A method for generating public health video monitoring results

Abstract

The application relates to the technical field of image recognition, and discloses a public health video supervision result generation method. After public health video supervision data is acquired, public supervision time axis processing and action end position conversion are performed, regional contact candidate results and unified supervision observation results are generated, the regional contact candidate results are converted into time sequence contact segments, a health state inheritance graph containing sub-region coverage candidate relationships is constructed, the current sub-region coverage candidate relationship is taken as a backtracking entry, disposal write permission verification is performed on an upstream state chain of the same inheritable health state object, and finally, the sub-region supervision state records of the public contact surface sub-regions are updated according to the disposal write permission results, and public health video supervision results carrying upstream state chain identifiers are generated. The method can avoid that the preceding sequence risks that are not cut off by the reset confirmation relationship are mistakenly written into the qualified closed state.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a method for generating public health video surveillance results. Background Technology

[0002] In public health facilities, video surveillance is increasingly being used for daily hygiene management in areas such as hospital waiting areas, school cafeterias, public restrooms, shopping mall mother and baby rooms, and pharmacy dispensing windows. These scenarios typically involve continuous changes in pedestrian flow, repeated use of public contact surfaces, and the overlapping of cleaning operations and pedestrian traffic. For example, the armrests of waiting benches may be touched by different people in a short period of time; one side of a cafeteria table may be tidying up plates while someone is still picking up or putting down items on the other side; and cleaning tools in public restrooms may move continuously between the floor, the edge of the sink, and the cleaning storage area. Existing video surveillance systems typically collect image sequences through cameras and use methods such as target detection, human posture recognition, hand contact recognition, cleaning action recognition, and trajectory coverage analysis to determine whether there are wiping, mopping, spraying, or tidying actions in the footage. This helps generate monitoring results indicating whether public contact surfaces are clean, pending treatment, or require review.

[0003] In practical applications, the generation of public health video surveillance results is also affected by the previous contact status of cleaning tools, gloves, or hands. Existing video recognition solutions can usually identify whether the cleaning trajectory covers local areas such as desktops, handrails, buttons, screens, or the edges of handwashing stations, and can also update the local status based on subsequent personnel contact. However, when generating public health surveillance results, it is still necessary to further address the issue of status inheritance caused by the continuous movement of objects performing cleaning actions between different areas. When wiping cloths, mops, clamps, gloves, or hands first contact risk source areas such as the edge of trash cans, sewage floors, or recycling areas, and then cover new public contact areas, generating a qualified closure result based solely on local cleaning trajectories may cause the previous risk status to transfer to the new monitored object with the cleaning action. If the video fails to confirm evidence of status reset such as replacement, cleaning, disinfection, disposal, or handwashing, there may be a discrepancy between the closure result of the public contact area and the on-site hygiene status, which may affect the accuracy of subsequent inspection scheduling, disinfection verification, and accountability. Therefore, it is necessary to establish a hygiene status inheritance judgment for objects performing cleaning actions and use it as a prerequisite control condition for the closure of the monitoring status of public contact areas. Summary of the Invention

[0004] This application proposes a method for generating public health video surveillance results to address the problems raised in the background art.

[0005] To achieve the above objectives, this application adopts the following technical solution: a method for generating public health video surveillance results, comprising the following steps:

[0006] S1. Obtain public health video surveillance data, adopt unified surveillance observation processing, perform public surveillance time axis processing on continuous video data in public health video surveillance data, convert the action end of inheritable health status objects into action end position records under scene surveillance coordinate system, and then generate regional contact candidate results and unified surveillance observation results from the action end position records.

[0007] S2 performs state inheritance fragmentation processing on the unified regulatory observation results, transforms the regional contact candidate results of inheritable health status objects into temporal contact fragments, and constructs a health status inheritance graph containing sub-region coverage candidate relationships according to the inheritance path between temporal contact fragments.

[0008] S3, using the current sub-area coverage candidate relationship as the backtracking entry point, performs upstream state chain permission verification on the sanitary state inheritance graph, preventing risk contact segments that have not been cut off by the reset confirmation relationship from being written into the qualified closed state of clean coverage, and generating disposal write permission results;

[0009] S4. Perform sub-regional regulatory status evidence chain update processing on the disposal written into the permit result, update the sub-regional coverage candidate relationship that has passed the permit verification to a qualified closed state and write it into the permit basis record, and generate a public health video supervision result carrying the upstream status chain identifier.

[0010] The beneficial effects of this invention are as follows:

[0011] 1. This invention solves the problem in existing video surveillance where public contact area sub-areas are easily mistakenly written as qualified closed states based solely on partial cleaning coverage. It achieves correlation verification between cleaning closure results and the previous health status of the treated object, and makes subsequent public health video surveillance results more binding on the evidence chain. This is achieved by first unifying public health video surveillance data, then constructing a health status inheritance graph, and finally backtracking and verifying the processing order of writing permission based on the current sub-area coverage candidate relationship.

[0012] 2. This invention organizes risk contact segments, reset confirmation segments, blind zone acceptance segments, and sub-area coverage candidate segments into the same hygiene state inheritance diagram, and uses reset confirmation relationships and blind zone uncertainties to control whether the risk inheritance relationship continues to enter the permission judgment. This solves the problem of neglecting the preceding risks and misattributing the risk sources when cleaning tools, gloves, or hands continuously circulate in multiple areas. It prevents risk contacts that have not been effectively reset and cut off from writing qualified closures, and also provides clear upstream source evidence for the generation of pending closure states and risk transfer states.

[0013] 3. This invention addresses the problem of low-reliability contacts, uncertain blind spot acceptance, or abnormal positioning being directly written back to the sub-area status in online supervision by using the disposal written into the permission result as a prerequisite for updating the status of public contact surface sub-areas. It limits the permission status, non-permission type, and upstream status chain identifier to the corresponding public contact surface sub-area for evidence chain update. This enables differentiated updates for qualified closure, pending review closure, risk transfer, and insufficient coverage, and also makes the public health video supervision results more reliable in subsequent review and status tracing. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort:

[0015] Figure 1 This is a flowchart of the method of the present invention;

[0016] Figure 2 This is the S3 flowchart of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] like Figure 1 and Figure 2 As shown, this invention discloses a method for generating public health video surveillance results, including the following steps:

[0019] S1 uses public health video surveillance data as its processing object. It reads continuous video data, camera calibration parameters, and scene area configuration records of the target public health scene. Continuous video data serves as the video observation input, camera calibration parameters serve as the coordinate conversion basis, and scene area configuration records serve as the configuration basis for the boundaries of the surveillance area and the public contact surface sub-area. Image recognition processing is implemented using conventional methods in the field. Its input is continuous video data, and the processing action is to identify inheritable health status objects and locate the action end of the inheritable health status objects. The output results are the observation records of inheritable health status objects and the position records of the action end. When the detection confidence is lower than the minimum available confidence, the corresponding video frame only forms an observation segment to be confirmed and does not enter the generation of regional contact candidate results. The minimum available confidence is a dimensionless result between 0 and 1, with a value range of 0.5 to 0.8, determined by the historical review records of the same scene, the image recognition processing verification records, and the scene area configuration records. When the historical review records are insufficient, the initial available confidence in the image recognition processing verification records is used.

