behavior detection model pre-trained based on artificial intelligence technology. In surveillance videos, abnormal behavior refers to activities that are significantly different from normal behavior patterns, which may indicate potential security threats, emergencies, or abnormal situations that require attention. For example, falling, wandering, asking for help, fighting, illegal intrusion, violent behavior, etc.

[0031] The process of training an AI model can be as follows: Collect surveillance videos of different surveillance objects. These surveillance videos come from various surveillance cameras and cover different scenes and time periods. Preprocess the collected surveillance videos, including video decoding, frame extraction, image enhancement, noise removal and other steps to improve the efficiency and accuracy of subsequent processing. Extract key features from video frames, such as human body contours, motion trajectories, behavior patterns, etc. Use a surveillance video set containing normal and abnormal behavior labels to train an AI model based on deep learning models such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). Apply the trained AI model to new surveillance videos, and perform real-time analysis on each frame or each video segment to detect whether there is abnormal behavior. If abnormal behavior is detected, the AI ​​model will output the corresponding abnormal label and detection confidence.

[0032] Traditional surveillance video requires manual observation of the screen for a long time, which is easy to cause fatigue and miss detection. Automated detection based on AI models can greatly reduce the manual burden and improve monitoring efficiency.

[0033] It should be understood that in a complex security monitoring environment, multiple surveillance cameras can be deployed to cover different angles and areas. By automatically integrating and analyzing video data from multiple surveillance cameras, monitoring events can be reviewed and verified from all angles and angles. Specifically, when a surveillance camera captures an abnormal event, the system can quickly call up the video footage of other related surveillance cameras for time synchronization and spatial matching to obtain more comprehensive and accurate event information. Through this multi-angle cross-verification, the system can significantly reduce the false alarm rate caused by the limited viewing angle of a single surveillance camera or interference from environmental factors, while enhancing the ability to identify hidden events or abnormal behaviors in complex scenarios and reducing the occurrence of missed reports.

[0034] The multi-angle event verification mechanism not only improves the accuracy and reliability of security monitoring, but also provides richer and more detailed evidence support for subsequent emergency response and incident investigation.

[0035] S22, when abnormal behavior is detected, obtaining a video clip related to the abnormal behavior from the monitoring video.

[0036] A video clip is a continuous video content extracted from a complete surveillance video, which includes the entire process of abnormal behavior.

[0037] When the AI ​​model detects activities in the surveillance video that may match the preset abnormal behavior pattern, the AI ​​model generates event candidates based on the detected abnormal behavior. The event candidate is associated with a specific video clip that contains the video content in the time period that the AI ​​model believes may be the abnormal behavior. At the same time, the AI ​​model gives each event candidate a detection confidence, which indicates the degree of certainty about this event candidate.

[0038] S23, sending the video clip to a monitoring manager for manual verification, and receiving a verification feedback mark from the monitoring manager.

[0039] When the AI ​​model detects abnormal behavior and generates relevant video clips, the video clips will be automatically sent to the monitoring manager's terminal device or monitoring platform or according to preset conditions (such as the detection confidence reaches a certain threshold).

[0040] Because the sent video clips include sufficient contextual information, monitoring managers can understand the circumstances before and after the abnormal behavior occurs. For example, the video content may include a few seconds to a few minutes before and after the abnormal behavior occurs.

[0041] After receiving the video clips, the monitoring managers will carefully watch and evaluate the content and determine whether the abnormal behavior detected by the AI ​​model falls into the category of abnormalities based on their own expertise and experience.

[0042] After completing manual verification, the monitoring manager will mark the video clips or abnormal behaviors in the video clips according to the verification results. The markings include classifications such as "normal" and "abnormal", as well as detailed descriptions or annotations.

[0043] Verification feedback marks can be recorded in the local database for subsequent analysis and statistics. At the same time, these marks can also be used as one of the data sources for AI model training and optimization, helping to improve the detection accuracy and robustness of AI models.

[0044] In the abnormal behavior detection process, the detected video clips related to the abnormal behavior are sent to the monitoring manager for manual verification, and their verification feedback marks are received to ensure that the abnormal behavior detected by the AI ​​model is manually reviewed, thereby improving the accuracy and reliability of detection, helping to reduce the occurrence of false alarms and missed alarms, and providing data support for the continuous optimization of the AI ​​model.

