Laboratory personnel behavior abnormity early warning system based on video monitoring analysis

Abstract

The invention discloses a laboratory personnel behavior abnormity early warning system based on video monitoring analysis, and the system comprises an image collection module which comprises a plurality of types of image collection devices disposed at different positions in a laboratory, and is used for collecting the image information of laboratory personnel in real time; the behavior analysis module is used for analyzing and processing the acquired image information of the laboratory personnel and judging whether abnormal behaviors exist or not; and the warning module is used for triggering a warning signal when the abnormal behavior is detected. According to the invention, the automation, precision and intelligence level of laboratory personnel behavior monitoring is effectively improved, and the technical blank of abnormal behavior detection of the existing system in a complex scene is filled.

Description

Technical Field

[0001] The present invention relates to the field of database security, and in particular to an abnormal behavior early warning system for laboratory personnel based on video surveillance analysis. Background Art

[0002] With the increasing automation and safety management requirements of laboratories, real-time monitoring of personnel behavior and abnormal warnings have become important requirements for ensuring laboratory safety. Traditional laboratory monitoring systems mostly rely on manual video inspections or single-type cameras, which have the following technical bottlenecks:

[0003] Insufficient environmental adaptability: It is difficult to clearly capture people's behavioral characteristics under low-light and complex lighting conditions, and lacks accurate capture of three-dimensional spatial posture;

[0004] Limited abnormal behavior analysis capabilities: The system can only identify simple preset actions and cannot dynamically learn complex behavior patterns. It also lacks effective detection of hidden risks such as abnormal coordinated behavior by multiple people and failure to wear protective equipment.

[0005] Lack of early warning linkage mechanism: The correlation analysis between environmental data and human behavior is not integrated, the early warning scenario is single, and there is a lack of real-time positioning and multi-terminal notification capabilities;

[0006] Inefficient data processing: Raw image data is subject to significant noise interference, making it difficult for traditional algorithms to efficiently process massive video streams. Model updates rely on manual retraining and are unable to adapt to the dynamic changes in laboratory scenarios.

[0007] Therefore, a laboratory personnel behavior abnormality early warning system based on video surveillance analysis is proposed. Summary of the Invention

[0008] The present invention solves the problems existing in the prior art through the following technical solutions, which include:

[0009] Image acquisition module, which includes various types of image acquisition devices set up in different locations in the laboratory, used to collect image information of laboratory personnel in real time;

[0010] Behavior analysis module, used to analyze and process the collected image information of laboratory personnel to determine whether there is abnormal behavior;

[0011] The warning module is used to trigger a warning signal when abnormal behavior is detected.

[0012] Furthermore, the image acquisition device includes:

[0013] High-definition cameras, set up in key areas of the laboratory, are used to collect high-definition video sequences of personnel;

[0014] Infrared thermal imagers are installed in low-light or no-light areas in the laboratory to obtain thermal imaging characteristics of personnel;

[0015] 3D depth camera, installed above the laboratory table or in the aisle, is used to collect the three-dimensional spatial coordinate data of people in the laboratory;

[0016] The three-dimensional space coordinate data includes depth information of the human head, neck, shoulders, elbows, wrists, chest, waist, hips, thighs, knees, calves, and ankles.

[0017] Furthermore, the behavior analysis module includes:

[0018] Motion trajectory analysis unit, used to extract the center of mass motion trajectory of the person based on continuous frame images and judge abnormal behavior through trajectory characteristic parameters;

[0019] The posture recognition unit is used to detect the spatial position of the human head, neck, shoulders, elbows, wrists, chest, waist, hips, thighs, knees, calves and ankles, and identify abnormal postures by calculating joint angles;

[0020] The behavior pattern learning unit is used to train and generate pattern models of normal and abnormal behaviors using historical image data.