[0020] Using continuous video data as the processing object, the acquisition time, acquisition order, and camera frame rate of the video frames in the continuous video data are read, and common monitoring timeline processing is performed. For video frames acquired by the same camera, the time deviation is expressed in frames and the corresponding actual time is recorded synchronously, controlled within 1 to 3 frames. For multi-camera joint coverage scenarios, the time deviation is expressed in seconds, controlled within 0.1 to 0.5 seconds. For multi-camera time deviation in scenarios involving short-term hand contact, the deviation is controlled within 0.1 to 0.2 seconds. For cleaning operation scenarios involving mops, wiping cloths, and spray bottles, the deviation is controlled within 0.2 to 0 seconds. 0.5 seconds. The aforementioned value is determined based on the camera frame rate, device time synchronization record, network transmission delay, and historical review record. When the historical review record is insufficient, the initial time deviation range is determined based on the camera frame rate and device time synchronization record, and corrected after the review record is accumulated. Video segments with time deviations exceeding the corresponding range are not included in the merging process of the same temporal contact segment, but are retained as observation segments to be confirmed. The public monitoring time axis is output to S2 to determine whether the regional contact candidate results can be merged into temporal contact segments, and output to S3 to limit the backtracking order of the upstream state chain.

[0021] After establishing a public monitoring timeline, the processing object is the record of the action point position of the inheritable hygiene status object. Camera calibration parameters and scene area configuration records are read, and position conversion in the scene monitoring coordinate system is performed. For near-planar public contact surfaces such as desktops, window sills, and screen surfaces, the action point position in the video image coordinates is converted to the action point position record in the scene monitoring coordinate system based on camera calibration parameters and calibration control points. For non-planar public contact surfaces such as curved handrails, door handles, and sink edges, the corresponding public contact surface sub-area is used as the processing object. Its semantic segmentation contour, calibration boundary, and at least three calibration control points arranged along the main extension direction of the contact surface are read. When the public contact surface has a tubular, arc-shaped, or strip-shaped extension, the centerline direction of the semantic segmentation contour or the axis projection direction marked in the scene area configuration record is determined as the first direction of the local contact surface coordinates, and the circumferential or width direction after unfolding along the surface normal is determined as the second direction. When the centerline direction cannot be automatically determined, the manually calibrated main extension direction is read as the unfolding direction. The edge end projects the 3D or 2D position of each calibrated control point in the scene monitoring coordinate system to the local contact surface coordinates, and then converts it back to the scene monitoring coordinate system according to the same expansion relationship. The maximum deviation of the control point position after the round trip conversion is used as the mapping error. When the mapping error is not higher than half of the contact confirmation distance range, the position results under the local contact surface coordinates are used for judgment when the regional contact candidate results, effective coverage quality, and sub-area monitoring status evidence chain update involve the public contact surface sub-area. When the correspondence between the local contact surface coordinates and the scene monitoring coordinate system cannot be established, or the mapping error of the correspondence is higher than half of the contact confirmation distance range, the corresponding public contact surface sub-area will not be included in the permission basis for the disposal and written into the permission result, and will be transferred to the generation path of the candidate result to be reviewed in subsequent processing.

[0022] After unifying the time and coordinate standards, a regulatory area base map is generated, using scene area configuration records as the processing object. This base map defines the technical roles that inheritable hygiene status objects assume when entering different regulatory areas. The risk source area represents the region where inheritable hygiene status objects can inherit prior risks; its boundaries are determined by hygiene management procedures, historical review records, calibration records, and abnormal contaminant identification results. For risk source areas formed by abnormal contaminants, semantic segmentation or texture anomaly detection is first performed on suspected contaminant areas in continuous video data at the edge. When there are labeled samples for categories such as liquid residue, food scraps, garbage edges, and sewage floors in the scene configuration, an abnormal contaminant mask is output using a semantic segmentation model. When there are no labeled samples, candidate contours for abnormal contaminants are formed using color deviation, local texture mutation, and background modeling difference, and these candidate contours are then submitted to historical review records or manual review for confirmation. The abnormal contaminant contour recognition error is formed by the maximum distance between the contour boundary and the review boundary or the upper bound of the calibration error. The buffer distance is measured in meters, and its range for online license verification is 0.02 meters to 0.5 meters. Within this range, a value no less than the abnormal dirt contour recognition error is preferred. When the abnormal dirt contour recognition error is greater than 0.5 meters, the abnormal dirt contour is not directly included in the risk source area for online license verification. Instead, the corresponding area is transferred to the unconfirmed risk source area, and the boundary of the risk source area is determined based on manual review records or recalibration results. For high-frequency contact surfaces with clear boundaries, such as buttons, screens, and window sills, the buffer distance is between 0.02 meters and 0.1 meters. For public contact surfaces with continuous contact boundaries, such as desktops, sink edges, and handrails, the buffer distance is between 0.05 meters and 0.2 meters. For areas where liquid diffusion or residue transfer occurs, such as sewage floors, garbage can edges, and dish recycling areas, the buffer distance is between 0.2 meters and 0.5 meters. The specific values ​​are determined based on the abnormal dirt outline recognition error, the scene hygiene management procedure, and historical review records; when historical review records are insufficient, the scene hygiene management procedure and the abnormal dirt recognition error are used as the initial value basis.

[0023] The reset confirmation area is used to represent the area that can support the previous risk cut-off judgment. Its boundary is formed by the location of the cleaning pool, the location of the disinfectant placement point, the location of the tool disposal point, the location of the glove replacement point, and the location of the handwashing area. The blind zone is used to represent the area where the video cannot continuously confirm the acceptance relationship and action status of the object being handled. Its boundary is formed by the camera obstruction boundary, the pillar obstruction area, the passage behind the door, and the area of ​​visual field breakage. The public contact surface area is used to represent the physical surface of the entity that needs to output the sub-area supervision status. After the supervision area base map is output to S2, the risk source area is used to form risk contact segments, the reset confirmation area is used to form reset confirmation segments, the blind zone is used to form the acceptance relationship to be confirmed, and the public contact surface sub-area is used to form the sub-area coverage candidate relationship.

[0024] Taking the public contact surface area as the treatment object, a public contact surface sub-area is formed based on the shape of the public contact surface, the accessible boundary, the historical contact hot zone, and the minimum effective coverage width of the action end. The unit of the minimum effective coverage width of the action end is meters, which is determined by the actual width of the action end of the cleaning tool, hygiene management procedures, effective coverage samples in historical review records, and calibration samples in the same scene. The value range is 0.03 meters to 0.25 meters. The minimum effective coverage width of the action end for hands and gloves is 0.03 meters to 0.08 meters, the minimum effective coverage width of the action end for wiping cloth is 0.08 meters to 0.15 meters, the minimum effective coverage width of the action end for mop is 0.15 meters to 0.25 meters, and the spray bottle uses the candidate area of ​​the spray landing point as the coverage range of the action end, and does not use the width of the spray bottle body as the minimum effective coverage width of the action end. The candidate spray landing point region is formed by the nozzle position, nozzle orientation, spraying action recognition results, coordinates of the plane or local contact surface where the common contact surface sub-area is located, and historical spray landing point statistics. At the edge, the spray bottle posture and nozzle tip position are identified in consecutive video frames. The spray center point is determined based on the intersection of the directional ray pointing from the nozzle tip to the common contact surface sub-area and the unfolded surface of the plane or local contact surface where the common contact surface is located. Then, an elliptical or fan-shaped candidate spray landing point region is formed based on the spraying distance, nozzle opening angle, or spray diffusion radius in historical verification records. The overlap ratio between the candidate spray landing point region and the effective contact area is used for subsequent effective coverage quality calculations. When the nozzle orientation, spray center point, or spray diffusion radius cannot be determined, the observation record of this spraying tool object is only retained as an observation segment awaiting confirmation.