[0045] In an optional implementation, before sending the video clip to a monitoring manager for manual verification, the method further includes:

[0046] Obtaining a detection confidence level corresponding to the abnormal behavior;

[0047] Classifying the abnormal behavior based on the video clip to obtain a classification type;

[0048] extracting a plurality of behavioral features from the video clip;

[0049] Correcting the detection confidence based on the multiple behavior features to obtain a target confidence;

[0050] Comparing the target confidence with a confidence threshold range corresponding to the classification type;

[0051] When the target confidence is within the preset confidence threshold range, the video clip is sent to a monitoring manager for manual verification.

[0052] When the AI ​​model detects abnormal behavior, it generates a detection confidence corresponding to the abnormal behavior. The detection confidence is extracted from the output of the AI ​​model.

[0053] The abnormal behavior is classified based on the video clip to obtain a classification type. The classification type may be a result of classifying the abnormal behavior according to a preset category, such as "falling", "intrusion" and the like.

[0054] Extract key frames from the video clips. Key frames usually contain the information that best represents the video content. Key frames can be selected by frame difference analysis, motion detection and other methods. Then extract multiple behavioral features from the key frames. Behavioral features refer to data points that can describe abnormal behavioral characteristics, such as human posture, motion trajectory, speed change, environmental context, etc. For example, the motion pattern of the target can be analyzed by optical flow method, trajectory tracking and other technologies, and motion features such as speed, acceleration, and direction of movement can be extracted. It is possible to combine time and space information to extract high-level features such as the spatiotemporal trajectory and behavior pattern of the monitored object.

[0055] The detection confidence is corrected based on the multiple behavioral features to obtain the target confidence. The target confidence can more accurately reflect the confidence of the authenticity of the abnormal behavior. Weights are assigned in advance to each behavioral feature and the initial detection confidence. The weight assignment can be adjusted according to the actual situation. For example, if a certain feature is particularly important in judging abnormal behavior, a higher weight can be assigned. Each behavioral feature is quantified into a specific numerical value. For example, the human body posture can obtain the joint angle through the posture estimation algorithm, and then calculate the deviation from the normal posture. The motion trajectory can be obtained through the tracking algorithm, and its complexity or abnormality can be calculated. The speed change can be quantified by calculating the statistical characteristics of the speed sequence (such as the standard deviation). The environmental context can be quantified by analyzing the light intensity and object position in the image. Use the weighted summation method to combine the quantified behavioral features and the corresponding weights to correct the initial confidence.

[0056] The confidence threshold range is a predefined confidence standard used to determine whether manual verification is required. Different classification types correspond to different confidence threshold ranges. The target confidence is compared with the confidence threshold range corresponding to the classification type. When the target confidence is within the confidence threshold range corresponding to the classification type, the video clip is sent to the monitoring manager for manual verification.

[0057] Classification is to better understand the nature of abnormal behavior, so that relevant behavioral features can be extracted more accurately and the detection confidence can be modified in a targeted manner. Different abnormal behaviors may require attention to different features, so classification helps improve the efficiency and accuracy of subsequent processing.

[0058] The purpose of correcting the confidence is to improve the accuracy of the AI ​​model's judgment of abnormal behavior. The initial detection confidence may be affected by many factors, such as environmental noise, obstructions, etc. By extracting multiple behavioral features from the video clip and correcting the confidence based on these multiple behavioral features, the authenticity of the abnormal behavior can be more comprehensively evaluated, thereby obtaining a more reliable target confidence.

[0059] When the target confidence is less than the lower limit of the confidence threshold range, there is no need to send the abnormal behavior video clip to the monitoring manager for manual verification. The target confidence is less than the lower limit of the confidence threshold range, indicating that the degree of confidence in the authenticity of the detected abnormal behavior is very low, far below the preset acceptable level. In this case, sending the video clip to the monitoring manager for manual verification may not be efficient, because it is likely that the video clip does not contain real abnormal behavior, or even if it does, its importance is relatively low. Therefore, in order to avoid wasting the time and resources of the monitoring manager, such low-confidence video clips are not sent for manual review.

[0060] When the target confidence is greater than the upper limit of the confidence threshold range, there is no need to send the abnormal behavior video clip to the monitoring manager for manual verification, and the alarm system is directly triggered to alarm. The target confidence is greater than the upper limit of the confidence threshold range, indicating that the degree of confidence in the authenticity of the detected abnormal behavior is very high, far exceeding the preset suspicious threshold. Since the target confidence has far exceeded the threshold that requires manual verification, the system can directly consider this as an urgent and important abnormal behavior without manual review. At this time, the alarm system is directly triggered so that timely response measures can be taken, such as sending nursing staff to the scene for processing.

