Method and system for monitoring abnormal entry of family farm based on multi-event split-screen monitoring

Abstract

The application discloses a kind of based on multi-event split-screen monitoring's family farm approach abnormal supervision method and system, comprising: obtaining approach information, generation operation step sequence is automatically completed by central server and is assessed priority according to emergency degree and risk score, obtain structured event queue;Active event and the total number of camera are obtained, matching split-screen specification template, combined with show weight descending distribution monitoring block and using information gain processing overflow, output dynamic split-screen layout;Motion detection is carried out to video stream, using weighted dynamic time warping algorithm and standard template time alignment, calculate deviation and trigger visual alarm;Deviation behavior is input into violation feature library and judged, intercept evidence video, and update farm risk score;Based on event load and abnormal distribution, resource adaptive scheduling is carried out, adjustment strategy is generated, and performance portrait is generated in combination with special officer response condition, and reinforcement strategy is obtained by optimizing scheduling. Thus, the accuracy and timeliness of family farm supervision are improved.

Description

Technical Field

[0001] This invention relates to the field of smart farm management technology, and in particular to a method and system for monitoring abnormal entry into family farms based on multi-event split-screen monitoring. Background Technology

[0002] With the promotion of large-scale and intensive farming models in family farms, the demand for unmanned and refined supervision of multi-channel entry behavior on farms is becoming increasingly urgent. Family farms typically have multiple dispersed entry points, including the main gate driveway, the entrance to the livestock pens, the entrance to the storage area, and personnel disinfection channels. These entrances are spatially independent but form a continuous biosecurity defense line in terms of business logic.

[0003] Currently, monitoring solutions for such scenarios generally employ fixed-view rotation or single motion detection alarm modes. In fixed-view rotation mode, the monitoring screen switches between different channels sequentially according to a preset order. Video specialists can only focus on the real-time status of a single entrance at any given time, making it impossible to have a global awareness of concurrent entry events at multiple entrances. This often leads to the omission of key abnormal behaviors during rotation intervals. Meanwhile, single motion detection alarms only apply threshold judgments to instantaneous motion features within a single video stream, lacking the ability to analyze the behavioral correlations of targets such as people, vehicles, and livestock across entrances and time periods. They cannot effectively identify complex anomalies such as drivers leaving without properly disembarking after entering the area, or livestock leaving their enclosures outside of grazing periods, resulting in high false alarm and false negative rates. Furthermore, existing technologies lack dynamic adaptability in the allocation of monitoring resources. The density of events occurring at family farms exhibits significant temporal imbalances. Peak periods for personnel and material arrivals are concentrated in specific timeframes. Fixed monitoring strategies result in insufficient regulatory capacity during peak periods and idle manpower during off-peak periods, making it impossible to adaptively adjust based on real-time event load and abnormal distribution patterns.

[0004] Therefore, there is an urgent need for a method that can dynamically organize multi-source entry events into a visual layout for multi-entry concurrent scenarios, perform in-depth anomaly analysis on cross-channel behavior sequences, and adaptively schedule regulatory resources based on real-time situation, so as to improve the accuracy and timeliness of family farm supervision. Summary of the Invention

[0005] This invention overcomes the shortcomings of the prior art and provides a method and system for monitoring abnormal entry into family farms based on multi-event split-screen monitoring. Its main purpose is to improve the accuracy and timeliness of family farm monitoring.

[0006] To achieve the above objectives, the first aspect of this invention provides a method for monitoring abnormal entry into family farms based on multi-event split-screen monitoring, comprising: The original entry reporting information is obtained by using the intelligent reporting terminal deployed at the family farm's entry channel. The information is then automatically completed in the central monitoring server based on the basic data configuration library to generate an ordered sequence of operation steps. Priority is assessed according to the urgency of the event and the farm's risk score, and finally a structured entry event queue is obtained. Based on the structured entry event queue, the current active events and the total number of deduplicated associated cameras are obtained, matched with the predefined split-screen specification template, and the comprehensive display weight of each event is calculated synchronously. Adjacent rectangular monitoring blocks are allocated in the split-screen grid in descending order of weight, and the screen overflow is handled with an information gain strategy to output a dynamic split-screen monitoring layout. The camera video stream associated with each event in the structured entry event queue is obtained, and the boundary trigger event sequence is obtained by performing motion detection in the region of interest. The weighted dynamic time warping algorithm is used to align the timing with the standard step template and calculate the comprehensive execution deviation. The comprehensive execution deviation event is determined and the split-screen visual alarm is rendered. The system obtains the observed behavior sequence corresponding to the comprehensive execution deviation event, inputs it into the pre-built violation behavior feature library for violation judgment, and when a violation is determined, it extracts the violation evidence video from the camera video stream, uses the exponential weighted moving average model to recursively update the farm dynamic risk score, and outputs the violation record and the updated farm risk score. Based on the structured entry event queue and violation records, adaptive scheduling of regulatory resources is carried out for event load and abnormal distribution. Regulatory adjustment strategies including abnormal judgment thresholds and split-screen refresh frequency are generated. Simultaneously, individual performance profiles are generated by video specialists based on their response to alarm events, and human resource support strategies are obtained to optimize the current regulatory layout.

[0007] In this solution, the original entry reporting information is obtained by using intelligent reporting terminals deployed at the family farm's entry channel. The information is then automatically completed in the central monitoring server based on the basic data configuration database to generate an ordered sequence of operational steps. Priority is assessed according to event urgency and farm risk score, ultimately resulting in a structured entry event queue, which specifically includes: When the smart reporting terminal deployed at the entrance of each family farm is triggered when personnel or materials enter the farm, it reads the fixed farm unique identifier and channel number and obtains the timestamp from the real-time clock module. It binds the triggered operation type code value and the pre-stored standard operating procedure identifier to form the original reporting message and transmits it to the central monitoring server. The central monitoring server extracts the farm identification code and standard operating procedure identification number from the received original reporting message as a joint query key, and initiates a step-by-step search to the basic data configuration database. The basic data configuration database stores the group-level unified operating standards, the four-level company-level supplementary operating requirements, and the service department-level detailed rules for the layout of specific farm enclosures. During the step-by-step retrieval process, the basic operation step skeleton is obtained through the corresponding process identifier number. Then, the incremental steps corresponding to the farm identifier code are merged into the basic operation step skeleton according to the rules. Then, the corresponding regional classification code and camera identifier set are read and attached to the attribute fields of each step node to generate an ordered operation step sequence. A globally unique event number is assigned to the current inbound event using a distributed sequence generator. The event number, along with the ordered work step sequence, event type, and timestamp, are encapsulated into a structured event object. The urgency coefficient constant corresponding to the event type and the recent historical risk score are read and weighted summed to obtain the priority field of the current event object. The structured event objects with assigned priorities are pushed into the binary heap queue of pending events. The heap order is arranged in descending order of priority scores, and finally the structured incoming event queue carrying complete metadata and feature vectors is output.