[0025] Using inheritable sanitation status objects in continuous video data as the processing objects, this method identifies inheritable sanitation status objects and determines their effective ends. Inheritable sanitation status objects are those that can carry prior sanitation status and affect subsequent cleaning coverage permits during public health operations. For cleaning tool objects, the effective end is determined as the cleaning end that comes into contact with public contact surfaces; for clamp objects, the effective end is determined as the clamping end; for hand-related objects, the effective end is determined as the hand mask or palm area; for spraying tool objects, the effective end is determined as the candidate spray landing point area. An inheritable sanitation status object observation record is generated for each inheritable sanitation status object. This observation record includes object category, object location, object mask, effective end location, appearance description, timestamp, tracking identifier, and detection confidence. For spraying tool objects, the effective end is determined as the candidate spray landing point area. The spraying direction is determined based on the spray bottle target frame, nozzle endpoint, nozzle orientation, and spraying action recognition results. The nozzle endpoint can be formed by the end key point detection results of the spray bottle target frame or by the slender end contour of the spray bottle segmentation mask. When the spraying direction intersects with the effective contact area of ​​the common contact surface sub-area, the intersection point is taken as the spray center point, and a candidate spray landing area is formed based on the spraying action duration, the distance from the nozzle to the contact surface, the historical spray landing radius, or the spray coverage range in the sanitation management procedures. The spraying duration must be no less than the minimum effective treatment time corresponding to the spray bottle. If the candidate spray landing area overlaps with the effective contact area, and the corresponding spraying action recognition result reaches the action recognition usable threshold, a corresponding area contact candidate result is generated. If the target frame of the spray bottle passes through the common contact surface sub-area but the nozzle orientation does not point towards the effective contact area, no cleaning coverage candidate is generated.

[0026] To ensure that inheritable hygiene status objects can be correctly inherited between different video segments, the observation records of inheritable hygiene status objects in adjacent video segments are used as the processing objects to generate object identity continuity quantities. The edge end reads the action end position records, object appearance descriptions, object movement directions, and scene region connectivity relationships from adjacent video segments, converting spatial continuity, motion consistency, appearance similarity, and region entry order reachability into dimensionless results between 0 and 1. Spatial continuity is formed by the inverse ratio of the distance between the end position of the action end in adjacent video segments and the beginning position of the action end in the next video segment relative to the upper bound of the reachable distance; the smaller the distance, the higher the value. Motion consistency is formed by the inverse ratio of the angle between the motion directions of the action ends in two video segments relative to 180 degrees; the more consistent the directions, the higher the value. Appearance similarity is formed by the cosine similarity of the object appearance feature vectors or the similarity of the color texture histogram. Region entry order reachability is determined based on the scene region connectivity relationships to determine whether the previous region can reach the next region within the time interval; 1 is taken if reachable, 0 if unreachable, and the result corresponding to the shortest reachable path is taken when multiple reachable paths exist. Each dimensionless result is used to form a continuous quantity of object identity based on the fusion weights determined by historical review records. When historical review records are insufficient, spatial continuity and appearance similarity are used as necessary inputs, with initial weights of no less than 0.3 for both. Motion consistency and region entry order accessibility are used as auxiliary inputs and allocated within the remaining weights, with each weight being a dimensionless value between 0 and 1 and the weights summing to 1. If motion consistency or region entry order accessibility is missing, the weights of the remaining input items are renormalized to form a continuous quantity of object identity. If spatial continuity or appearance similarity is missing, no continuous quantity of object identity is formed, and the corresponding adjacent video segments are transferred to the observation segments to be confirmed. The acceptance threshold for continuous object identity is determined by historical labeled samples, camera calibration error evaluation results, and historical review records, with a value range of 0.6 to 0.85. When historical review records are insufficient, and the camera coverage is continuous, the calibration error is less than half of the contact confirmation distance range, and the object appearance similarity reaches 0.75 or above, the initial acceptance threshold is 0.6 to 0.7. When the occlusion ratio is high, the object appearance similarity is less than 0.75, or there are multiple branches in the regional connectivity path, the initial acceptance threshold is 0.75 to 0.85.

[0027] Using the effective contact area of ​​the action end position record and the common contact surface sub-region as the processing object, regional contact candidate results are generated, and the action end position record is converted to the coordinate scale corresponding to the effective contact area. When the action end position record enters the effective contact area, a regional contact candidate result is formed; when the distance between the action end position record and the boundary of the effective contact area falls within the contact confirmation distance range, a regional contact candidate result is also formed. The unit of the contact confirmation distance range is meters, which is determined by the scene area configuration record and coordinate calibration error, and the value range is from 0.02 meters to 0.10 meters; approximate planar common contact. The distance value of the surface area is used in the scene supervision coordinate system, and the distance value of the non-planar public contact surface area is used in the local contact surface coordinate system. When the detection confidence of the action end is lower than the minimum available confidence, no regional contact candidate result is generated, and the corresponding segment is transferred to the observation segment to be confirmed. When the effective contact area is missing or the coordinate caliber cannot be matched, no permission basis for writing the disposal into the permission result is generated, and the corresponding public contact surface area is transferred to the generation path of the candidate result to be reviewed for closure. The regional contact candidate result is used as the candidate input for generating the time-series contact segment in S2, and does not enter the generation of the disposal writing into the permission result in S3.

[0028] Therefore, S1 outputs unified regulatory observation results, which include a public regulatory timeline, a scene regulatory coordinate system, a regulatory area base map, a set of public contact surface sub-regions, observation records of inheritable hygiene status objects, continuous quantities of object identity, regional contact candidate results, and local contact surface coordinate correspondences. After the unified regulatory observation results enter S2, they are used to generate time-series contact segments and construct a hygiene status inheritance graph. Among them, regional contact candidate results serve as candidate inputs for contact confirmation, continuous quantities of object identity serve as segments for segment inheritance, the regulatory area base map serves as the source of risk inheritance, reset confirmation, and blind spot confirmation areas, and the set of public contact surface sub-regions serves as the basis for generating sub-region coverage candidate relationships.

[0029] S2 takes the unified regulatory observation results output by S1 as the processing object, and reads the public regulatory time axis, scene regulatory coordinate system, regulatory area base map, public contact surface sub-region set, inheritable sanitary state object observation records, object identity continuous quantity, regional contact candidate results, and local contact surface coordinate correspondence in the unified regulatory observation results. The public regulatory time axis plays a role in the time constraint of fragment merging, the scene regulatory coordinate system and local contact surface coordinates play a role in the spatial constraint of contact judgment, the regulatory area base map plays a role in the regional source of fragment type, the inheritable sanitary state object observation records play a role in the processing object attribution, the regional contact candidate results play a role in the candidate input of contact confirmation, and the object identity continuous quantity plays a role in the admission judgment of state acceptance edge. S2 converts the candidate observations formed by S1 into time-series contact fragments and sanitary state inheritance graphs that can be back-read by S3.

[0030] First, contact confirmation processing is performed using the regional contact candidate results as the processing object. For each regional contact candidate result, the corresponding action end position record, contact area boundary, candidate frame start and end time, and action recognition result are read. The proximity between the action end and the contact area boundary is formed according to relative distance normalization: when the action end position enters the effective contact area, the proximity is 1; when the action end is outside the effective contact area and the distance between it and the effective contact area boundary does not exceed the contact confirmation distance range, the proximity is formed by 1 minus the ratio of this distance to the contact confirmation distance range; when this distance exceeds the contact confirmation distance range, the proximity is 0. The duration of the action end entering the effective contact area is formed by the ratio of the actual duration to the minimum identifiable contact duration, and is truncated to 1 when it is greater than 1. Action recognition results with a confidence level between 0 and 1 are used directly; when the action recognition result is a category label or raw score, it is first calibrated or normalized to between 0 and 1 using a validation set. When the confidence level of the action recognition result is lower than the action recognition usable threshold, it will not be included in the fusion of contact confirmation results. At the edge, the proximity, duration, and action recognition results that reach the action recognition usable threshold are used to form a contact confirmation result according to the fusion weights determined by historical review records. When historical review records are insufficient, proximity and duration are used as necessary inputs, and action recognition results are used as auxiliary inputs. When both proximity and duration are missing, no contact confirmation result is formed, and the corresponding region's contact candidate results are retained as observation segments to be confirmed.