[0061] Only when the target confidence is greater than the lower limit of the confidence threshold range and less than the upper limit of the confidence threshold range, that is, the target confidence is within the confidence threshold range, does it need to send the abnormal behavior video clip to the monitoring manager for manual verification. When the target confidence is between these two thresholds, it indicates that the judgment of abnormal behavior is in a relatively vague but still meaningful range. The confidence in this range is neither too low (so that it can be ignored) nor too high (so that it can be directly confirmed), but requires further verification to ensure accuracy. In this case, it is necessary to send the video clip to the monitoring manager for manual verification. Because although the AI ​​model has a certain degree of confidence in abnormal behavior, it has not reached the level that can be directly confirmed or ignored. Through manual review, it is possible to more accurately determine whether abnormal behavior really exists and whether further action is needed. Doing so can not only avoid the waste of resources caused by false alarms, but also ensure that important abnormal behaviors are responded to in a timely manner. .

[0062] The above optional implementation provides a more refined process to decide whether to send the detected abnormal behavior video clips to the monitoring manager for manual verification. By comprehensively considering multiple factors such as detection confidence, behavior classification, behavior feature extraction, and confidence correction, the efficiency and accuracy of verification can be improved.

[0063] In an optional implementation, sending the video clip to a monitoring manager for manual verification includes:

[0064] Determine the target monitoring manager identification according to the classification type;

[0065] The video clip is sent to a target monitoring manager corresponding to the target monitoring manager identification for manual verification.

[0066] Based on the previous classification results of the abnormal behavior video clips (i.e., the classification type), the identification of the target monitoring manager responsible for handling the type of abnormal behavior is found and determined. A mapping or allocation mechanism can be set up to associate different types of abnormal behaviors with specific monitoring managers or teams. In this way, the system can ensure that each video clip is sent to the monitoring manager who is most suitable for handling its type of abnormality, thereby improving processing efficiency and accuracy.

[0067] After the target monitoring manager ID is determined, the video clip of the abnormal behavior will be sent to the target monitoring manager corresponding to the ID. For example, this can be done through channels such as email, instant messaging tools, or a dedicated monitoring management platform. After receiving the video clip, the monitoring manager will conduct a manual review to confirm whether the abnormal behavior actually exists and whether further action is required. In this way, the monitoring manager's expertise and experience can be fully utilized to review and verify the detection results of the AI ​​system, thereby improving the reliability and accuracy of the entire monitoring system.

[0068] The above optional implementation method, which determines the target monitoring manager based on the type of abnormal behavior, helps to achieve the optimal configuration and efficient use of monitoring resources. Different types of abnormal behaviors may require different processing methods and expertise, so assigning them to the most suitable monitoring manager for processing can ensure that the problem is solved in a timely and effective manner. At the same time, this can also improve the work efficiency of monitoring managers, because they can focus on handling the abnormal types they are good at and familiar with, without wasting time and energy on unfamiliar areas.

[0069] In an optional implementation, sending the video clip to a target monitoring manager corresponding to the target monitoring manager identifier for manual verification includes:

[0070] The video clip is encrypted and sent to a target monitoring manager corresponding to the target monitoring manager identifier for manual verification.

[0071] In order to enhance the security of video clips during transmission, protect the privacy of the monitored objects in the video clips, and ensure that only authorized target monitoring managers can access and view these sensitive information, the video clips will be encrypted before being sent to the target monitoring managers for manual verification.

[0072] During the encryption process, one or more encryption algorithms can be used to encrypt the video clips to generate encrypted video files. The encryption algorithm can be symmetric encryption (such as AES) or asymmetric encryption (such as RSA). During the encryption process, the system will also generate one or more keys for subsequent decryption operations. Ensure that only authorized target monitoring managers can obtain the keys.

[0073] After the encryption process is completed, the encrypted video clips are sent to the target monitoring manager corresponding to the target monitoring manager ID through a secure communication channel (such as encrypted email, secure file transfer protocol, etc.). During the transmission process, even if the data is intercepted, it cannot be directly viewed by unauthorized personnel because the video content has been encrypted and protected.