[0008] In this solution, the process of obtaining the current active events and the total number of deduplicated associated cameras based on the structured entry event queue, matching them with a predefined split-screen specification template, simultaneously calculating the comprehensive display weight for each event, allocating adjacent rectangular monitoring blocks in the split-screen grid in descending order of weight, handling screen overflow with an information gain strategy, and outputting a dynamic split-screen monitoring layout specifically includes: The active events that are currently being executed or whose planned time falls into the activation time window are selected from the structured entry event queue. The camera identifier set associated with each event is extracted and deduplicated to count the total number of video streams to be displayed. The minimum accommodating split layout is selected by comparing it with the predefined split screen specification template. If the maximum number of splits is exceeded, the fixed display set and the carousel set are divided according to the priority score. For each active event, a comprehensive display weight is calculated. The comprehensive display weight is obtained by multiplying the event type sensitivity coefficient, the complement mapping value of the farm normalized biosafety level score, and the emergency quantification value calculated by the proportion of completed steps and the planned time offset through a nonlinear function. All active events are then arranged in descending order according to the obtained weights. The split-screen view is abstracted into a two-dimensional grid matrix. A greedy maximum empty rectangle search algorithm is used to assign a rectangular monitoring block composed of adjacent cells to each active event. The cells in the block are filled with the real-time video stream of the corresponding camera in the order of the ordered operation steps, forming a spatial arrangement in the same direction as the target active event flow. If the remaining grid area cannot fully accommodate all the camera windows required for an event during the allocation process, the information gain is measured by the frequency with which each camera is associated as a source of evidence in historical violation records. The key control point images are selected and placed in the main display window, and the video streams of the remaining cameras are compressed into thumbnails and floated in a picture-in-picture format. The system integrates the video stream identifiers, grid coordinates, and thumbnail mapping relationships of each window to generate the current dynamic split-screen monitoring layout. When the structured entry event queue is added or deleted, or when the overall display weight of any event fluctuates beyond a set threshold, incremental layout recalculation is triggered to update the split-screen monitoring layout.

[0009] In this solution, the steps of obtaining the camera video streams associated with each event in the structured entry event queue, performing motion detection in the region of interest to obtain the boundary trigger event sequence, using a weighted dynamic time warping algorithm and a standard step template for time sequence alignment and calculating the comprehensive execution deviation, determining the comprehensive execution deviation event, and rendering visual alarms for split-screen displays specifically include: In the structured entry event queue, the camera video streams associated with each event are obtained. The camera identifier set attached to each step node is parsed from the ordered operation step sequence of each event and a video stream subscription relationship is established. In each video frame, an irregular region of interest corresponding to the step area classification code is defined. The motion detection module based on Gaussian mixture model is used to analyze video frames one by one. When a foreground target enters or exits a defined irregular region of interest, a boundary trigger event containing region code, entry / exit direction mark and frame-level timestamp is generated and aggregated into a boundary trigger event sequence according to time sequence. From the ordered sequence of work steps attached to the structured entry event queue, the median of the expected time interval and the regional classification code of each step node are read in sequence. A standard step template composed of the expected regional code and the standard duration constraint is constructed. The arrangement order of the template elements is strictly consistent with the standard process of the work steps. The weighted dynamic time warping algorithm is used to calculate the optimal alignment path between the boundary trigger event sequence and the standard step template. A cumulative cost matrix is ​​constructed with the length of the boundary trigger event sequence as the number of rows and the length of the standard step template as the number of columns. The cost of each cell in the matrix is ​​composed of the weighted sum of the region coding matching cost and the time interval deviation cost. The path with the minimum cumulative cost is searched in the cumulative cost matrix as the optimal mapping relationship. When the cumulative cost of the optimal path is lower than the matching threshold of the target farm, the event segments in the boundary trigger event sequence are assigned to the corresponding steps according to the regularized path index and the actual start and end times are written back. If the matching threshold is exceeded, a step abnormality warning signal is generated. After a successful match, the deviation of each matched step is calculated based on the actual duration and the standard duration. Then, the deviations of all steps are weighted and averaged using the biosafety criticality coefficient of each step as the weight to obtain the overall execution deviation of the target event. When the overall execution deviation exceeds the preset attention threshold, the target event is determined to be an overall execution deviation event, and a visual alarm signal is sent to the split-screen rendering pipeline to change the border color scale of the monitoring block corresponding to the target event.

[0010] In this solution, the process of obtaining the observed behavior sequence corresponding to the comprehensive execution deviation event, inputting it into a pre-built violation behavior feature database for violation determination, and when a violation is determined to exist, extracting violation evidence video from the camera video stream, recursively updating the farm's dynamic risk score using an exponentially weighted moving average model, and outputting the violation record and the updated farm risk score, specifically includes: Extract the corresponding converged boundary trigger event sequence from the event records of the comprehensive execution deviation event as the observation behavior sequence, and extract all relevant spatiotemporal state machine instances from the violation behavior feature library according to the event type code; In the violation behavior feature library, each violation rule is a state machine consisting of several state nodes and directed transition edges with time constraints. The state nodes represent the occurrence of target appearance or target disappearance events in the region of interest corresponding to the specific region encoding, and the directed transition edges specify the order of transition between states and the maximum allowed time interval threshold. The observed behavior sequence is used as the input event stream. Boundary trigger events are read in order of timestamp. Each time an event is read, all active state machines are traversed to determine whether the current event meets the triggering condition and to perform state transition. When any state machine transitions to the accepted state marked as a violation along the directed edge, a violation judgment is triggered and the corresponding violation type identifier is recorded. After a violation is detected, the video data segment of a preset duration is extracted based on the boundary trigger event timestamp corresponding to the predecessor state node associated with the acceptance state, which serves as the starting point for evidence extraction. This generates a structured violation record containing a violation type identifier, a violation occurrence timestamp, a farm identifier, an event number, and a violation evidence video, and writes it into the negative list database. The current risk score of the target farm is obtained and multiplied by a preset attenuation factor to obtain the attenuated historical component. The preset fixed contribution coefficient corresponding to the newly occurred violation is multiplied by the severity weight of the violation to obtain the current increment. This increment is added to the attenuated historical component to generate the updated dynamic risk score of the farm.