[0031] The contact confirmation threshold is a dimensionless threshold ranging from 0.5 to 0.8, determined by historical verification records, scene area configuration records, and image recognition processing verification records. When historical verification records are insufficient, the contact confirmation threshold is determined based on the area boundary error, camera frame rate, and shortest identifiable contact duration in the scene area configuration records. The initial available confidence in the image recognition processing verification records is only used as the lower limit for the action recognition result to enter the contact confirmation result. In scenes with frequent occlusion, obvious reflection, or unstable boundary of the action end, the contact confirmation threshold is between 0.7 and 0.8. In scenes with a fixed camera, stable action end contour, and clear boundary of the effective contact area, the contact confirmation threshold is between 0.5 and 0.7. When the contact confirmation result does not reach the contact confirmation threshold, the corresponding contact candidate result is not included in the temporal contact segment merging process and is retained as an observation segment to be confirmed.

[0032] After the contact confirmation result reaches the contact confirmation threshold, consecutive candidate frame segments that meet the contact confirmation conditions are processed as the processing objects for segment aggregation. Adjacent candidate frame segments are read according to the common monitoring time axis, and it is checked whether adjacent candidate frame segments correspond to the same inheritable sanitary status object, the same contact area, and the same coordinate system. For approximately planar common contact surface sub-regions, segment aggregation uses the position result under the scene monitoring coordinate system; for non-planar common contact surface sub-regions, segment aggregation uses the position result under the local contact surface coordinate system. When the time interval between adjacent candidate frame segments does not exceed the segment break interval, the segment break interval is not less than the time deviation allowed by the common monitoring time axis processing in S1, and the continuity of object identity meets the acceptance condition, adjacent candidate frame segments are merged into a temporal contact segment. The unit of the segment break interval is seconds, and the value range is 0.1 seconds. The time interval is 1.0 seconds; for short-term hand contact scenarios, it is 0.1 to 0.3 seconds; for continuous handling scenarios corresponding to wiping cloths and spray bottles, it is 0.3 to 0.7 seconds; and for continuous mopping scenarios corresponding to mops, it is 0.5 to 1.0 seconds. These values ​​are determined based on the camera frame rate, the movement speed of the action end, and historical verification records. When historical verification records are insufficient, the initial value is determined by the shortest identifiable action duration corresponding to the camera frame rate and the action end type. If the coordinate diameter of adjacent candidate frame segments is inconsistent, the action end position record is missing, or the continuity of object identity does not meet the acceptance condition, then segment merging is not performed, and the corresponding candidate frame segments are retained separately. After segment aggregation, for each temporal contact segment, the inheritable hygiene status object, segment start and end time, segment area, contact confirmation result, and segment source are recorded, so that the subsequent hygiene status inheritance diagram accepts the status on a segment-by-segment basis.

[0033] Subsequently, the segment type is determined using the time-series contact segments and the regulatory area base map as the processing objects. When the time-series contact segment corresponds to a risk source area, it is formed as a risk contact segment; when the time-series contact segment corresponds to a reset confirmation area, it is formed as a candidate segment for reset confirmation; when the time-series contact segment corresponds to a blind zone, it is formed as a blind zone receiving segment; when the time-series contact segment corresponds to a public contact surface sub-area, it is formed as a sub-area coverage candidate segment. The segment type is provided by the regulatory area base map and is not determined separately by the action recognition result. If the same time-series contact segment spans multiple areas, the edge end counts the number of frames in each area where the action end continuously meets the contact confirmation threshold according to the public regulatory time axis, and converts the number of frames into effective dwell time. When the active device oscillates repeatedly near the region boundary, the region switching time is not determined by a single frame crossing, but by consecutive confirmation frames and backoff tolerance. When the active device enters the new region for at least two consecutive video frames, and the distance from the original region boundary exceeds a preset proportion of the contact confirmation distance range, the frame time of the first entry into the new region is determined as the region crossing time. If the active device travels back and forth between two adjacent regions and there are insufficient consecutive confirmation frames, the original segment type is maintained and the segment is marked as a boundary oscillation segment. When the longest effective dwell time exceeds the second longest effective dwell time and the difference reaches the minimum distinguishing duration, the region corresponding to the longest effective dwell time is used to determine the main segment type. When the difference does not reach the minimum distinguishing duration, the temporal contact segment is split into adjacent segments according to the region crossing time, and the segment types of the split segments are retained respectively.

[0034] For risk contact segments, the risk source area and the risk contact segment are used as the processing objects to form the risk source intensity and the risk contact inheritance basis. The edge reads the formation basis of the risk source area boundary, the proportion of the risk source area confirmed to have risk in historical review records, and the scene hygiene management procedures. The formation basis of the risk source area boundary is mapped to a basic source value between 0 and 1 according to the source type. Risk source areas that are manually calibrated or explicitly marked by hygiene management procedures take a higher basic source value; risk source areas identified by abnormal contaminants and manually reviewed take a medium-to-high basic source value; risk source areas identified by abnormal contaminants but not manually reviewed only enter the pending confirmation risk source area. The confirmation proportion in the historical review records is directly used as a historical confirmation value between 0 and 1; the scene hygiene management procedures are mapped to a procedure risk value between 0 and 1 according to the risk area type. When all three exist, the risk source intensity is formed according to the fusion weight determined by the historical review records; when historical review records are insufficient, the risk source area type and scene hygiene management procedures are used as the main inputs, and the abnormal contaminant identification results are used as auxiliary inputs. If any input is missing and cannot be supplemented by manual review or scene hygiene management procedures, the corresponding temporal contact segment will not be included in the risk inheritance relationship and will be transferred to the pending confirmation inheritance relationship. The value range of the risk source intensity is 0 to 1, and it is only used as the basis for S2 to establish the risk inheritance relationship, and is not directly output as the result of public health video supervision.

[0035] For candidate segments of the reset confirmation segment, the processing objects are the candidate segments of the reset confirmation segment, the reset confirmation area, and the observation records of inheritable sanitary status objects before and after the reset. A reset confidence condition judgment is performed, and the location of the reset confirmation area corresponding to the candidate segment of the reset confirmation segment, the action recognition result corresponding to the candidate result of the area contact, the observation records of inheritable sanitary status objects before and after the reset, and the object identity continuity quantity are read. For reset actions of the same type used by the original disposal object, the entry of the same inheritable sanitary status object before and after the reset into the subsequent cleaning coverage segment is used as the basis for acceptance. For reset actions of the type replaced by a new disposal object, the entry of the new disposal object into the subsequent cleaning coverage segment under the same operation process is used as the basis for acceptance. The consistency of the same operation process is confirmed based on the time sequence, location accessibility, and tool flow relationship between the reset confirmation segment and the subsequent cleaning coverage segment. When the personnel trajectory can be stably identified in continuous video data, the continuity of the personnel trajectory is considered. As an auxiliary confirmation basis for the same operation process, the reset confirmation area contact confirmation, reset action recognition, and post-reset disposal object acceptance are all satisfied. When the confidence level of the reset action recognition result reaches the action recognition usable threshold, the candidate fragments of the reset confirmation segment form a reset confirmation relationship that satisfies the reset credibility condition. If any of the results is missing, or the confidence level of the reset action recognition result does not reach the action recognition usable threshold, a reset confirmation relationship is not formed, and the corresponding candidate fragment is retained as a temporal contact segment that does not cut off the preceding risk. The reset credibility threshold is a dimensionless threshold of 0.6 to 0.85, determined by historical review records, scene hygiene management procedures, and image recognition processing verification records. When historical review records are insufficient, the initial threshold is determined by the operational requirements for cleaning, disinfection, disposal, and replacement actions in the hygiene management procedures. The reset confirmation relationship is output to the hygiene status inheritance graph and used to form the reset evidence credibility in subsequent S3 and determine whether the preceding risk has been cut off.