[0021] Furthermore, the abnormal behavior judgment method of the motion trajectory analysis unit includes:

[0022] Calculate the change in the coordinates of the center of mass of the person in the continuous frame image, Δx = x t -x t-1 , Δy=y t -y t-1 ;

[0023] Among them, x t is the horizontal coordinate of the center of mass of the person in the current frame image, x t-1 is the horizontal coordinate of the center of mass of the person in the previous frame of image;

[0024] y t is the vertical coordinate of the center of mass of the person in the current frame image, y t-1 is the vertical coordinate of the center of mass of the person in the previous frame of image;

[0025] Abnormality judgment is performed based on trajectory complexity and speed change rate. The trajectory complexity calculation formula is:

[0026]

[0027] Where N is the number of trajectory points, Δx i is the change in the horizontal coordinate of the center of mass of the i-th trajectory point relative to the horizontal coordinate of the center of mass of the previous trajectory point;

[0028] Δy i is the change in the vertical coordinate of the center of mass of the i-th trajectory point relative to the vertical coordinate of the center of mass of the previous trajectory point;

[0029] The speed change rate calculation formula is: Among them, v t is the speed of the person at the current moment, v t-1 is the movement speed of the personnel at the previous moment;

[0030] When Complexity>C threshold , and |Δv|>V threshold It is judged as abnormal behavior;

[0031] Among them, C threshold is the preset trajectory complexity threshold, V threshold is the preset speed change rate threshold.

[0032] Furthermore, the abnormal posture judgment method of the posture recognition unit includes:

[0033] Detect the coordinates of the human head, neck, shoulders, elbows, wrists, chest, waist, hips, thighs, knees, calves and ankles;

[0034] Calculate the joint angles between the head and neck, neck and shoulder, shoulder and elbow, elbow and wrist, shoulder and chest, chest and waist, waist and hip, hip and thigh, thigh and knee, knee and calf, and calf and ankle. The calculation formula is:

[0035]

[0036] Among them, (x p1 ,y p1 ) is the horizontal and vertical coordinates of the first key point constituting the joint in the image coordinate system, (x p2 ,y p2 ) is the horizontal and vertical coordinates of the second key point constituting the joint in the image coordinate system, (x p3 ,y p3 ) are the horizontal and vertical coordinates of the third key point constituting the joint in the image coordinate system.

[0037] Furthermore, the abnormal behavior judgment method of the behavior pattern learning unit includes:

[0038] Use clustering algorithms to classify historical behavior data and generate normal behavior clusters and abnormal behavior clusters;

[0039] Compute the central eigenvectors of each class of clusters:

[0040] Where M is the number of samples in the cluster, F ij is the feature vector of the jth sample in cluster i;

[0041] Extract feature vectors from newly collected data and calculate the cosine similarity with the centers of various clusters. When the similarity with the abnormal cluster is greater than S threshold It is judged as abnormal behavior when S threshold is the preset similarity threshold.

[0042] Furthermore, the warning module includes:

[0043] Local warning unit, including sound and light alarm device, used to issue visual and sound warning signals on the laboratory site;

[0044] A remote notification unit, configured to send notifications of abnormal events, including the time, spatial location, and associated image data of the abnormality;

[0045] The spatial positioning subunit is used to convert the image coordinates into the actual laboratory space coordinates. The conversion formula is:

[0046] f is the focal length of the camera, X image is the horizontal coordinate of the person in the image coordinate system, Y image is the vertical coordinate of the person in the image coordinate system, Z image is the depth coordinate of the person in the image coordinate system, X is the horizontal coordinate of the person in the actual space coordinate system of the laboratory, and Y is the vertical coordinate of the person in the actual space coordinate system of the laboratory.

[0047] Furthermore, the behavior analysis module also includes a behavior correlation analysis unit for detecting abnormal collaborative behavior of multiple people, specifically including:

[0048] Calculate the relative positions and motion parameters between multiple people in real time. For person A and person B, the relative distance formula is:

[0049] Among them, x A is the horizontal coordinate of person A at the current moment, x B is the horizontal coordinate of person B at the current moment, y A is the vertical coordinate of person A at the current moment, y B is the vertical coordinate of person B at the current moment;

[0050] When D AB <D collab , or ΔV AB <V collab , it is determined that there is a tendency of coordinated abnormal behavior, among which D collab is the preset relative distance threshold, Vcollab is the preset relative speed threshold.

[0051] Compared with the existing technology, the present invention has the following advantages: the laboratory personnel behavior abnormality warning system based on video surveillance analysis realizes the comprehensive collection of laboratory personnel image information through the combination of high-definition cameras, infrared thermal imagers and 3D depth cameras, covering high-definition video, thermal imaging features and three-dimensional spatial coordinate data, ensuring the effectiveness of monitoring in complex lighting environments (such as low light and dark scenes), and providing multi-dimensional data support for subsequent behavior analysis.