[0011] In this solution, the adaptive scheduling of regulatory resources based on structured entry event queues and violation records, oriented towards event load and anomaly distribution, generates regulatory adjustment strategies including anomaly judgment thresholds and split-screen refresh frequencies. Simultaneously, individual performance profiles are generated based on video specialists' response to alarm events to optimize the current regulatory layout and obtain manpower reinforcement strategies. Specifically, this includes: The system runtime is divided into continuous time slices with a fixed duration. The actual number of events and the number of violations in each time slice are extracted from the structured entry event queue and violation records, respectively. The historical sequence of the same time slice in several past periods is traced back, and the event arrival density prediction baseline and the abnormal event ratio prediction baseline are generated using the exponential smoothing algorithm. The number of actual arrival events is counted in real time within the current time slice. When the number of actual arrival events exceeds the product of the event arrival density prediction baseline and the preset expansion coefficient, it is determined that the event is in a high load state. A first type of adjustment instruction is generated, which includes lowering the comprehensive execution deviation alarm threshold and increasing the refresh frame rate of the split-screen monitoring layout, and then issued for execution. Within the current time slice, the proportion of the number of actual triggered comprehensive execution deviation warning events to the total number of actual arrival events is counted. When the proportion exceeds the product of the abnormal event ratio prediction baseline and the preset amplification factor, it is determined to be in an abnormal high-occurrence state, and a second type of adjustment instruction is generated to lock the high-abnormal farm video window and prevent it from being rotated and push reinforcement requests. The system collects the response time difference of each alarm event by video specialists, marks events that exceed the preset management time limit as response timeouts, aggregates the average response time of each specialist on a weekly basis, and obtains the ratio of the number of missed events to the total number of violations from the negative list database as the missed rate to build an individual performance profile. The individual performance profiles are used as weighting factors to input into the scheduling optimization model. The load weight matrix is ​​constructed using the event arrival density prediction baseline and the abnormal event ratio prediction baseline for each time slot. The capability weight matrix is ​​constructed using the specialist performance scores. The KM algorithm is used to solve for the maximum weighted perfect match between specialists and time slots, and the manpower reinforcement strategy is output.

[0012] A second aspect of the present invention provides a family farm entry anomaly monitoring system based on multi-event split-screen monitoring. The system includes: a memory, a processor, and a communication interface. The memory contains a family farm entry anomaly monitoring method program based on multi-event split-screen monitoring. When the processor executes the family farm entry anomaly monitoring method program based on multi-event split-screen monitoring, it implements the steps of the family farm entry anomaly monitoring method based on multi-event split-screen monitoring as described in any of the above claims. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0014] Figure 1This is a flowchart of the first method of a family farm entry anomaly monitoring method based on multi-event split-screen monitoring, provided in an embodiment of the present invention. Figure 2 This is a flowchart of a second method for monitoring abnormal entry into a family farm based on multi-event split-screen monitoring, as provided in an embodiment of the present invention. Figure 3 This is a block diagram of a family farm entry anomaly monitoring system based on multi-event split-screen monitoring, provided in an embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0017] Figure 1 This is a flowchart of the first method of a family farm entry anomaly monitoring method based on multi-event split-screen monitoring, provided in an embodiment of the present invention. like Figure 1 As shown, the present invention provides a first method flowchart for a family farm entry anomaly monitoring method based on multi-event split-screen monitoring, including: S102: The original entry reporting information is obtained by using the intelligent reporting terminal deployed at the family farm's entry channel. The information is automatically completed in the central monitoring server based on the basic data configuration library to generate an ordered sequence of operation steps. Priority is assessed according to the urgency of the event and the farm's risk score, and finally a structured entry event queue is obtained. S104. Based on the structured entry event queue, obtain the current active events and the total number of deduplicated associated cameras, match them with the predefined split-screen specification template, synchronously calculate the comprehensive display weight of each event, allocate adjacent rectangular monitoring blocks in the split-screen grid in descending order of weight, process screen overflow with information gain strategy, and output dynamic split-screen monitoring layout. S106, obtain the camera video stream associated with each event in the structured entry event queue, perform motion detection in the region of interest to obtain the boundary trigger event sequence, use the weighted dynamic time warping algorithm and standard step template to perform time sequence alignment and calculate the comprehensive execution deviation, determine the comprehensive execution deviation event and perform split-screen visual alarm rendering; S108: Obtain the observation behavior sequence corresponding to the comprehensive execution deviation event, input it into the pre-built violation behavior feature library for summary and violation judgment, when a violation behavior is judged, extract the violation evidence video from the camera video stream, use the exponential weighted moving average model to recursively update the farm dynamic risk score, and output the violation record and the updated farm risk score. S110, based on the structured entry event queue and violation records, performs adaptive scheduling of regulatory resources oriented towards event load and abnormal distribution, generates regulatory adjustment strategies including abnormal judgment thresholds and split-screen refresh frequency, and simultaneously generates individual performance profiles of video specialists based on their response to alarm events, thereby optimizing the current regulatory layout and obtaining manpower support strategies.

[0018] Furthermore, in a preferred embodiment of the present invention, the process of obtaining original entry reporting information using an intelligent reporting terminal deployed at the family farm's entry channel, automatically completing and generating an ordered sequence of operation steps in the central monitoring server based on the basic data configuration database, prioritizing the process according to event urgency and farm risk score, and finally obtaining a structured entry event queue, specifically includes: When the smart reporting terminal deployed at the entrance of each family farm is triggered when personnel or materials enter the farm, it reads the fixed farm unique identifier and channel number and obtains the timestamp from the real-time clock module. It binds the triggered operation type code value and the pre-stored standard operating procedure identifier to form the original reporting message and transmits it to the central monitoring server. The central monitoring server extracts the farm identification code and standard operating procedure identification number from the received original reporting message as a joint query key, and initiates a step-by-step search to the basic data configuration database. The basic data configuration database stores the group-level unified operating standards, the four-level company-level supplementary operating requirements, and the service department-level detailed rules for the layout of specific farm enclosures. During the step-by-step retrieval process, the basic operation step skeleton is obtained through the corresponding process identifier number. Then, the incremental steps corresponding to the farm identifier code are merged into the basic operation step skeleton according to the rules. Then, the corresponding regional classification code and camera identifier set are read and attached to the attribute fields of each step node to generate an ordered operation step sequence. A globally unique event number is assigned to the current inbound event using a distributed sequence generator. The event number, along with the ordered work step sequence, event type, and timestamp, are encapsulated into a structured event object. The urgency coefficient constant corresponding to the event type and the recent historical risk score are read and weighted summed to obtain the priority field of the current event object. The structured event objects with assigned priorities are pushed into the binary heap queue of pending events. The heap order is arranged in descending order of priority scores, and finally the structured incoming event queue carrying complete metadata and feature vectors is output.

[0019] It should be noted that in the scenario of multi-channel entry supervision of family farms, the traditional reporting method usually only generates a simple entry record, which lacks binding with subsequent operation processes, associated cameras and risk levels, making it difficult for the central supervision system to conduct in-depth analysis and automated scheduling of entry events.