[0074] After receiving the encrypted video clips, the target monitoring manager uses the corresponding decryption key and algorithm to decrypt the video file to restore the original video content. After decryption is completed, the monitoring manager can manually review and verify the video clips. This method of encrypted transmission and decryption verification not only protects the confidentiality of the video clips, but also ensures that only legitimate recipients can access the data.

[0075] S24, triggering an alarm system when the verification feedback mark matches a preset alarm trigger mark.

[0076] After the video clip is manually verified by the target monitoring manager, the manager will generate a verification feedback tag for the video clip based on factors such as whether the abnormal behavior in the video content actually exists and whether it needs urgent processing. This tag is usually a code, identifier or status information used to indicate the verification result of the video clip and subsequent processing suggestions.

[0077] When the verification feedback tag is generated, it is matched with the preset alarm trigger tag. The matching process can be to compare the codes, identifiers or status information of the two tags to determine whether they are consistent or conform to a certain specific logical relationship (such as inclusion relationship, equivalence relationship, etc.).

[0078] If the verification feedback tag matches the preset alarm trigger tag, the alarm system is triggered immediately. The alarm system may notify relevant personnel in a variety of ways, including sending emails, SMS notifications, pop-up alarm windows, and starting sound or light signals. The content of the alarm usually includes key information such as the summary of the video clip that triggered the alarm, the type of abnormal behavior, the time and place of occurrence, so that relevant personnel can quickly understand the situation and respond.

[0079] After the alarm is triggered, a series of subsequent processing processes can be initiated, such as recording alarm logs, updating monitoring status, automatically dispatching emergency response teams, etc. These processing processes are designed to ensure that relevant personnel can respond to emergencies in a timely and effective manner and reduce potential risks and losses.

[0080] In an optional embodiment, the method further comprises:

[0081] The AI ​​model is updated based on the verification feedback mark.

[0082] Collect verification feedback marks provided by the target monitoring manager. The verification feedback marks represent whether the abnormal behavior in the video clip actually exists and whether the model's judgment is correct.

[0083] The collected verification feedback marks are analyzed to determine in which cases the model made the correct judgment and in which cases there were false positives or negatives. Based on the analysis results, the parameters of the AI ​​model are adjusted, for example, feature weights are modified, classification thresholds are adjusted, learning algorithms are updated, etc., to improve the model's recognition ability in similar scenarios. That is, the verification feedback marks of monitoring managers are used to further train and improve the AI ​​model, thus forming an automated feedback loop to continuously improve the performance of the model.

[0084] In other embodiments, new training data (including correctly labeled abnormal behavior samples and normal behavior samples) can also be used to retrain the AI ​​model. Retraining can help the AI ​​model learn more about the characteristics of abnormal behavior and reduce the occurrence of false positives and false negatives.

[0085] After updating the AI ​​model, perform a performance evaluation on the AI ​​model to ensure that the updated AI model is more accurate and efficient in identifying abnormal behaviors than before the update.

[0086] By updating the AI ​​model based on validation feedback tags, the AI ​​model can continuously learn and improve to adapt to the changing monitoring environment and abnormal behavior patterns. This self-learning and optimization capability is critical to improving the overall performance and reliability of the monitoring system.

[0087] In addition, updating AI models can also help reduce the occurrence of false positives and missed negatives, and improve the accuracy and credibility of the monitoring system. When AI models can more accurately identify abnormal behavior, monitoring managers can focus more on handling real emergencies instead of being distracted by a large number of false positives.

[0088] In an optional embodiment, the method further comprises:

[0089] A monitoring interface is displayed, wherein the monitoring interface is used to display the video clip and provide the monitoring manager with a function of performing verification feedback marking.

[0090] The monitoring interface can load and play video clips and provide control functions such as play, pause, stop, fast forward, fast rewind, etc., so that monitoring managers can view video content as needed.