[0052] Motion trajectory analysis combines trajectory complexity and velocity change rate to effectively identify abnormal movement patterns;

[0053] Posture recognition calculates joint angles through key point coordinates to accurately capture abnormal limb postures. The behavior pattern learning unit uses machine learning algorithms to train historical data, build dynamic behavior models, and improve the generalization detection capabilities of unknown abnormal behaviors. The integration realizes a three-dimensional analysis of human behavior, significantly improving the accuracy and comprehensiveness of abnormal behavior detection.

[0054] It integrates local sound and light alarm and remote notification functions, supports real-time on-site warnings of abnormal events and remote information synchronization, and combines spatial positioning technology to convert image coordinates into actual spatial coordinates, making it easier for managers to quickly locate the location of abnormalities, shorten response time, and improve laboratory safety management efficiency.

[0055] By calculating the relative positions and motion parameters of multiple people in real time and identifying abnormal collaborative behavior trends, it fills the limitations of traditional single-target monitoring and is suitable for security risk warnings in multi-person collaborative scenarios, further enhancing the practicality and security of the system.

[0056] Through the integration of multiple technologies and intelligent analysis, a complete closed loop covering data collection, behavior recognition, and early warning response has been built, effectively improving the automation level and safety protection capabilities of abnormal behavior monitoring of laboratory personnel and reducing manual monitoring costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0058] The following is a detailed description of an embodiment of the present invention. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process. However, the protection scope of the present invention is not limited to the following embodiment.

[0059] like Figure 1 As shown, this embodiment provides a technical solution: a laboratory personnel behavior abnormality warning system based on video surveillance analysis, comprising:

[0060] Image acquisition module, which includes various types of image acquisition devices set up in different locations in the laboratory, used to collect image information of laboratory personnel in real time;

[0061] Behavior analysis module, used to analyze and process the collected image information of laboratory personnel to determine whether there is abnormal behavior;

[0062] The warning module is used to trigger a warning signal when abnormal behavior is detected.

[0063] The image acquisition device includes:

[0064] High-definition cameras, set up in key areas of the laboratory, are used to collect high-definition video sequences of personnel;

[0065] Infrared thermal imagers are installed in low-light or no-light areas in the laboratory to obtain thermal imaging characteristics of personnel;

[0066] 3D depth camera, installed above the laboratory table or in the aisle, is used to collect the three-dimensional spatial coordinate data of people in the laboratory;

[0067] The three-dimensional spatial coordinate data includes depth information of the human head, neck, shoulders, elbows, wrists, chest, waist, hips, thighs, knees, calves, and ankles;

[0068] High-definition cameras provide high-resolution video information, ensuring accurate capture of behavioral details (such as hand operations and body movement trajectories) under normal lighting conditions, providing basic visual data for behavioral analysis;

[0069] Infrared thermal imagers fill in monitoring blind spots in low-light or dark environments, achieve 24-hour uninterrupted monitoring through thermal signal imaging, avoid monitoring failures caused by changes in ambient light, and improve the system's adaptability in complex lighting conditions;

[0070] The combination of the two covers all lighting scenarios from strong light to no light, ensuring that there are no time blind spots in monitoring.

[0071] 3D depth cameras acquire a person's three-dimensional spatial coordinate data (including depth information of key points such as the head, shoulders, and hands). Compared to traditional two-dimensional images, they can accurately restore the body's position and posture in space, providing three-dimensional spatial data for subsequent posture recognition and motion trajectory analysis, thus resolving the problem of behavioral analysis errors caused by the lack of depth information in two-dimensional images.

[0072] The three devices output video, thermal imaging, and three-dimensional coordinate data respectively, forming a multimodal data source, supporting the behavior analysis module to extract features from different dimensions (such as appearance features, thermal signal features, and spatial position features), thereby improving the comprehensiveness and accuracy of abnormal behavior detection.

[0073] Multiple types of equipment utilize hardware redundancy to reduce the impact of single device failure or environmental limitations on overall monitoring performance. For example, infrared thermal imagers can still stably output thermal signal data even when the camera is overexposed to strong light, ensuring system reliability in extreme environments.

[0074] Data complementarity avoids the detection blind spots of a single sensor. For example, the 3D depth camera's penetrating detection capability for scenes obstructed by limbs complements the high-definition camera's ability to capture details, improving the ability to analyze human behavior in complex scenarios.