[0020] This solution deploys intelligent reporting terminals at the entry points. When personnel or materials enter and trigger a button, the terminal reads the factory-installed unique farm identifier and channel number, obtains a millisecond-level timestamp via a real-time clock module, binds the operation type code corresponding to the button with a pre-stored standard operating procedure (SOP) identifier, encapsulates it into an original reporting message, and sends it to the central monitoring server via a narrowband IoT communication module using a message queue telemetry transmission protocol. This process avoids the delays and errors caused by manual entry, ensuring the real-time nature and reliability of entry event information. The central monitoring server extracts the farm identifier and SOP identifier from the message as a joint query key, performing a step-by-step search on the three-level basic data configuration database in memory. The configuration database uses a hash table to hierarchically store the group-level unified operating standards, the fourth-level company-level supplementary operating requirements, and the service department-level detailed rules for specific farm pen layouts. During retrieval, the basic operational step skeleton is first obtained from the group-level configuration. Then, the incremental steps supplemented by the fourth-level company are inserted into the skeleton according to the merging rules. Finally, the regional classification code and camera identifier set corresponding to the channel are read from the service department-level configuration and attached to the attribute fields of each step node, automatically generating a complete and strictly ordered operational step sequence. Subsequently, a globally unique and monotonically increasing event number is assigned to the entry event using a distributed sequence generator, and this number, along with the step sequence, event type, and timestamp, is encapsulated into a structured event object. Next, the urgency coefficient constant corresponding to the event type and the farm's recent historical risk score are read, weighted, and summed to obtain a priority score, which is then written into the event priority field. The event object is pushed into a binary heap-structured queue of pending events, with the heap order arranged in descending order of priority, ensuring that high-risk, high-urgency events are retrieved and processed first. This dynamic priority scheduling mechanism ensures that when multiple entry events occur concurrently, regulatory resources always prioritize operational processes with higher biosafety risks. Finally, a structured entry event queue carrying complete metadata and feature vectors is output. This avoids the performance overhead caused by repeated queries and splicing between multiple modules, and enhances the real-time performance and accuracy of the system.

[0021] Furthermore, in a preferred embodiment of the present invention, the step of obtaining the current active events and the total number of deduplicated associated cameras based on the structured entry event queue, matching them with a predefined split-screen specification template, simultaneously calculating the comprehensive display weight for each event, allocating adjacent rectangular monitoring blocks in the split-screen grid in descending order of weight, handling screen overflow with an information gain strategy, and outputting a dynamic split-screen monitoring layout specifically includes: The active events that are currently being executed or whose planned time falls into the activation time window are selected from the structured entry event queue. The camera identifier set associated with each event is extracted and deduplicated to count the total number of video streams to be displayed. The minimum accommodating split layout is selected by comparing it with the predefined split screen specification template. If the maximum number of splits is exceeded, the fixed display set and the carousel set are divided according to the priority score. For each active event, a comprehensive display weight is calculated. The comprehensive display weight is obtained by multiplying the event type sensitivity coefficient, the complement mapping value of the farm normalized biosafety level score, and the emergency quantification value calculated by the proportion of completed steps and the planned time offset through a nonlinear function. All active events are then arranged in descending order according to the obtained weights. The split-screen view is abstracted into a two-dimensional grid matrix. A greedy maximum empty rectangle search algorithm is used to assign a rectangular monitoring block composed of adjacent cells to each active event. The cells in the block are filled with the real-time video stream of the corresponding camera in the order of the ordered operation steps, forming a spatial arrangement in the same direction as the target active event flow. If the remaining grid area cannot fully accommodate all the camera windows required for an event during the allocation process, the information gain is measured by the frequency with which each camera is associated as a source of evidence in historical violation records. The key control point images are selected and placed in the main display window, and the video streams of the remaining cameras are compressed into thumbnails and floated in a picture-in-picture format. The system integrates the video stream identifiers, grid coordinates, and thumbnail mapping relationships of each window to generate the current dynamic split-screen monitoring layout. When the structured entry event queue is added or deleted, or when the overall display weight of any event fluctuates beyond a set threshold, incremental layout recalculation is triggered to update the split-screen monitoring layout.

[0022] It should be noted that in scenarios where multiple entrances to a family farm enter concurrently, traditional fixed-screen rotation or equally divided monitoring methods prevent video specialists from simultaneously monitoring the overall picture of multiple ongoing events. Events of different risk levels occupy equal screen space, and critical violations can easily be buried in low-risk footage. To address this issue, this solution provides an adaptive split-screen layout mechanism based on event weights and camera topology constraints.

[0023] Specifically, the system first filters all active events currently in progress or scheduled to activate within their designated time window from the structured event queue. It then extracts the camera identifier set associated with each event, removes duplicates, and calculates the total number of video streams that need to be displayed simultaneously. This number is compared with a predefined split-screen specification template library, which stores commonly used specifications from single-screen to multi-segment. The system selects the minimum split layout that can accommodate the current total number of streams. When the total number of streams exceeds the maximum supported number of segments, the active events are divided into a fixed display set and a carousel set based on their assigned priority scores. High-priority events continuously occupy the window, while low-priority events rotate sequentially according to preset time slices, ensuring that limited screen resources always prioritize high-risk events.

[0024] For each active event, its comprehensive display weight is calculated, which is obtained by multiplying three factors: the event type sensitivity coefficient reflects the difference in importance of different entry types in the biosafety system, for example, the sensitivity coefficient of dead pig handling is higher than that of ordinary materials; the complement mapping value of the farm's normalized biosafety level score gives farms with higher risks a higher display weight; and the urgency metric value calculated by the proportion of completed steps and the planned time offset using a nonlinear function incorporates both work progress and time urgency into the weight calculation. After multiplying the three factors, all active events are arranged in descending order of weight. In the layout and allocation phase, the system abstracts the split screen into a two-dimensional grid matrix and uses a greedy maximum empty rectangle search algorithm to sequentially allocate rectangular monitoring blocks composed of adjacent cells to each event. Within each block, the cells are filled with the real-time video streams of the corresponding cameras from left to right and from top to bottom according to the sequence of work steps, so that the monitoring screen expands in the same direction as the work process in space. Video specialists can intuitively track the complete movement chain from decontamination to changing clothes and then to the entrance of the production area without manual switching. When the remaining grid area cannot fully accommodate all camera windows required for an event, the information gain is calculated based on the frequency with which each camera is associated as evidence in historical violation records. Key control point images are prioritized for display, while the remaining cameras are displayed as picture-in-picture thumbnails floating in the corners of the block, supporting full-resolution pop-ups on mouse hover. Finally, the video stream identifiers, grid coordinates, and thumbnail mappings of each window are integrated to generate the current dynamic split-screen monitoring layout. Incremental recalculation is triggered when events are added or removed from the event queue, or when the weight of any event fluctuates beyond a threshold, avoiding screen jitter caused by global refreshes. By comprehensively displaying weights, event type risk, farm security level, and operation progress are uniformly quantified as sorting criteria, making the allocation of monitoring resources traceable. Using historical violation frequency as an information gain metric to handle screen overflow ensures that limited windows always prioritize displaying the most regulatory-valuable key control points, effectively improving monitoring efficiency and violation detection capabilities in multi-event concurrent scenarios.