[0091] The monitoring interface also provides a set of verification feedback tools, allowing monitoring managers to evaluate the content in the video clips and give corresponding verification feedback marks. The verification feedback tools may include buttons (such as "normal", "abnormal"), drop-down menus (for selecting abnormal types), text boxes (for entering detailed descriptions), etc. For example, the specific fields of the verification feedback tool screening strategy for the abnormal action of falling may include: falling, falling, slipping, tripping, sitting down, sliding down, falling down, fainting, sliding off the bed, falling to the ground, falling, sitting down, sitting on the ground, falling down, losing consciousness, transient syncope, sudden dizziness when getting up, slowly sliding to the ground, kneeling on both knees, buttocks, etc.). The specific fields of the verification feedback tool fall location may include: bedside, bathroom, toilet, corridor, sofa, living room, kitchen, out for inspection, etc. The specific fields of vital signs may include: heart rate, blood pressure. The monitoring manager can quickly mark the status of the video clip by clicking or selecting the verification feedback tool.

[0092] The monitoring interface can display the verification feedback marks given by the monitoring manager in real time. The verification feedback marks can be associated with the timestamp or frame number of the video clip, so that the specific video content can be quickly located when needed. At the same time, the monitoring interface can also provide summary information of the verification feedback marks, such as the number and type distribution of abnormal behaviors, to help monitoring managers understand the overall monitoring situation.

[0093] The function of displaying video clips through the monitoring interface and providing verification feedback marking to monitoring managers can significantly improve the user experience and monitoring efficiency of the monitoring system.

[0094] In an optional embodiment, the method further comprises:

[0095] The video clip and the verification feedback mark are stored in association.

[0096] The verification feedback mark is associated with the corresponding video clip, and the associated video clip and verification feedback mark are stored in a database of the system. The database can be a relational database or a non-relational database.

[0097] To facilitate subsequent data analysis, for example, through statistical analysis of a large number of video clips and verification feedback marks, we can find the distribution patterns, change trends and potential risk points of abnormal behaviors and create indexes for the stored data. In this way, when we need to find a specific video clip or verification feedback mark, we can quickly locate the relevant data and improve data processing efficiency.

[0098] Since the stored data may contain sensitive information, you can also set up permission management to allow authorized personnel to access and modify the stored data to ensure the security and integrity of the data.

[0099] Storing video clips and verification feedback tags in association can improve monitoring system performance and support subsequent data analysis.

[0100] The following will use specific examples to illustrate the application and advantages of the artificial intelligence-based security monitoring method provided in the embodiments of the present application in actual scenarios.

[0101] Suppose an AI-based safety monitoring system is installed in a nursing facility to monitor the activities of residents, especially to identify possible falls.

[0102] The system recognizes that this may be a fall event when a resident suddenly sits down and appears to be in pain, as captured by the surveillance camera. The system records the video clip of the event (e.g., 10 seconds of video before and after the fall) and the detection confidence score (e.g., 0.8). Since the detection confidence score is higher than the preset confidence threshold (e.g., 0.7), the system considers this to be an event candidate that requires further review. The event candidate is sent to the remote server and notifies the monitoring manager on duty. After receiving the notification, the monitoring manager accesses the system interface to view the video clip and the detection confidence score, taking into account the time of the event and the health status of the resident. The monitoring manager determines that this is a real fall event and marks it in the system for confirmation. Marked as a correct alarm, the marked alarm will form annotation information. After receiving the confirmation from the monitoring manager, the system immediately sends an alarm to the caregiver, and the resident receives timely medical assistance. It can be seen that the system can not only respond quickly to potential emergencies, but also improve the accuracy and reliability of detection through the participation of monitoring managers. In addition, the system records the confirmation of the monitoring manager as positive feedback for the training and optimization of the AI ​​model to reduce false positives and false negatives.

[0103] Assume that an AI-based security monitoring system is deployed in an elderly care facility. An elderly person accidentally falls when getting up at night. The surveillance camera captures this behavior. The system detects the fall behavior and generates an event candidate. Since the detection confidence is higher than the preset confidence threshold, the system sends a review request to the monitoring manager on duty. The monitoring manager logs into the system and views the video clip. The monitoring manager confirms that this is a real fall event and marks it as confirmed in the system. The system encrypts and stores the video clip of the event, the detection results, and the monitoring manager's confirmation feedback in the database.

[0104] At the same time, the system uses newly collected real alarm data to update the AI ​​model to improve the ability to identify fall events. Through continuous model training and monitoring manager feedback, the system is continuously optimized to reduce false alarms and missed alarms and improve overall performance.