[0036] In this embodiment, cutting off preceding risks refers to setting a stop-acceptance flag on risk inheritance relationships located before the reset confirmation relationship that meets the reset credibility conditions in the health status inheritance graph, so that the risk inheritance relationship will no longer enter the risk contact inheritance quantity formation process of the current sub-region coverage candidate relationship. The cutting-off process does not delete the original risk contact segment and the original risk inheritance relationship, but only prevents them from continuing to be used as the exclusion basis for a qualified closed state in the current upstream status chain permission verification. If the reset confirmation relationship does not meet the reset credibility conditions, the acceptance of the post-reset disposal object is not established, or the same operation process cannot be confirmed, the cutting-off process is not performed, and the preceding risks continue to enter the subsequent permission verification along the status acceptance edge.

[0037] For blind zone continuation segments, the blind zone continuation segment, the inheritable health status object observation records before and after the blind zone, and the public supervision timeline are used as processing objects to form blind zone uncertainty. The edge end reads the duration of the blind zone continuation segment, the object identity continuity before and after the blind zone, and the degree of missing available observations within the blind zone. The maximum inheritable duration of the blind zone is formed by the blind zone length, the reachable velocity range corresponding to the action end type, the scene area connectivity, and historical verification records, with the dimension in seconds; when historical verification records are insufficient, an initial value is formed by dividing the blind zone length by the conservative reachable velocity of the action end type. When the blind zone duration does not exceed the maximum inheritable duration of the blind zone, the duration uncertainty sub-quantity is formed according to the ratio of the blind zone duration to the maximum inheritable duration of the blind zone; when it exceeds the maximum inheritable duration of the blind zone, this sub-quantity is set to 1. The object identity continuity before and after the blind zone is converted inversely to a discontinuous sub-quantity, i.e., obtained by subtracting the object identity continuity from 1; the degree of missing available observations within the blind zone is formed based on the ratio of the number of missing frames to the number of frames that should be observed within the blind zone. At the edge, a blind zone uncertainty is formed based on the fusion weights determined by historical review records. When historical review records are insufficient, the duration uncertainty and discontinuity sub-quantities serve as necessary inputs, while the degree of observation missing serves as an auxiliary input. When the continuity of object identity after the blind zone meets the acceptance condition, the blind zone duration does not exceed the maximum acceptable duration of the blind zone, and the blind zone uncertainty is below the limit threshold, the corresponding acceptance relationship is allowed to continue to enter the upstream state chain permission verification. When the blind zone uncertainty reaches the limit threshold, the continuity of object identity before and after the blind zone is missing, or the blind zone boundary cannot be determined, a pending acceptance relationship is formed in the health state inheritance graph, and in S3, the corresponding current sub-region coverage candidate relationship is entered into the generation path of the pending review closure candidate result.

[0038] After determining the types of temporal contact segments, a sanitation state inheritance graph is constructed using these segments as graph nodes. Each graph node records the inheritable sanitation state object corresponding to the temporal contact segment, the segment's start and end times, the region to which the segment belongs, the contact confirmation result, the segment type, and the segment's source. State connection edges are established based on the temporal reachability, spatial reachability, and appearance continuity of the same inheritable sanitation state object between adjacent temporal contact segments. Temporal reachability is formed based on the time interval between two temporal contact segments. Spatial reachability is formed based on the reachable path length between corresponding positions of two temporal contact segments, the time interval between segments, and the reachable speed range corresponding to the action end type. Appearance continuity is formed based on the appearance description in the observation record of the inheritable sanitation state object. Temporal reachability is formed according to the time interval between two temporal contact segments and the maximum inheritable interval corresponding to the action end type. When the time interval does not exceed the maximum inheritable interval, the temporal reachability is formed by subtracting the ratio of the time interval to the maximum inheritable interval from 1; when it exceeds the maximum inheritable interval, the temporal reachability is 0. Spatial accessibility is determined based on the reachable path length between corresponding positions of two temporally connected segments, the time interval between segments, and the reachable velocity range corresponding to the type of action end. The reachable path length preferentially adopts the topological reachable path length under the scene region connectivity relationship. When a topological path is missing, the ground projection distance is used. Geometric straight-line distances across walls and obstructions cannot be directly used as substitutes. Appearance continuity is determined based on the appearance description in the observation records of inheritable sanitary state objects. When the appearance description is a color, texture, shape outline, or depth feature vector, a dimensionless result between 0 and 1 is used for similarity conversion. At the edge end, it is first determined whether temporal accessibility, spatial accessibility, appearance continuity, and the object identity continuity quantity output by S1 all meet their respective admission conditions. If all admission conditions are met, the state inheritance result is formed according to the fusion weight determined by the historical review records. When the historical review records are insufficient, a cascade judgment method is used, that is, after temporal accessibility, spatial accessibility, and appearance continuity all meet the initial admission conditions, the object identity continuity quantity output by S1 is used as the state inheritance result. When the state acceptance result reaches the threshold for the establishment of a state acceptance edge, a state acceptance edge is established; if there is a lack of action end position record, object appearance description, or scene area connectivity relationship, the state acceptance edge is not established, and a disposal object acceptance interruption relationship is formed between the corresponding time sequence contact segments.

[0039] After establishing state acceptance edges, risk inheritance relationships, reset confirmation relationships, blind zone acceptance relationships, and sub-area coverage candidate relationships are further formed. A risk inheritance relationship refers to the directed relationship established between a risk contact segment and a subsequent sub-area coverage candidate segment when a risk contact segment reaches a subsequent sub-area coverage candidate segment through one or more state acceptance edges, and there is no reset confirmation relationship in the acceptance path that has met the reset credibility condition and completed the acceptance by the reset post-reset disposal object. The risk inheritance relationship records the risk contact segment identifier, the subsequent sub-area coverage candidate segment identifier, the state acceptance edge identifier on the acceptance path, the risk source strength, and whether it is cut off by the reset confirmation relationship. When the acceptance path extends from the risk contact segment corresponding to the risk source area to the subsequent sub-area coverage candidate segment, and there is no reset confirmation segment that meets the reset credibility condition in the acceptance path, the risk inheritance relationship is retained along the acceptance path; when the acceptance path passes through a reset confirmation segment that meets the reset credibility condition, and the reset post-reset disposal object acceptance is established, a stop acceptance flag is set for the risk inheritance relationship before the reset confirmation segment.