[0075] By limiting specific types of image acquisition equipment and their collaborative working mechanisms, the entire system is provided with basic data acquisition capabilities for all time periods, all scenarios, and high precision. This is the core technical foundation for subsequent behavior analysis and abnormal warnings, and ensures the system's environmental adaptability, data richness, and detection reliability from the hardware level.

[0076] The behavior analysis module includes:

[0077] Motion trajectory analysis unit, used to extract the center of mass motion trajectory of the person based on continuous frame images and judge abnormal behavior through trajectory characteristic parameters;

[0078] The posture recognition unit is used to detect the spatial position of the human head, neck, shoulders, elbows, wrists, chest, waist, hips, thighs, knees, calves and ankles, and identify abnormal postures by calculating joint angles;

[0079] Behavior pattern learning unit, used to train and generate pattern models of normal and abnormal behaviors using historical image data;

[0080] The system extracts motion trajectories by analyzing the changes in the centroid coordinates of consecutive frames, quantifies the trajectory complexity and speed change rate, and effectively identifies abnormal movement patterns (such as irregular and rapid movement, frequent turns, etc.). This overcomes the limitation of traditional monitoring that relies solely on static images and enables temporal analysis of people's dynamic behavior.

[0081] The system calculates joint angles based on the spatial positions of key human body points (head, shoulders, hands, etc.), accurately capturing abnormal limb postures (such as excessive bending, abnormally raised hand, and body imbalance during operation). This system is suitable for monitoring standard operating postures in the laboratory (such as reagent collection and equipment debugging), filling the gap in the inability of simple trajectory analysis to identify local limb abnormalities.

[0082] A behavioral feature matrix is ​​constructed from the two dimensions of "overall motion trajectory and local limb posture", covering the dynamic process and static posture of human behavior, avoiding the problem of missed detection caused by single-dimensional analysis.

[0083] By using machine learning algorithms to train historical data, it automatically generates models of normal and abnormal behavior patterns that conform to specific laboratory scenarios. This overcomes the limitations of traditional rule engines that rely on manually preset thresholds (such as rule failures caused by differences in laboratory personnel's operating habits and equipment layout).

[0084] By using clustering and other algorithms to identify unknown abnormal behavior patterns, it has the ability to detect newly emerging abnormal behaviors (such as illegal operations not recorded in historical data), thereby improving the adaptability and intelligence level of the system;

[0085] Closed-loop optimization mechanism: As new data accumulates, the model can be dynamically updated to continuously optimize detection accuracy and avoid detection errors caused by changes in personnel behavior or laboratory process adjustments;

[0086] The motion trajectory analysis unit focuses on the trajectory of a person's center of mass, providing a macroscopic view of their movement in space and identifying any abnormal movements such as sudden acceleration or frequent reversals. The posture recognition unit delve deeper into the human posture, analyzing the positions of key points and joint angles to identify abnormal postures such as excessive bending and twisting of limbs. The behavioral pattern learning unit leverages historical image data to establish models of normal and abnormal behavior, detecting anomalies based on behavioral habits and patterns. These three units monitor human behavior from different dimensions, significantly improving the comprehensiveness of abnormal behavior detection.

[0087] Capturing Subtle Abnormal Behaviors: The three units work together to uncover subtle anomalies that are difficult to detect using a single analysis method. For example, trajectory analysis alone may not be able to determine whether a seemingly normal person's posture is abnormal, but combined with the posture recognition unit, it can more accurately identify problems. The behavior pattern learning unit, on the other hand, can learn from long-term data and discover potentially abnormal behavior patterns that are difficult to identify using conventional trajectory and posture analysis.

[0088] The abnormal behavior judgment method of the motion trajectory analysis unit includes:

[0089] Calculate the change in the coordinates of the center of mass of the person in the continuous frame image, Δx = x t -x t-1 , Δy=y t -y t-1 ;

[0090] Among them, x t is the horizontal coordinate of the center of mass of the person in the current frame image, x t-1is the horizontal coordinate of the center of mass of the person in the previous frame of image;

[0091] y t is the vertical coordinate of the center of mass of the person in the current frame image, y t-1 is the vertical coordinate of the center of mass of the person in the previous frame of image;

[0092] Abnormality judgment is performed based on trajectory complexity and speed change rate. The trajectory complexity calculation formula is:

[0093]

[0094] Where N is the number of trajectory points, Δx i is the change in the horizontal coordinate of the center of mass of the i-th trajectory point relative to the horizontal coordinate of the center of mass of the previous trajectory point;