[0105] This application first uses an AI model to detect abnormal behavior in surveillance videos. Compared with traditional manual monitoring, the AI ​​model can analyze videos continuously and without omission, quickly respond to and identify potential security threats, and reduce the possibility of human negligence and misjudgment. After detecting abnormal behavior, it can automatically and accurately extract video clips related to abnormal behavior from the surveillance video, which not only reduces the amount of video that the monitoring manager needs to view and improves work efficiency, but also ensures that the monitoring manager can directly focus on key information, that is, the specific context in which the abnormal behavior occurs, which helps the monitoring manager to understand the problem faster and make corresponding decisions. The extracted video clips are sent to the monitoring manager for manual verification, and the verification feedback mark of the manager is received. The introduction of manual verification links can further improve the accuracy and reliability of the system. The professional knowledge and experience of the monitoring manager can make up for the shortcomings of the AI ​​model in some complex or ambiguous scenarios and ensure the accuracy of the alarm triggering. At the same time, the manual verification feedback mechanism also helps to continuously optimize the performance of the AI ​​model and form a virtuous circle. When the verification feedback mark of the monitoring manager matches the preset alarm trigger mark, the system triggers the alarm system, ensuring that the alarm system will only be triggered when it is confirmed that there is real abnormal behavior, avoiding false alarms and missed alarms, and improving the effectiveness and credibility of the alarm system. At the same time, timely alarm notifications help to quickly respond to and handle security incidents and reduce potential losses.

[0106] This application combines the judgment of artificial intelligence and the feedback mechanism of monitoring managers, and is applied in nursing monitoring scenarios. It has been able to achieve one person combining artificial intelligence to host 500 users, and more users can be hosted in the future. At the same time, real alarms and false alarms are manually confirmed to confirm the correct labels and wrong labels for artificial intelligence training, so as to optimize the training accuracy of artificial intelligence.

[0107] See also Figure 3As shown, in one embodiment, a security monitoring device 30 based on artificial intelligence is provided, comprising:

[0108] The detection module 301 is used to detect abnormal behavior of the surveillance video containing the monitored object based on the AI ​​model.

[0109] The acquisition module 302 is used to acquire a video clip related to the abnormal behavior from the monitoring video when the abnormal behavior is detected.

[0110] The correction module 303 is used to obtain the detection confidence corresponding to the abnormal behavior; classify the abnormal behavior based on the video clip to obtain a classification type; extract multiple behavior features from the video clip; correct the detection confidence based on the multiple behavior features to obtain a target confidence; and compare the target confidence with a confidence threshold range corresponding to the classification type.

[0111] The feedback module 304 is configured to send the video clip to a monitoring manager for manual verification when the target confidence is within the preset confidence threshold range, and receive a verification feedback mark from the monitoring manager.

[0112] The alarm module 305 is used to trigger an alarm system when the verification feedback mark matches a preset alarm trigger mark.

[0113] The updating module 306 is used to update the AI ​​model based on the verification feedback mark.

[0114] The display module 307 is used to display a monitoring interface, wherein the monitoring interface is used to display the video clips and provide the monitoring manager with a function of performing verification feedback marking.

[0115] The associating module 308 is used to associate and store the video clip and the verification feedback mark.

[0116] This application first uses an AI model to detect abnormal behavior in surveillance videos. Compared with traditional manual monitoring, the AI ​​model can analyze videos continuously and without omission, quickly respond to and identify potential security threats, and reduce the possibility of human negligence and misjudgment. After detecting abnormal behavior, it can automatically and accurately extract video clips related to abnormal behavior from the surveillance video, which not only reduces the amount of video that the monitoring manager needs to view and improves work efficiency, but also ensures that the monitoring manager can directly focus on key information, that is, the specific context in which the abnormal behavior occurs, which helps the monitoring manager to understand the problem faster and make corresponding decisions. The extracted video clips are sent to the monitoring manager for manual verification, and the verification feedback mark of the manager is received. The introduction of manual verification links can further improve the accuracy and reliability of the system. The professional knowledge and experience of the monitoring manager can make up for the shortcomings of the AI ​​model in some complex or ambiguous scenarios and ensure the accuracy of the alarm triggering. At the same time, the manual verification feedback mechanism also helps to continuously optimize the performance of the AI ​​model and form a virtuous circle. When the verification feedback mark of the monitoring manager matches the preset alarm trigger mark, the system triggers the alarm system, ensuring that the alarm system will only be triggered when it is confirmed that there is real abnormal behavior, avoiding false alarms and missed alarms, and improving the effectiveness and credibility of the alarm system. At the same time, timely alarm notifications help to quickly respond to and handle security incidents and reduce potential losses.