[0040] When the receiving path passes through a blind zone receiving segment and the uncertainty of the blind zone reaches the limit threshold, a receiving relationship to be confirmed is formed. After the receiving relationship to be confirmed is output to S3, it is used to trigger a candidate result for closure to be reviewed. When the timing contact segment corresponds to a common contact surface sub-region, and the position record of the action end in the timing contact segment enters the effective contact area of ​​the common contact surface sub-region, or the distance between the position record of the action end and the boundary of the effective contact area falls within the contact confirmation distance range, and the contact confirmation result reaches the contact confirmation threshold, a candidate relationship for sub-region coverage is formed. After the candidate relationship for sub-region coverage is output to S3, it serves as the backtracking entry point for the upstream state chain permission verification process. If there are risk inheritance relationships, reset confirmation relationships, and receiving relationships to be confirmed in the same receiving path, it is first determined whether to enter the candidate result for closure to be reviewed based on the receiving relationship to be confirmed. If it does not enter the candidate result for closure to be reviewed, it is then determined whether to cut off the risk inheritance relationship based on the reset confirmation relationship. Risk inheritance relationships that are not cut off are retained in subsequent candidate relationships for sub-region coverage.

[0041] S2 ultimately outputs a temporal contact segment and a sanitation state inheritance graph. The temporal contact segment serves as the source of graph nodes. The sanitation state inheritance graph carries the state inheritance relationship between the same inheritable sanitation state object in risk contact, reset confirmation, blind zone acceptance, and sub-area coverage candidate. S3 reads the current sub-area coverage candidate relationship in the sanitation state inheritance graph and backtracks the upstream state chain along the state acceptance edge to generate the disposal write permission result. S2 transforms the area contact candidate result of S1 into a sanitation state inheritance graph input with time, space, and object acceptance interface diameter, so that whether clean coverage can be written into a qualified closed state is constrained by the upstream state chain.

[0042] S3 takes the current sub-area coverage candidate relationship in the hygiene status inheritance graph output by S2 as the processing object, reads the temporal contact segment, state acceptance edge, risk inheritance relationship, reset confirmation relationship, pending confirmation acceptance relationship and sub-area coverage candidate relationship in the hygiene status inheritance graph, and reads the contact confirmation result, risk source intensity, object identity continuity quantity, blind zone uncertainty quantity and the effective contact area of ​​the corresponding public contact surface sub-area formed by S2. The current sub-area coverage candidate relationship indicates that the inheritable hygiene status object forms a clean coverage candidate for the public contact surface sub-area in the current temporal contact segment. This candidate relationship serves as the backtracking entry for the disposal write permission verification, and does not directly cause the public contact surface sub-area to enter the qualified closed state.

[0043] First, the temporal contact segment corresponding to the current sub-region coverage candidate relationship is used as the backtracking entry point. The upstream state chain of the same inheritable health state object is read forward along the state acceptance edge of the health state inheritance graph. The input source of the upstream state chain is the health state inheritance graph formed by S2. The processing action is to read the preceding temporal contact segment segment by segment along the state acceptance edge. The judgment criteria are whether the state acceptance edge is established and whether there is a disposal object acceptance interruption relationship. The output result is the upstream state chain corresponding to the current sub-region coverage candidate relationship. During the backtracking process, the state acceptance edge is read forward according to the start time of the current temporal contact segment until a reset confirmation relationship that meets the reset confidence condition, a disposal object acceptance interruption relationship, the starting node of the health state inheritance graph, or the preset backtracking time is reached.

[0044] When the preset backtracking time is reached, if there are still inheritable sanitation status objects that are continuously connected to the current sub-area coverage candidate relationship, and no starting node of the sanitation status inheritance graph or a reset confirmation relationship that meets the reset credibility condition has been obtained, then the current sub-area coverage candidate relationship will be transferred to the generation path of the candidate result to be reviewed. The preset backtracking time is measured in minutes, with a value range of 5 to 60 minutes, determined by the cleaning operation cycle of the public health scenario, historical review records, and scenario sanitation management procedures. If historical review records are insufficient, the shortest cleaning inspection cycle in the scenario sanitation management procedures will be used as the initial backtracking time. When the initial backtracking time is less than 5 minutes... The clock value is taken in 5-minute intervals. When the initial backtracking duration is greater than 60 minutes, it is taken in 60-minute intervals. For hospital waiting areas, pharmacy windows and number retrieval screens, the clock value is taken in 5 to 20 minutes. For school cafeterias and mother and baby rooms, the clock value is taken in 10 to 40 minutes. For public restrooms and areas near trash cans, the clock value is taken in 20 to 60 minutes. When the shortest cleaning and inspection cycle in the scene's hygiene management procedure exceeds 60 minutes, the preset backtracking duration is still set at 60 minutes as the upper limit for online license verification. The part that exceeds the upper limit for online license verification and still maintains the continuous acceptance of the object is not directly used for license status judgment, and the corresponding current sub-area coverage candidate relationship is transferred to the candidate result to be reviewed and closed.

[0045] During the backtracking process, when a reset confirmation relationship that meets the reset trust conditions is encountered, the risk contact segment before the reset confirmation relationship is stopped from being included in the current disposal write permission verification; when a disposal object inherits an interruption relationship, the timing contact segment before the interruption is stopped from being inherited into the current sub-area coverage candidate relationship; when the backtracking path passes through a blind zone and the uncertainty of the blind zone reaches the limit condition, the current sub-area coverage candidate relationship is transferred to the generation path of the closed candidate result to be reviewed, and the upstream state chain formed therefrom enters the generation process of the disposal write permission result, which is used to limit whether the current clean coverage can inherit the previous state.

[0046] After forming the upstream state chain, the current sub-region coverage candidate relationship and the corresponding public contact surface sub-region are used as processing objects to generate effective coverage quality. The action end position record, the effective contact area of ​​the public contact surface sub-region, the contact confirmation result, and the coverage duration under the public supervision time axis in the current time-series contact segment are read. The action end position record is converted to the coordinate scale corresponding to the public contact surface sub-region. First, the area ratio of the effective contact area of ​​the action end coverage is formed. Then, the coverage duration is converted into a dimensionless persistence result between 0 and 1 according to the minimum effective handling time of the public contact surface sub-region. The dimensionless persistence result is formed according to the ratio of the coverage duration to the minimum effective handling time. When the ratio is greater than 1, it is truncated to 1. When the coverage duration is missing or the minimum effective handling time cannot be formed, no dimensionless persistence result is generated, and the current sub-region coverage candidate relationship is transferred to the generation path of the candidate result to be reviewed and closed.

[0047] For public contact surface sub-areas with small effective contact areas, such as buttons, door handles, and screens, the area ratio should be weighted at 0.4 to 0.6. For continuous public contact surface sub-areas, such as desktops, handrails, and window sills, the coverage duration should be weighted at 0.3 to 0.5. For scenarios with reflections, obstructions, or unstable boundary conditions, the contact confirmation result should be weighted at 0.3 to 0.5. When multiple scenario conditions are met simultaneously, the edge end first determines the candidate weights for the area ratio based on the closure granularity of the public contact surface sub-area, then determines the candidate weights for the dimensionless continuous result based on the integrity of the coverage duration, and finally determines the candidate weights for the contact confirmation result based on the stability of the contact confirmation result. If the sum of the three candidate weights is not equal to 1, they are normalized by dividing by the sum of the three to make the final weight sum 1. If any of the area ratio, dimensionless continuous result, or contact confirmation result is missing, and an initial value cannot be generated based on historical review records or scenario hygiene management procedures, the permission status will not be generated and the current sub-area coverage candidate relationship will be transferred to the generation path of the pending review closure candidate result.