[0095] Δy i is the change in the vertical coordinate of the center of mass of the i-th trajectory point relative to the vertical coordinate of the center of mass of the previous trajectory point;

[0096] Δv=v t -v t-1

[0097] The speed change rate calculation formula is: t-1 , where v t is the speed of the person at the current moment, v t-1 is the movement speed of the personnel at the previous moment;

[0098] When Complexity>C threshold , and |Δv|>V threshold It is judged as abnormal behavior;

[0099] Among them, C threshold is the preset trajectory complexity threshold, V threshold is the preset speed change rate threshold;

[0100] The historical image data stored in the database module (including raw data such as personnel trajectories and postures in normal and abnormal scenarios) and the feature data output by the behavior analysis module (such as trajectory complexity and joint angle anomalies) provide sufficient training material for the behavior pattern learning unit, solving the one-sided problem of single real-time data detection. For example, by using long-term accumulation of experimental operation images, the model can accurately learn normal behavior patterns in different experimental stages (preparation, operation, and conclusion), avoiding misjudgments caused by short-term data deviations;

[0101] As new data continues to be imported into the database, the behavioral pattern learning unit can perform incremental training based on historical data and automatically update the anomaly detection threshold (such as adjusting the normal movement speed range of personnel in the low-temperature laboratory according to seasonal changes), enabling the system to adapt to changes in the laboratory environment (such as equipment layout adjustments, operating process updates) or changes in personnel habits (such as new employees' operating modes), avoiding the lag of traditional fixed rule engines.

[0102] The database module associates and stores raw image data, feature extraction results (such as the center of mass coordinate sequence and joint angle matrix at a specific moment), and anomaly detection logs (the specific feature combination that triggered the warning), forming a complete behavioral analysis chain. When a safety accident occurs in the laboratory, the complete process of abnormal behavior can be reproduced by retrieving the trajectory-posture-pattern data at a specific time, providing an objective basis for accident cause analysis and responsibility determination.

[0103] For laboratories that need to meet strict compliance requirements (such as biosafety laboratories and hazardous chemical operation rooms), the long-term data storage function of the database module can meet the regulatory agency's requirements for the retention period of operation records and avoid compliance risks caused by missing data.

[0104] The behavior analysis module can call upon historical abnormal case data in the database to perform secondary verification on current detection results. For example, when a combination of rapid movement and "raising one hand high" is detected in real time, the system can search the historical database to see if a similar pattern has been flagged as illegal reagent dumping. This improves the accuracy of identifying complex abnormal behaviors and reduces false positives based on single features.

[0105] By analyzing historical behavioral data of the same person or area, gradual anomalies can be identified (such as an employee's occasional irregular operating posture gradually evolving into habitual violations), which makes up for the deficiency of real-time detection that only focuses on immediate behavior and provides early warning of long-term safety hazards.

[0106] The abnormal posture judgment method of the posture recognition unit includes:

[0107] Detect the coordinates of the human head, neck, shoulders, elbows, wrists, chest, waist, hips, thighs, knees, calves and ankles;

[0108] Calculate the joint angles between the head and neck, neck and shoulder, shoulder and elbow, elbow and wrist, shoulder and chest, chest and waist, waist and hip, hip and thigh, thigh and knee, knee and calf, and calf and ankle. The calculation formula is:

[0109]

[0110] Among them, (x p1 ,y p1) is the horizontal and vertical coordinates of the first key point constituting the joint in the image coordinate system, (x p2 ,y p2 ) is the horizontal and vertical coordinates of the second key point constituting the joint in the image coordinate system, (x p3 ,y p3 ) are the horizontal and vertical coordinates of the third key point constituting the joint in the image coordinate system.

[0111] The abnormal behavior judgment method of the behavior pattern learning unit includes:

[0112] Use clustering algorithms to classify historical behavior data and generate normal behavior clusters and abnormal behavior clusters;

[0113] Compute the central eigenvectors of each class of clusters:

[0114] Where M is the number of samples in the cluster, F ij is the feature vector of the jth sample in cluster i;

[0115] Extract feature vectors from newly collected data and calculate the cosine similarity with the centers of various clusters. When the similarity with the abnormal cluster is greater than S threshold It is judged as abnormal behavior when S threshold is the preset similarity threshold;