[0117] In one embodiment, a monitoring device is provided, whose internal structure diagram can be as follows: Figure 4 As shown. The monitoring device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor of the monitoring device is used to provide computing and control capabilities. The memory of the monitoring device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the monitoring device is used to communicate with an external client via a network connection.

[0118] In one embodiment, when the processor executes the computer program, the following steps are implemented:

[0119] Detect abnormal behaviors in surveillance videos containing surveillance objects based on AI models;

[0120] When abnormal behavior is detected, obtaining a video clip related to the abnormal behavior from the surveillance video;

[0121] Sending the video clip to a monitoring manager for manual verification, and receiving a verification feedback mark from the monitoring manager;

[0122] When the verification feedback mark matches the preset alarm trigger mark, the alarm system is triggered.

[0123] In one embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0124] Detect abnormal behaviors in surveillance videos containing surveillance objects based on AI models;

[0125] When abnormal behavior is detected, obtaining a video clip related to the abnormal behavior from the surveillance video;

[0126] Sending the video clip to a monitoring manager for manual verification, and receiving a verification feedback mark from the monitoring manager;

[0127] When the verification feedback mark matches the preset alarm trigger mark, the alarm system is triggered.

[0128] It should be noted that the functions or steps that can be implemented by the above-mentioned computer-readable storage medium or monitoring device can be correspondingly referred to in the aforementioned method embodiment. In order to avoid repetition, they will not be described one by one here.

[0129] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0130] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0131] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A security monitoring method based on artificial intelligence, the method comprising: Detect abnormal behaviors in surveillance videos containing surveillance objects based on AI models; When abnormal behavior is detected, obtaining a video clip related to the abnormal behavior from the surveillance video; Sending the video clip to a monitoring manager for manual verification, and receiving a verification feedback mark from the monitoring manager; When the verification feedback mark matches the preset alarm trigger mark, the alarm system is triggered.

2. The artificial intelligence-based security monitoring method according to claim 1, characterized in that: The method further comprises: The AI ​​model is iteratively updated based on the verification feedback mark.

3. The artificial intelligence-based security monitoring method according to claim 1, characterized in that: Before sending the video clip to a monitoring manager for manual verification, the method further includes: Obtaining a detection confidence level corresponding to the abnormal behavior; Classifying the abnormal behavior based on the video clip to obtain a classification type; extracting a plurality of behavioral features from the video clip; Correcting the detection confidence based on the multiple behavior features to obtain a target confidence; Comparing the target confidence with a confidence threshold range corresponding to the classification type; When the target confidence is within the preset confidence threshold range, the video clip is sent to a monitoring manager for manual verification.

4. The artificial intelligence-based security monitoring method according to claim 3 is characterized in that: The sending the video clip to the monitoring manager for manual verification includes: Determine the target monitoring manager identification according to the classification type; The video clip is sent to a target monitoring manager corresponding to the target monitoring manager identification for manual verification.

5. The artificial intelligence-based security monitoring method according to claim 4, characterized in that: The sending the video clip to the target monitoring manager corresponding to the target monitoring manager identifier for manual verification includes: The video clip is encrypted and sent to a target monitoring manager corresponding to the target monitoring manager identifier for manual verification.

6. The artificial intelligence-based security monitoring method according to any one of claims 1 to 5, characterized in that: The method further comprises: A monitoring interface is displayed, wherein the monitoring interface is used to display the video clip and provide the monitoring manager with a function of performing verification feedback marking.

7. The artificial intelligence-based security monitoring method according to any one of claims 1 to 5, characterized in that: The method further comprises: The video clip and the verification feedback mark are stored in association.

8. A security monitoring device based on artificial intelligence, characterized in that: The device comprises: A detection module is used to detect abnormal behaviors of surveillance videos containing surveillance objects based on AI models; An acquisition module, used for acquiring video clips related to the abnormal behavior from the monitoring video when abnormal behavior is detected; A feedback module, used to send the video clip to a monitoring manager for manual verification and receive a verification feedback mark from the monitoring manager; The alarm module is used to trigger the alarm system when the verification feedback mark matches the preset alarm trigger mark.

9. A monitoring device, characterized in that: The monitoring device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the artificial intelligence-based security monitoring method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence-based security monitoring method as described in any one of claims 1 to 7 are implemented.

Images