[0025] Furthermore, in a preferred embodiment of the present invention, the steps of obtaining the camera video streams associated with each event in the structured entry event queue, performing motion detection in the region of interest to obtain the boundary trigger event sequence, using a weighted dynamic time warping algorithm and a standard step template for time sequence alignment and calculating the comprehensive execution deviation, determining the comprehensive execution deviation event, and performing split-screen visual alarm rendering specifically include: In the structured entry event queue, the camera video streams associated with each event are obtained. The camera identifier set attached to each step node is parsed from the ordered operation step sequence of each event and a video stream subscription relationship is established. In each video frame, an irregular region of interest corresponding to the step area classification code is defined. The motion detection module based on Gaussian mixture model is used to analyze video frames one by one. When a foreground target enters or exits a defined irregular region of interest, a boundary trigger event containing region code, entry / exit direction mark and frame-level timestamp is generated and aggregated into a boundary trigger event sequence according to time sequence. From the ordered sequence of work steps attached to the structured entry event queue, the median of the expected time interval and the regional classification code of each step node are read in sequence. A standard step template composed of the expected regional code and the standard duration constraint is constructed. The arrangement order of the template elements is strictly consistent with the standard process of the work steps. The weighted dynamic time warping algorithm is used to calculate the optimal alignment path between the boundary trigger event sequence and the standard step template. A cumulative cost matrix is ​​constructed with the length of the boundary trigger event sequence as the number of rows and the length of the standard step template as the number of columns. The cost of each cell in the matrix is ​​composed of the weighted sum of the region coding matching cost and the time interval deviation cost. The path with the minimum cumulative cost is searched in the cumulative cost matrix as the optimal mapping relationship. When the cumulative cost of the optimal path is lower than the matching threshold of the target farm, the event segments in the boundary trigger event sequence are assigned to the corresponding steps according to the regularized path index and the actual start and end times are written back. If the matching threshold is exceeded, a step abnormality warning signal is generated. After a successful match, the deviation of each matched step is calculated based on the actual duration and the standard duration. Then, the deviations of all steps are weighted and averaged using the biosafety criticality coefficient of each step as the weight to obtain the overall execution deviation of the target event. When the overall execution deviation exceeds the preset attention threshold, the target event is determined to be an overall execution deviation event, and a visual alarm signal is sent to the split-screen rendering pipeline to change the border color scale of the monitoring block corresponding to the target event.

[0026] It's important to note that the challenge in supervising operations at family farms lies in the fact that compliance at each step cannot be determined solely by a single video clip. Instead, it requires meticulous comparison of the entry and exit sequences of personnel and materials across multiple areas against standard procedures. Traditional methods rely on video specialists manually reviewing each segment and making subjective judgments, which is not only inefficient but also makes it difficult to quantify the extent of deviations. This solution provides a complete detection chain, from motion sensing to timing alignment and automatic quantification of deviations.

[0027] Specifically, the monitoring server retrieves the camera video streams associated with each event from the structured entry event queue. First, it parses the camera identifier set attached to the ordered sequence of operational steps for each event, establishing a subscription relationship with the corresponding channel's video stream. Within each video frame, it delineates irregular regions of interest (ROIs) based on the region classification code for that step. For example, the area at the entrance of the decontamination room, the interior of the changing room, and the entrance of the disinfection channel are each designated as an independent detection area. This ensures that the focus of motion detection is precisely limited to the key locations where steps occur, avoiding interference from other irrelevant areas in the frame. Subsequently, it performs frame-by-frame analysis based on a Gaussian mixture model. This model maintains multiple Gaussian distributions for each pixel to dynamically adapt to changes in illumination and background disturbances. When the centroid of a foreground target is detected crossing the boundary from outside the ROI into the interior or moving out of the interior, a boundary trigger event is generated, containing a region code, an entry / exit direction marker, and a timestamp accurate to the frame. All such events generated by all associated cameras are aggregated sequentially into a boundary trigger event sequence, depicting the spatiotemporal transfer trajectory of the target between multiple step regions.

[0028] Simultaneously, standard step templates are extracted from the ordered sequence of work steps attached to the structured entry event queue. The median of the expected time interval and the region classification code for each step node are read sequentially to construct a template sequence composed of the expected region code and standard duration constraints. The order of template elements strictly follows the standard workflow of the work steps. For example, the standard template for personnel entry is: decontamination room, shower room, changing room, disinfection channel, and production area entrance, with each region corresponding to an expected standard stay duration. After obtaining the actual observation sequence and standard templates, a weighted dynamic time warping algorithm is used to calculate the optimal alignment path between them. The algorithm constructs a cumulative cost matrix with the length of the boundary trigger event sequence as the number of rows and the length of the standard step template as the number of columns. The cost of each cell in the matrix consists of two parts: the matching cost between the region code of the current boundary trigger event and the expected region code of the template element, and the deviation cost between the time interval of the current event and the standard duration of the template element. The two parts are weighted and summed according to the biosafety criticality coefficient of the step. For example, the time deviation cost weight of the shower step is significantly higher than that of the changing step. The path with the minimum cumulative cost is searched in the cost matrix, and this path is the optimal mapping relationship between the actual execution and the planned process.

[0029] When the cumulative cost of the optimal path is lower than the matching threshold dynamically determined by the target farm based on the standard deviation of historical operation fluctuations, the event segments in the boundary trigger event sequence are assigned to the corresponding steps according to the regularized path index, and the actual start and end times are written back. If the threshold is exceeded, a warning signal for missing steps or disordered sequence is generated. After successful matching, the ratio of the actual duration to the standard duration of each step is used to calculate the step deviation using an asymmetric deviation function with logarithmic transformation. This function applies different penalty curvatures for excessively long and short durations. The deviations of each step are then weighted and averaged using a biosafety criticality coefficient to obtain the comprehensive execution deviation of the target event. When this deviation exceeds the preset attention threshold, the system determines it as a comprehensive execution deviation event and sends a visual alarm signal to the split-screen rendering pipeline, with the border color of the event monitoring block gradually changing from green to red, automatically highlighting the abnormal event in the split-screen monitoring layout. This achieves accurate perception of the entry and exit of operation steps, avoids false triggering of fixed area detection in complex scenarios, and reduces the cognitive burden on video specialists and the delay in anomaly detection.

[0030] Furthermore, in a preferred embodiment of the present invention, the step of obtaining the observed behavior sequence corresponding to the comprehensive execution deviation event, inputting it into a pre-built violation behavior feature database for violation determination, and when a violation is determined to exist, extracting violation evidence video from the camera video stream, recursively updating the farm dynamic risk score using an exponentially weighted moving average model, and outputting the violation record and the updated farm risk score, specifically includes: Extract the corresponding converged boundary trigger event sequence from the event records of the comprehensive execution deviation event as the observation behavior sequence, and extract all relevant spatiotemporal state machine instances from the violation behavior feature library according to the event type code; In the violation behavior feature library, each violation rule is a state machine consisting of several state nodes and directed transition edges with time constraints. The state nodes represent the occurrence of target appearance or target disappearance events in the region of interest corresponding to the specific region encoding, and the directed transition edges specify the order of transition between states and the maximum allowed time interval threshold. The observed behavior sequence is used as the input event stream. Boundary trigger events are read in order of timestamp. Each time an event is read, all active state machines are traversed to determine whether the current event meets the triggering condition and to perform state transition. When any state machine transitions to the accepted state marked as a violation along the directed edge, a violation judgment is triggered and the corresponding violation type identifier is recorded. After a violation is detected, the video data segment of a preset duration is extracted based on the boundary trigger event timestamp corresponding to the predecessor state node associated with the acceptance state, which serves as the starting point for evidence extraction. This generates a structured violation record containing a violation type identifier, a violation occurrence timestamp, a farm identifier, an event number, and a violation evidence video, and writes it into the negative list database. The current risk score of the target farm is obtained and multiplied by a preset attenuation factor to obtain the attenuated historical component. The preset fixed contribution coefficient corresponding to the newly occurred violation is multiplied by the severity weight of the violation to obtain the current increment. This increment is added to the attenuated historical component to generate the updated dynamic risk score of the farm.