[0116] The early warning module automatically generates multi-level warnings based on the characteristic data output by the behavior analysis module (such as the duration of abnormal behavior, the scope of impact, and the accident level of historical similar cases). For example, if a person is detected staying in a hazardous chemical storage area for a long time and in an abnormal posture (such as frequently bending over to pick up and put away chemicals), the system can combine historical data to determine whether there is a risk of illegal use and trigger an audible and visual alarm of the corresponding level;

[0117] Dynamic response strategies adapt to scenario requirements: For different laboratory scenarios (such as high-temperature electric furnace rooms and biological safety cabinet operation areas), differentiated linkage rules can be preset (such as the first-level warning only triggers the regional warning light, the second-level warning automatically cuts off the power supply of the equipment, and the third-level warning links the access control blockade and starts the fire sprinkler), solving the scenario adaptability problem of traditional fixed threshold warnings, ensuring that the response measures neither excessively interfere with normal operations nor block the risk chain before the danger escalates.

[0118] Zero-delay response with full system coordination: The early warning module communicates real-time data with laboratory infrastructure (such as fume hoods, gas leak detectors, and fire alarm systems), automatically executing control instructions while confirming abnormal behavior. For example, when a person is detected entering a highly toxic reagent area without protective equipment and a level 3 warning is triggered, the system will simultaneously complete the following: close the access control to the area to prevent more people from entering; activate the ventilation system to accelerate air replacement; and send a location alarm message to the administrator's mobile phone, compressing manual decision-making time to seconds and minimizing the scope of the accident.

[0119] Reverse control prevents the spread of risk: By linking devices in reverse (e.g., cutting off reagent supply lines to abnormal areas and locking down lab equipment in violation of regulations), further dangerous behavior can be physically prevented. For example, upon detecting unauthorized overclocking of high-temperature equipment, the system automatically reduces its power and triggers a forced cooling process, mitigating the risk of fire or explosion caused by human error or equipment failure. This upgrades safety management from post-event accountability to pre-event blocking.

[0120] For abnormal scenarios with high repetition rates and clear risks (such as contact with corrosive reagents without wearing gloves according to procedures), the equipment linkage mechanism can automatically complete standard response actions, relieving safety management personnel of the monitoring pressure. For complex or ambiguous abnormal behaviors (such as large groups of people gathering with signs of physical conflict), the system uses graded warnings to accurately push information to designated responsible persons, assisting in rapid human decision-making, forming an efficient collaborative model where "machines handle standardized risks, while humans handle complex scenarios."

[0121] Real-time verification of operational compliance: Linked control can serve as a digital gatekeeper for operational processes. For example, if laboratory personnel have not completed pre-job training and signed in, the access control system automatically prohibits them from entering high-risk areas. When the ventilation equipment is not started, the electronic lock of the toxic reagent cabinet is prohibited from being opened. This eliminates compliance risks caused by human oversight at the source and ensures that laboratory operations strictly follow standard operating procedures.

[0122] The warning module includes:

[0123] Local warning unit, including sound and light alarm device, used to issue visual and sound warning signals on the laboratory site;

[0124] A remote notification unit, configured to send notifications of abnormal events, including the time, spatial location, and associated image data of the abnormality;

[0125] The spatial positioning subunit is used to convert the image coordinates into the actual laboratory space coordinates. The conversion formula is:

[0126] f is the focal length of the camera, X imageis the horizontal coordinate of the person in the image coordinate system, Y image is the vertical coordinate of the person in the image coordinate system, Z image is the depth coordinate of the person in the image coordinate system, X is the horizontal coordinate of the person in the actual space coordinate system of the laboratory, and Y is the vertical coordinate of the person in the actual space coordinate system of the laboratory;

[0127] The local warning unit uses audio and visual alarms (such as flashing lights and buzzing alarms) to emit strong signals within the laboratory, instantly attracting personnel's attention. This is particularly useful in noisy environments where personnel are focused on operations. This intuitive warning method avoids the delays associated with manual monitoring, ensuring that abnormal events are immediately detected by on-site personnel, buying time for initial risk control and reducing the likelihood of an incident escalating.

[0128] The remote notification unit pushes key information about abnormal events (such as the time of occurrence, specific location, and related images) to management terminals (such as mobile phones and computers) in real time via the network, solving the problem of traditional early warning systems that alarm on-site but are not perceived remotely. Even if managers are not on-site in the laboratory, they can quickly determine the risk level and initiate the disposal process based on complete information (such as screenshots of abnormal behavior and precise location descriptions), shortening decision-making time and improving cross-space collaboration efficiency.