[0048] The effective coverage quality value ranges from 0 to 1, serving as a permissible input for determining whether cleaning coverage has reached the permissible effective contact area of ​​the public contact surface sub-area. The effective coverage quality threshold is a dimensionless threshold of 0.5 to 0.8, determined by the closure granularity of the public contact surface sub-area, the minimum effective coverage width at the action end, historical review records, and scene hygiene management procedures. When historical review records are insufficient, an initial threshold is determined using the effective contact area ratio of the public contact surface sub-area, the minimum effective coverage width at the action end, and the minimum effective handling time. The minimum effective handling time is measured in seconds, ranging from 0.5 seconds to 5 seconds, and is not less than the time resolution of the public supervision timeline corresponding to the current time-series contact segment, and the short-term wiping area corresponding to hands or gloves. The effective coverage quality threshold is set to 0.5 to 1.5 seconds for the scene, 1 to 3 seconds for continuous wiping scenes corresponding to the wiping cloth, 2 to 5 seconds for continuous mopping scenes corresponding to the mop, and 1 to 3 seconds for spraying scenes corresponding to the spray bottle. For public contact surface sub-areas such as buttons, screens, and door handles with small effective contact areas and clear contact positions, the effective coverage quality threshold is set to 0.7 to 0.8. For continuous public contact surface sub-areas such as desktops, handrails, and window sills, the effective coverage quality threshold is set to 0.5 to 0.7. When the effective contact area is missing, the coordinate caliber cannot be matched, or the coverage duration cannot be converted, the action of not generating a permission status is written into the permission result, and the current sub-area coverage candidate relationship is transferred to the generation path of the candidate result to be reviewed and closed.

[0049] Next, taking the upstream state chain as the processing object, a risk contact inheritance quantity is generated. This involves reading the risk inheritance relationship that has entered the current disposal write permission verification in the upstream state chain, the risk source strength corresponding to the risk contact segment, the contact confirmation result of the risk contact segment, and the state inheritance result from the risk contact segment to the current sub-region coverage candidate relationship. Risk contact segments that have been severed by a reset confirmation relationship that meets the reset trust condition and are no longer inherited by the current sub-region coverage candidate relationship are not included in the risk contact inheritance quantity formation process. Only risk contact segments in the same upstream state chain that have entered the current disposal write permission verification are included in the risk contact inheritance quantity formation process. For each included risk contact segment, the risk source strength, contact confirmation result, and state inheritance result are unified to a dimensionless result between 0 and 1. When all three exist... The risk inheritance result corresponding to the risk contact segment is formed according to the fusion weight or priority determined by the historical review records. When using fusion weight, each fusion weight is a dimensionless value between 0 and 1, and the weight sum is 1. When the historical review records are insufficient, the initial fusion weight is determined according to the risk source area type, contact confirmation threshold and state inheritance edge establishment threshold. Among them, the initial weight of the risk source intensity corresponding to the risk source area with residual transfer characteristics is not less than 0.4. The total weight of the contact confirmation result and state inheritance result corresponding to the risk source area with clear boundaries and only general contact risk is not less than 0.5. When there are multiple segment risk inheritance results, the highest value is taken as the risk contact inheritance amount. This value selection rule is used to retain the highest preceding risk in the same upstream state chain and does not dilute multiple risk contact segments by average.

[0050] The risk inheritance threshold ranges from 0 to 1, and is responsible for determining whether the pre-existing risk that has not been severed prevents clean coverage from being written into a qualified closed state. The risk inheritance limit threshold is a dimensionless threshold of 0.4 to 0.7, determined by historical review records, the type of risk source area, and the scenario hygiene management procedures. When historical review records are insufficient, the initial threshold is determined using the initial risk source intensity and hygiene management procedures formed in S2. For risk source areas with residual transfer characteristics, the risk inheritance limit threshold is 0.4 to 0.55. For risks with clear boundaries and only general exposure risk... In the source area, the risk inheritance restriction threshold is set between 0.55 and 0.7. The lower the risk inheritance restriction threshold, the stricter the exclusion of the risk inheritance relationship corresponding to the risk source area. When there is no risk contact segment in the upstream state chain, the risk contact inheritance amount is 0. When any result of risk source intensity, contact confirmation result or state acceptance result is missing, and an initial value cannot be formed based on the scenario hygiene management procedure or historical review record, the corresponding risk inheritance relationship will not enter the permission state judgment, and the current sub-area coverage candidate relationship will be transferred to the generation path of the candidate result to be reviewed and closed.

[0051] Further, using the reset confirmation relationship in the upstream state chain as the processing object, a reset evidence credibility quantity is generated. The edge end reads the reset confirmation area contact confirmation result, reset action identification result, reset post-reset handling object acceptance result, and same operation process confirmation result from the reset confirmation fragment, and first confirms whether the reset confirmation fragment comes from a reset confirmation relationship in S2 that meets the reset credibility condition. When the risk contact inheritance quantity reaches the risk inheritance limit threshold, the risk contact fragment that reaches the risk inheritance limit threshold is used as the verification object, and the reset confirmation relationship located after the risk contact fragment and before the current sub-region coverage candidate relationship is read; if there are multiple risk contact fragments that reach the risk inheritance limit threshold, the subsequent reset confirmation relationship is read for each risk contact fragment. If any risk contact fragment that reaches the risk inheritance limit threshold lacks a subsequent reset confirmation relationship that meets the reset credibility condition, the reset evidence credibility quantity is not used to cut off the preceding risk. The contact confirmation result for the reset confirmation area directly adopts the contact confirmation result formed by S2; the reset action recognition result adopts the normalized value of the action recognition confidence; the result of the post-reset treatment object acceptance adopts the object identity continuity quantity when the treated object before and after the reset or the new treated object enters the subsequent cleaning coverage segment; the confirmation result of the same operation process is formed by the time sequence, location accessibility, tool flow relationship and personnel trajectory continuity between the reset confirmation segment and the subsequent cleaning coverage segment. If the time sequence is not valid, the location is unreachable or the tool flow relationship is interrupted, the confirmation result of the same operation process is 0. When all four dimensionless results meet their respective admission conditions, the reset evidence credibility subquantity is formed according to the fusion weight determined by the historical review records; when the historical review records are insufficient, the contact confirmation result for the reset confirmation area, the reset action recognition result and the result of the post-reset treatment object acceptance are used as necessary inputs, and the confirmation result of the same operation process is used as an auxiliary input. When multiple risk contact segments have corresponding reset evidence credibility subquantities, the lowest value among the reset evidence credibility subquantities is used as the reset evidence credibility quantity corresponding to the current sub-area coverage candidate relationship, which is used to limit the impact of the weakest reset evidence in the same upstream state chain on the previous risk cutoff judgment.

[0052] The reset evidence credibility value ranges from 0 to 1, and it serves as a permissive factor in determining whether the prior risk has been severed from the reset confirmation relationship. The reset credibility threshold is a dimensionless threshold of 0.6 to 0.85, determined by historical review records, scene hygiene management procedures, and image recognition processing verification records. When historical review records are insufficient, the initial threshold is determined by the operational requirements for cleaning, disinfection, disposal, and replacement actions in the hygiene management procedures. For scenes involving disposal, glove replacement, and wiping cloth replacement, the reset credibility threshold is 0.75 to 0.85. For scenes involving cleaning, disinfection, and handwashing, and where the reset confirmation area boundaries are clear... In this scenario, the reset confidence threshold is set between 0.6 and 0.75. A higher threshold is used for reset actions involving object replacement to limit the reliability of the connection between the new disposal object and the same operation process. A lower threshold is used for reset actions involving the continued use of the original disposal object to limit the entry of the same disposal object into the subsequent cleaning coverage segment after the disposal is completed in the reset confirmation area. When there is no reset confirmation relationship in the upstream state chain, the reset evidence confidence value is set to 0. When any result of reset confirmation area contact confirmation, reset action identification, or post-reset disposal object acceptance is missing, the reset evidence confidence value is not used to cut off the preceding risk.