[0031] It should be noted that after identifying events with abnormal execution through the comprehensive deviation index, it is necessary to further determine whether the abnormality constitutes a clear violation, and update the farm's dynamic risk file accordingly. Simply relying on the deviation value can only reflect time-consuming anomalies and cannot identify violation types with specific behavioral patterns, such as missing steps or crossing regions.

[0032] Specifically, firstly, the previously converged boundary trigger event sequences are extracted from the event records of the comprehensive execution deviation events as observation behavior sequences. Simultaneously, based on the event type encoding of the event, all spatiotemporal state machine instances related to that type are retrieved and loaded from the violation behavior feature library. In the violation behavior feature library, each violation rule is modeled as a finite state machine consisting of several state nodes and directed transition edges with time constraints. State nodes represent events where a target is detected or disappears within the region of interest corresponding to a specific region encoding. Directed transition edges specify the order of transitions from one state to another and the maximum allowed time interval threshold for the transition. Taking the violation rule of a person entering the changing area without showering as an example, its state machine includes an initial state, a shower area appearance state, and a changing area appearance state. If the observation sequence directly triggers the changing area appearance state and remains there without passing through the shower area appearance state, the state machine enters the acceptance state along the preset violation edge, and the violation is determined to be established. The recognition engine takes the observed behavior sequence as the input event stream, reads the boundary trigger events one by one in the order of timestamps, and traverses all loaded active state machines each time it reads them. It determines whether the trigger condition of a certain outgoing edge is met based on the region code and timestamp carried by the current event. If it is met, it performs a state transition and records the transition timestamp. When any state machine transitions along a directed edge to an accepted state marked as a violation, it immediately triggers a violation judgment and records the violation type identifier corresponding to the accepted state.

[0033] After a violation is detected, the start time of the violation is determined by using the boundary trigger event timestamp corresponding to the predecessor state node associated with the acceptance state. A video data segment with a preset delay from that starting point to the current time is extracted from the circular video buffer of the corresponding camera. This segment is then compressed and encoded to generate a violation evidence video with a time watermark. Subsequently, a structured violation record is generated, containing a violation type identifier, violation occurrence timestamp, farm identifier, event number, and violation evidence video storage index. This record is written to the negative list database. In the risk score update phase, the farm's current risk score is read as the score value of the previous period. This score is multiplied by a preset weekly decay factor to obtain the decayed historical component, allowing the influence of historical violation records on the current score to gradually diminish over time. Simultaneously, the preset fixed contribution coefficient corresponding to the newly occurred violation is multiplied by the severity weight of the current violation to obtain the current increment. The decayed historical component is added to the current increment to obtain the updated dynamic risk score of the farm, completing the full closed loop from violation detection to risk quantification.

[0034] Figure 2 This is a flowchart of a second method for monitoring abnormal entry into a family farm based on multi-event split-screen monitoring, as provided in an embodiment of the present invention. like Figure 2 As shown, the present invention provides a second method flowchart for a family farm entry anomaly monitoring method based on multi-event split-screen monitoring, including: S202 divides the system runtime into continuous time slices with a fixed duration, extracts the actual number of events and the number of violations in each time slice from the structured arrival event queue and violation records, traces back the historical sequence of the same time slice in several past periods, and uses the exponential smoothing algorithm to generate the event arrival density prediction baseline and the abnormal event ratio prediction baseline. S204 counts the actual number of events arriving in real time within the current time slice. When the actual number of events arriving exceeds the product of the event arrival density prediction baseline and the preset expansion coefficient, it is determined that the event is in a high load state. A first type of adjustment instruction is generated, which includes lowering the comprehensive execution deviation alarm threshold and increasing the refresh frame rate of the split-screen monitoring layout, and then sent out for execution. S206 calculates the proportion of the number of actual triggered comprehensive execution deviation warning events to the total number of actual arrival events within the current time slice. When the proportion exceeds the product of the abnormal event ratio prediction baseline and the preset amplification factor, it is determined to be in an abnormal high-occurrence state, and generates a second type of adjustment instruction to lock the high-abnormal farm video window and prevent it from being rotated and push reinforcement requests. The S208 video collection specialist collects the response time difference for each alarm event, marks events that exceed the preset management time limit as response timeouts, aggregates the average response time of each specialist on a weekly basis, and obtains the ratio of the number of missed events to the total number of violations from the negative list database as the missed rate to build an individual performance profile. S210 inputs the individual performance profile as a weighting factor into the scheduling optimization model, constructs a load weight matrix with the event arrival density prediction baseline and the abnormal event ratio prediction baseline for each time slot, constructs a capability weight matrix with the specialist performance score, and uses the KM algorithm to solve for the maximum weight perfect match between specialists and time slots, and outputs a manpower reinforcement strategy.

[0035] It should be noted that in actual regulatory operations, events occurring at family farms are not evenly distributed. Different types of events, such as personnel disinfection, material transfer, and dead pig disposal, exhibit significant peak and trough variations throughout the day. Fixed monitoring strategies and rigid staffing schedules are ill-suited to this dynamically changing load. When the number of concurrent events surges during peak periods, video specialists cannot meticulously verify each event, increasing the risk of missed anomalies. Conversely, during periods of concentrated anomaly outbreaks, failure to promptly identify risk windows and deploy additional personnel can lead to a loss of regulatory control. This solution designs a closed-loop mechanism for adaptive scheduling of regulatory resources based on dual predictions of event load and anomaly distribution.

[0036] Specifically, the entire day's operation time is first divided into continuous time slices of fixed duration, for example, with a 15-minute granularity. The actual number of arriving events and the number of confirmed violations within each time slice are extracted from the structured arrival event queue and negative list database, respectively. For each time slice, the historical event quantity sequence and historical violation quantity sequence of the same time slice over several past periods are traced back. An exponential smoothing algorithm with trend adjustment is applied to give higher weight to recent data and capture the linear trend of the sequence, generating the event arrival density prediction baseline and abnormal event ratio prediction baseline for the current time slice. After entering the current time slice, the actual number of arriving events is continuously counted. Once this value exceeds the product of the prediction baseline and the preset inflation coefficient, it is determined that the current event is in a high-load state, and the first type of adjustment instruction is automatically issued. The alarm threshold for comprehensive execution deviation is appropriately lowered from the default level to expand the initial screening range, while the refresh frame rate of the split-screen monitoring layout is increased to reduce screen latency. At the same time, the real-time ratio of the number of comprehensive execution deviation warning events actually triggered within the time slice to the total number of actual arrival events is simultaneously counted. When this ratio exceeds the product of the abnormal ratio prediction baseline and the preset amplification factor, it is determined that the current state is in a state of high abnormality. A second type of adjustment instruction is generated, locking all video windows associated with the high-abnormal farms in the split screen and preventing them from being rotated. A reinforcement request containing a list of farm identifiers and current abnormal statistics is sent to the duty dispatch terminal through the message push channel.