[0129] By using technical means to convert the positions of people in the monitoring screen into the actual spatial coordinates of the laboratory, the specific location where abnormal behavior occurs can be accurately identified.

[0130] The behavior analysis module also includes a behavior correlation analysis unit for detecting abnormal collaborative behavior of multiple people, specifically including:

[0131] Calculate the relative positions and motion parameters between multiple people in real time. For person A and person B, the relative distance formula is:

[0132] Among them, x A is the horizontal coordinate of person A at the current moment, x B is the horizontal coordinate of person B at the current moment, y A is the vertical coordinate of person A at the current moment, y B is the vertical coordinate of person B at the current moment;

[0133] When D AB <D collab , or ΔV AB <V collab , it is determined that there is a tendency of coordinated abnormal behavior, among which D collab is the preset relative distance threshold, V collab is the preset relative speed threshold;

[0134] The establishment of a behavioral correlation analysis unit focuses on the detection of abnormal collaborative behavior among multiple people, filling the technical gap of traditional single-target monitoring and significantly improving the system's security protection capabilities in complex collaborative scenarios. This unit calculates the relative positions (relative distances) and motion parameters (relative speeds) of multiple people in real time, constructs a dynamic correlation model, and can accurately identify abnormal collaborative tendencies among multiple people (such as collective illegal operations and organized dangerous behaviors). Compared with the behavior analysis of a single person, its core advantages are:

[0135] It effectively solves the risk monitoring needs when multiple people cooperate in the laboratory, and avoids the problem of missed detection of collaborative anomalies caused by independent analysis of a single target. It is especially suitable for high-risk experimental scenarios that require collaboration among multiple people (such as chemical synthesis and precision equipment debugging).

[0136] By quantifying parameters such as relative distance and speed, combined with preset collaborative behavior thresholds, real-time dynamic analysis of the correlation between multiple people's behaviors can be achieved. This can quickly identify risky behaviors such as abnormal aggregation and synchronous abnormal movement, providing immediate decision-making basis for safety management.

[0137] It makes up for the limitation of traditional systems that only focus on individual behavior, upgrades from "single individual monitoring" to collaborative analysis of group behavior, enhances the detection ability of complex risks such as organized violations and multi-person coordination errors, and further strengthens the laboratory safety line.

[0138] As an organic part of the behavior analysis module, this unit supports data interaction and collaborative verification with other functional units (such as trajectory analysis and posture recognition). It can flexibly adjust detection parameters according to scenarios such as laboratory layout and personnel density to improve the overall adaptability of the system.

[0139] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0140] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0141] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A laboratory personnel behavior abnormality warning system based on video surveillance analysis, characterized in that: include: Image acquisition module, which includes various types of image acquisition devices set up in different locations in the laboratory, used to collect image information of laboratory personnel in real time; Behavior analysis module, used to analyze and process the collected image information of laboratory personnel to determine whether there is any abnormal behavior; The warning module is used to trigger a warning signal when abnormal behavior is detected.

2. The laboratory personnel behavior abnormality early warning system based on video surveillance analysis according to claim 1 is characterized by: The image acquisition device includes: High-definition cameras, set up in key areas of the laboratory, are used to collect high-definition video sequences of personnel; Infrared thermal imagers are installed in low-light or no-light areas in the laboratory to obtain thermal imaging characteristics of personnel; 3D depth camera, installed above the laboratory table or in the aisle, is used to collect the three-dimensional spatial coordinate data of people in the laboratory; The three-dimensional space coordinate data includes depth information of the human head, neck, shoulders, elbows, wrists, chest, waist, hips, thighs, knees, calves, and ankles.

3. The laboratory personnel behavior abnormality early warning system based on video surveillance analysis according to claim 1 is characterized by: The behavior analysis module includes: Motion trajectory analysis unit, used to extract the center of mass motion trajectory of the person based on continuous frame images and judge abnormal behavior through trajectory characteristic parameters; The posture recognition unit is used to detect the spatial position of the human head, neck, shoulders, elbows, wrists, chest, waist, hips, thighs, knees, calves and ankles, and identify abnormal postures by calculating joint angles; The behavior pattern learning unit is used to train and generate pattern models of normal and abnormal behaviors using historical image data.