[0037] At the manpower scheduling level, the system continuously collects the response time difference from generation to confirmation for each alarm event by each video specialist. Events exceeding the management time limit are marked as response timeouts. The average response time of each specialist is aggregated weekly. Simultaneously, the system retrieves the number of missed events that were not identified in real time but were confirmed after review during the specialist's assigned time period from the negative list database, calculates the missed event rate, and inputs the average response time and missed event rate into a rank-based evaluation model to generate individual performance profiles. The scheduling optimization model constructs a load weight matrix based on the event arrival density prediction baseline and the anomaly ratio prediction baseline for each time slot, and a capability weight matrix based on specialist performance scores. The matching of specialists and time slots is abstracted as a bipartite graph, and the Kuhn-Munkres algorithm is used to solve for the maximum weight perfect match, so that high-performing specialists are preferentially assigned to time slots predicted to be high-load or high-anomaly. Finally, the optimal specialist scheduling scheme that balances load balancing and individual capabilities is output. This solves the dilemma of insufficient manpower during peak periods and avoids idle manpower during off-peak periods, improving both resource utilization efficiency and the timeliness of anomaly detection in the entire monitoring system.

[0038] Figure 3 An embodiment of the present invention provides a family farm entry anomaly monitoring system 3 based on multi-event split-screen monitoring. The system includes: a memory 301, a processor 302, and a communication interface 303. The memory 301 contains a family farm entry anomaly monitoring method program based on multi-event split-screen monitoring. When the family farm entry anomaly monitoring method program based on multi-event split-screen monitoring is executed by the processor 302, it implements the steps of the family farm entry anomaly monitoring method based on multi-event split-screen monitoring as described above.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring abnormal entry into family farms based on multi-event split-screen monitoring, characterized in that, include: The original entry reporting information is obtained by using the intelligent reporting terminal deployed at the family farm's entry channel. The information is then automatically completed in the central monitoring server based on the basic data configuration library to generate an ordered sequence of operation steps. Priority is assessed according to the urgency of the event and the farm's risk score, and finally a structured entry event queue is obtained. Based on the structured entry event queue, the current active events and the total number of deduplicated associated cameras are obtained, matched with the predefined split-screen specification template, and the comprehensive display weight of each event is calculated synchronously. Adjacent rectangular monitoring blocks are allocated in the split-screen grid in descending order of weight, and the screen overflow is handled with an information gain strategy to output a dynamic split-screen monitoring layout. Obtain the camera video streams associated with each event in the structured entry event queue, parse the camera identifier set attached to each step node from the ordered operation step sequence of each event and establish a video stream subscription relationship, and delineate the irregular region of interest corresponding to the step region classification code in each video frame; The motion detection module based on Gaussian mixture model is used to analyze video frames one by one. When a foreground target enters or exits a defined irregular region of interest, a boundary trigger event containing region code, entry / exit direction mark and frame-level timestamp is generated and aggregated into a boundary trigger event sequence according to time sequence. From the ordered sequence of work steps attached to the structured entry event queue, the median of the expected time interval and the regional classification code of each step node are read in sequence. A standard step template composed of the expected regional code and the standard duration constraint is constructed. The arrangement order of the template elements is strictly consistent with the standard process of the work steps. The weighted dynamic time warping algorithm is used to calculate the optimal alignment path between the boundary trigger event sequence and the standard step template. A cumulative cost matrix is ​​constructed with the length of the boundary trigger event sequence as the number of rows and the length of the standard step template as the number of columns. The cost of each cell in the matrix is ​​composed of the weighted sum of the region coding matching cost and the time interval deviation cost. The path with the minimum cumulative cost is searched in the cumulative cost matrix as the optimal mapping relationship. When the cumulative cost of the optimal path is lower than the matching threshold of the target farm, the event segments in the boundary trigger event sequence are assigned to the corresponding steps according to the regularized path index and the actual start and end times are written back. If the matching threshold is exceeded, a step abnormality warning signal is generated. After a successful match, the deviation of each matched step is calculated based on the actual duration and the standard duration. Then, the deviations of all steps are weighted and averaged using the biosafety criticality coefficient of each step as the weight to obtain the overall execution deviation of the target event. When the overall execution deviation exceeds the preset attention threshold, the target event is determined to be an overall execution deviation event, and a visual alarm signal is sent to the split-screen rendering pipeline to change the border color scale of the monitoring block corresponding to the target event. The system obtains the observed behavior sequence corresponding to the comprehensive execution deviation event, inputs it into the pre-built violation behavior feature library for violation judgment, and when a violation is determined, it extracts the violation evidence video from the camera video stream, uses the exponential weighted moving average model to recursively update the farm risk score, and outputs the violation record and the updated farm risk score. Based on the structured entry event queue and violation records, adaptive scheduling of regulatory resources is carried out for event load and abnormal distribution. Regulatory adjustment strategies including abnormal judgment thresholds and split-screen refresh frequency are generated. Simultaneously, individual performance profiles are generated by video specialists based on their response to alarm events, and manpower support strategies are obtained to optimize the current regulatory schedule.

2. The method for monitoring abnormal entry into a family farm based on multi-event split-screen monitoring according to claim 1, characterized in that, The process involves using intelligent reporting terminals deployed at the family farm's entry point to obtain original entry reporting information. This information is then automatically completed in the central monitoring server based on a basic data configuration database to generate an ordered sequence of operational steps. Priority is assessed according to event urgency and farm risk score, ultimately resulting in a structured entry event queue, which specifically includes: When the smart reporting terminal deployed at the entrance of each family farm is triggered when personnel or materials enter the farm, it reads the fixed farm unique identifier and channel number and obtains the timestamp from the real-time clock module. It binds the triggered operation type code value and the pre-stored standard operating procedure identifier to form the original reporting message and transmits it to the central monitoring server. The central monitoring server extracts the farm identification code and standard operating procedure identification number from the received original reporting message as a joint query key, and initiates a step-by-step search to the basic data configuration database. The basic data configuration database stores the group-level unified operating standards, the four-level company-level supplementary operating requirements, and the service department-level detailed rules for the layout of specific farm enclosures. During the step-by-step retrieval process, the basic operation step skeleton is obtained through the corresponding process identifier number. Then, the incremental steps corresponding to the farm identifier code are merged into the basic operation step skeleton according to the rules. Then, the corresponding regional classification code and camera identifier set are read and attached to the attribute fields of each step node to generate an ordered operation step sequence. A globally unique event number is assigned to the current inbound event using a distributed sequence generator. The event number, along with the ordered job step sequence, event type, and timestamp, are encapsulated into a structured event object. The urgency coefficient constant corresponding to the event type and the recent historical farm risk score are read and weighted summed to obtain the priority field of the current event object. The structured event objects with assigned priorities are pushed into the binary heap queue of pending events. The heap order is arranged in descending order of priority scores, and finally the structured incoming event queue carrying complete metadata and feature vectors is output.

3. The method for monitoring abnormal entry into a family farm based on multi-event split-screen monitoring as described in claim 1, characterized in that, The process involves obtaining the current active events and the total number of deduplicated associated cameras based on the structured entry event queue, matching them with a predefined split-screen specification template, simultaneously calculating the comprehensive display weight for each event, allocating adjacent rectangular monitoring blocks in the split-screen grid in descending order of weight, handling screen overflow using an information gain strategy, and outputting a dynamic split-screen monitoring layout. Specifically, this includes: The active events that are currently being executed or whose planned time falls into the activation time window are selected from the structured entry event queue. The camera identifier set associated with each event is extracted and deduplicated to count the total number of video streams to be displayed. The minimum accommodating split layout is selected by comparing it with the predefined split screen specification template. If the maximum number of splits is exceeded, the fixed display set and the carousel set are divided according to the priority score. For each active event, a comprehensive display weight is calculated. The comprehensive display weight is obtained by multiplying the event type sensitivity coefficient, the complement mapping value of the farm normalized biosafety level score, and the emergency quantification value calculated by the proportion of completed steps and the planned time offset through a nonlinear function. All active events are then arranged in descending order according to the obtained weights. The split-screen view is abstracted into a two-dimensional grid matrix. A greedy maximum empty rectangle search algorithm is used to assign a rectangular monitoring block composed of adjacent cells to each active event. The cells in the block are filled with the real-time video stream of the corresponding camera in the order of the ordered operation steps, forming a spatial arrangement in the same direction as the target active event flow. If the remaining grid area cannot fully accommodate all the camera windows required for an event during the allocation process, the information gain is measured by the frequency with which each camera is associated as a source of evidence in historical violation records. The key control point images are selected and placed in the main display window, and the video streams of the remaining cameras are compressed into thumbnails and floated in a picture-in-picture format. The system integrates the video stream identifiers, grid coordinates, and thumbnail mapping relationships of each window to generate the current dynamic split-screen monitoring layout. When the structured entry event queue is added or deleted, or when the overall display weight of any event fluctuates beyond a set threshold, incremental layout recalculation is triggered to update the split-screen monitoring layout.

4. The method for monitoring abnormal entry into a family farm based on multi-event split-screen monitoring according to claim 1, characterized in that, The process involves acquiring the observed behavior sequence corresponding to the comprehensive execution deviation event, inputting it into a pre-built violation behavior feature database for violation determination, and when a violation is determined, extracting violation evidence video from the camera video stream, recursively updating the farm risk score using an exponentially weighted moving average model, and outputting the violation record and the updated farm risk score. Specifically, this includes: Extract the corresponding converged boundary trigger event sequence from the event records of the comprehensive execution deviation event as the observation behavior sequence, and extract all relevant spatiotemporal state machine instances from the violation behavior feature library according to the event type code; In the violation behavior feature library, each violation rule is a state machine consisting of several state nodes and directed transition edges with time constraints. The state nodes represent the occurrence of target appearance or target disappearance events in the region of interest corresponding to the specific region encoding, and the directed transition edges specify the order of transition between states and the maximum allowed time interval threshold. The observed behavior sequence is used as the input event stream. Boundary trigger events are read in order of timestamp. Each time an event is read, all active state machines are traversed to determine whether the current event meets the triggering condition and to perform state transition. When any state machine transitions to the accepted state marked as a violation along the directed edge, a violation judgment is triggered and the corresponding violation type identifier is recorded. After a violation is detected, the video data segment of a preset duration is extracted based on the boundary trigger event timestamp corresponding to the predecessor state node associated with the acceptance state, which serves as the starting point for evidence extraction. This generates a structured violation record containing a violation type identifier, a violation occurrence timestamp, a farm identifier, an event number, and a violation evidence video, and writes it into the negative list database. The current risk score of the target farm is obtained and multiplied by a preset attenuation factor to obtain the attenuated historical component. The preset fixed contribution coefficient corresponding to the newly occurred violation is multiplied by the severity weight of the violation to obtain the current increment. This increment is then added to the attenuated historical component to generate the updated farm risk score.

5. A method for monitoring abnormal entry into family farms based on multi-event split-screen monitoring as described in claim 1, characterized in that, The adaptive scheduling of regulatory resources based on structured entry event queues and violation records, oriented towards event load and anomaly distribution, generates regulatory adjustment strategies including anomaly judgment thresholds and split-screen refresh frequencies. Simultaneously, individual performance profiles are generated based on video specialists' response to alarm events to optimize the current regulatory schedule and obtain manpower reinforcement strategies. Specifically, this includes: The system runtime is divided into continuous time slices with a fixed duration. The actual number of events and the number of violations in each time slice are extracted from the structured entry event queue and violation records, respectively. The historical sequence of the same time slice in several past periods is traced back, and the event arrival density prediction baseline and the abnormal event ratio prediction baseline are generated using the exponential smoothing algorithm. The number of actual arrival events is counted in real time within the current time slice. When the number of actual arrival events exceeds the product of the event arrival density prediction baseline and the preset expansion coefficient, it is determined that the event is in a high load state. A first type of adjustment instruction is generated, which includes lowering the comprehensive execution deviation alarm threshold and increasing the refresh frame rate of the split-screen monitoring layout, and then issued for execution. Within the current time slice, the proportion of the number of actual triggered comprehensive execution deviation warning events to the total number of actual arrival events is counted. When the proportion exceeds the product of the abnormal event ratio prediction baseline and the preset amplification factor, it is determined to be in an abnormal high-occurrence state, and a second type of adjustment instruction is generated to lock the high-abnormal farm video window and prevent it from being rotated and push reinforcement requests. The system collects the response time difference of each alarm event by video specialists, marks events that exceed the preset management time limit as response timeouts, aggregates the average response time of each specialist on a weekly basis, and obtains the ratio of the number of missed events to the total number of violations from the negative list database as the missed rate to build an individual performance profile. The individual performance profiles are used as weighting factors to input into the scheduling optimization model. The load weight matrix is ​​constructed using the event arrival density prediction baseline and the abnormal event ratio prediction baseline for each time slot. The capability weight matrix is ​​constructed using the specialist performance scores. The KM algorithm is used to solve for the maximum weighted perfect match between specialists and time slots, and the manpower reinforcement strategy is output.

6. A family farm entry anomaly monitoring system based on multi-event split-screen monitoring, characterized in that, The system includes: a memory, a processor, and a communication interface. The memory contains a program for monitoring abnormal entry of family farms based on multi-event split-screen monitoring. When the program for monitoring abnormal entry of family farms based on multi-event split-screen monitoring is executed by the processor, it implements the steps of the method for monitoring abnormal entry of family farms based on multi-event split-screen monitoring as described in any one of claims 1-5.

Images