4. The laboratory personnel behavior abnormality early warning system based on video surveillance analysis according to claim 3 is characterized by: The abnormal behavior judgment method of the motion trajectory analysis unit includes: Calculate the change in the coordinates of the center of mass of the person in the continuous frame image, Δx = x t -x t-1 , Δy=y t -y t-1 ; Among them, x t is the horizontal coordinate of the center of mass of the person in the current frame image, x t-1 is the horizontal coordinate of the center of mass of the person in the previous frame of image; y t is the vertical coordinate of the center of mass of the person in the current frame image, y t-1 is the vertical coordinate of the center of mass of the person in the previous frame of image; Abnormality judgment is performed based on trajectory complexity and speed change rate. The trajectory complexity calculation formula is: Where N is the number of trajectory points, Δx i is the change in the horizontal coordinate of the center of mass of the i-th trajectory point relative to the horizontal coordinate of the center of mass of the previous trajectory point; Δy i is the change in the vertical coordinate of the center of mass of the i-th trajectory point relative to the vertical coordinate of the center of mass of the previous trajectory point; The speed change rate calculation formula is: Among them, v t is the speed of the person at the current moment, v t-1 is the movement speed of the personnel at the previous moment; When Complexity>C threshold , and |Δv|>V threshold It is considered abnormal behavior; Among them, C threshold is the preset trajectory complexity threshold, V threshold is the preset speed change rate threshold.

5. The laboratory personnel behavior abnormality warning system based on video surveillance analysis according to claim 3 is characterized by: The abnormal posture judgment method of the posture recognition unit includes: Detect the coordinates of the human head, neck, shoulders, elbows, wrists, chest, waist, hips, thighs, knees, calves and ankles; Calculate the joint angles between the head and neck, neck and shoulder, shoulder and elbow, elbow and wrist, shoulder and chest, chest and waist, waist and hip, hip and thigh, thigh and knee, knee and calf, and calf and ankle. The calculation formula is: Among them, (x p1 ,y p1 ) is the horizontal and vertical coordinates of the first key point constituting the joint in the image coordinate system, (x p2 ,y p2 ) is the horizontal and vertical coordinates of the second key point constituting the joint in the image coordinate system, (x p3 ,y p3 ) are the horizontal and vertical coordinates of the third key point constituting the joint in the image coordinate system.

6. The laboratory personnel behavior abnormality warning system based on video surveillance analysis according to claim 1 is characterized by: The abnormal behavior judgment method of the behavior pattern learning unit includes: Use clustering algorithms to classify historical behavior data and generate normal behavior clusters and abnormal behavior clusters; Compute the central eigenvectors of each class of clusters: Where M is the number of samples in the cluster, F ij is the feature vector of the jth sample in cluster i; Extract feature vectors from newly collected data and calculate the cosine similarity with the centers of various clusters. When the similarity with the abnormal cluster is greater than S threshold It is judged as abnormal behavior when S threshold is the preset similarity threshold.

7. The laboratory personnel behavior abnormality early warning system based on video surveillance analysis according to claim 1 is characterized by: The warning module includes: Local warning unit, including sound and light alarm device, used to issue visual and sound warning signals on the laboratory site; A remote notification unit, configured to send notifications of abnormal events, including the time, spatial location, and associated image data of the abnormality; The spatial positioning subunit is used to convert the image coordinates into the actual laboratory space coordinates. The conversion formula is: f is the focal length of the camera, X image is the horizontal coordinate of the person in the image coordinate system, Y image is the vertical coordinate of the person in the image coordinate system, Z image is the depth coordinate of the person in the image coordinate system, X is the horizontal coordinate of the person in the actual space coordinate system of the laboratory, and Y is the vertical coordinate of the person in the actual space coordinate system of the laboratory.

8. The laboratory personnel behavior abnormality early warning system based on video surveillance analysis according to claim 1 is characterized by: The behavior analysis module also includes a behavior correlation analysis unit for detecting abnormal collaborative behavior of multiple people, specifically including: Calculate the relative positions and motion parameters between multiple people in real time. For person A and person B, the relative distance formula is: Among them, x A is the horizontal coordinate of person A at the current moment, x B is the horizontal coordinate of person B at the current moment, y A is the vertical coordinate of person A at the current moment, y B is the vertical coordinate of person B at the current moment; When D AB <D collab , or ΔV AB <V collab , it is determined that there is a tendency of coordinated abnormal behavior, among which D collab is the preset relative distance threshold, V collab is the preset relative speed threshold.

Images