[0053] Taking the current sub-region coverage candidate relationship's corresponding connection path as the processing object, read the path object identity continuity quantity and blind zone uncertainty quantity. The path object identity continuity quantity is the path-based value of the object identity continuity quantity at the current sub-region coverage candidate relationship, and its source is the state connection result corresponding to each state connection edge in S2, no longer using the symbolic representation of the adjacent video segment object identity continuity quantity in S1. When the connection path includes multiple state connection edges, the lowest value among the state connection results corresponding to each state connection edge is used as the path object identity continuity quantity. This lowest value is used to limit the weakest connection link in the entire connection path; when the connection path contains only one state connection edge, the state connection result corresponding to that state connection edge is directly used. The connection threshold of the path object identity continuity quantity is a dimensionless threshold of 0.6 to 0.85, determined by historical review records, camera calibration error evaluation results, and scene area configuration records; when historical review records are insufficient, the initial connection threshold of the object identity continuity quantity in S1 is used. The continuous quantity of object identity written into the subsequent processing license result refers to the continuous quantity of path object identity corresponding to the candidate relationship of the current sub-region coverage.

[0054] The blind zone uncertainty originates from the blind zone connection segment in S2 and serves to correct whether the connection relationship after passing through the blind zone enters the review process. When the connection path includes multiple blind zone connection segments, the highest value of the blind zone uncertainty corresponding to each blind zone connection segment is used as the blind zone uncertainty corresponding to the current sub-region coverage candidate relationship. The blind zone uncertainty limit threshold is a dimensionless threshold of 0.4 to 0.7, determined by camera occlusion distribution, historical review records, and scene area configuration records. When historical review records are insufficient, the blind zone duration and camera field of view break length are used to determine the initial limit threshold. For scenarios with frequent occlusion, long blind zone duration, or long field of view break length, the blind zone uncertainty limit threshold is set to 0.4 to 0.55. For scenarios with continuous camera coverage and clear blind zone boundaries, the blind zone uncertainty limit threshold is set to 0.55 to 0.7. When the object identity continuity does not reach the connection threshold, or the blind zone uncertainty reaches the limit threshold, the current sub-region coverage candidate relationship does not enter the permission state judgment and is transferred to the generation path of the closed candidate result to be reviewed.

[0055] After obtaining the object identity continuity quantity, risk contact inheritance quantity, reset evidence credibility quantity, effective coverage quality, and blind zone uncertainty quantity, a disposal write permission result is generated. Each input quantity originates from the same upstream state chain corresponding to the current sub-region coverage candidate relationship, and all are dimensionless results between 0 and 1. The risk contact inheritance quantity is formed by the risk contact segment entering the current disposal write permission verification in the same upstream state chain; the reset evidence credibility quantity is formed by the reset confirmation relationship in the same upstream state chain that can cut off the risk contact segment that reaches the risk inheritance limit threshold; the effective coverage quality is formed by the common contact surface sub-region corresponding to the current sub-region coverage candidate relationship; the blind zone uncertainty quantity is formed by the blind zone inheriting segment traversed by the current upstream state chain; the object identity continuity quantity is formed by the inheriting path between the current sub-region coverage candidate relationship and the upstream state chain. In this embodiment, the disposal write permission result is formed according to the following formula: ;

[0056] in, This represents the sequence number of the inheritable sanitary state object in the current upstream state chain. This indicates the sequence number of the common contact surface sub-region corresponding to the current sub-region coverage candidate relationship. This indicates the temporal contact segment number where the current sub-region coverage candidate relationship is located; Indicates the first The inheritable health status object in the first In the first temporal contact segment, the first The handling of each public contact area is recorded in the permit result; This indicates a continuous quantity representing the identity of the objects formed in the preceding text; This indicates the effective coverage quality formed in the preceding text; This represents the uncertainty in the blind spot formed in the preceding text; This indicates the amount of risk exposure inherited from the preceding text; This indicates the credibility of the reset evidence formed in the preceding text; This indicates the identity acceptance threshold determined above; This represents the effective coverage quality threshold determined above; This indicates the threshold for limiting the uncertainty of the blind zone as defined above; This indicates the risk inheritance limit threshold determined above; This indicates the reset confidence threshold determined above; An indicator function that takes the value 1 when the condition is met and 0 when the condition is not met; This indicates a logical AND operation. This indicates a logical OR.

[0057] This formula uses a gating judgment method to form the disposal and write permission result. The object identity continuity quantity plays an access role between the current clean coverage and the upstream state chain; the effective coverage quality plays a triggering role for the closure of the common contact surface sub-region; the blind zone uncertainty quantity plays a role in the review and correction; and the risk contact inheritance quantity and the reset evidence credibility quantity jointly play a role in the prior risk cut-off judgment. When there is no risk contact segment in the upstream state chain, the risk contact inheritance quantity is 0; when there is no reset confirmation relationship in the upstream state chain, the reset evidence credibility quantity is 0; when the receiving path corresponding to the current sub-region coverage candidate relationship does not pass through the blind zone receiving segment, the blind zone uncertainty quantity corresponding to the current sub-region coverage candidate relationship is recorded as... 0; When the current sub-region coverage candidate relationship's corresponding acceptance path passes through a blind zone acceptance segment but the blind zone duration, blind zone boundary, or the continuity of object identity before and after the blind zone cannot be formed, the current sub-region coverage candidate relationship is transferred to the generation path of the candidate result to be reviewed and closed; when the continuity of object identity is missing, the current sub-region coverage candidate relationship is transferred to the candidate result to be reviewed and closed. This permission judgment is used to generate the disposal write permission result and enter the sub-region supervision status evidence chain update processing in S4 to restrict clean coverage without cut-off prior risks from being written into the qualified closed status, while allowing clean coverage without prior risks to enter the permission status judgment when the object acceptance and effective coverage meet the conditions.

[0058] When the disposal write-in permit result is a permit status, the disposal write-in permit status is output to the permit result, and the quantitative results related to the permit judgment and the upstream status chain identifier are written to the permit basis record. The permit basis record includes at least the current sub-area coverage candidate relationship, the corresponding inheritable health status object, the current time-series contact segment, the effective coverage quality, the risk contact inheritance amount, and the reset evidence credibility amount. The permit basis record enters S4 to update the corresponding public contact surface sub-area to a qualified closed state, and is used to associate the evidence chain of this permit judgment in the subsequent public health video monitoring results. The permit basis record is only used as input for the sub-area monitoring status evidence chain update processing in S4, and is not used as the basis for correcting the risk source intensity, contact confirmation threshold, status inheritance edge establishment threshold, effective coverage quality threshold, risk inheritance restriction threshold, blind zone uncertainty restriction threshold, or reset credibility threshold.

[0059] When the disposal write-in result is non-permitted, the non-permitted type is determined in the order of evidence uncertainty, uncut prior risks, and insufficient coverage. When the amount of uncertainty in the blind zone reaches the limit threshold, or the amount of continuity of object identity does not reach the acceptance threshold, the current sub-area coverage candidate relationship is transferred to the pending review closure candidate result. When the amount of risk contact inheritance reaches the risk inheritance limit threshold, the amount of reset evidence credibility does not reach the reset credibility threshold, and the contact confirmation result of the current sub-area coverage candidate relationship reaches the contact confirmation threshold, the current sub-area coverage candidate relationship is transferred to the risk transfer candidate result. When it does not enter the risk transfer candidate result and the effective coverage quality does not reach the effective coverage quality threshold, the current sub-area coverage candidate relationship is transferred to the insufficient coverage candidate result. After the pending review closure candidate result, risk transfer candidate result, and insufficient coverage candidate result enter S4, they are used to generate the corresponding sub-area regulatory status evidence chain update processing. When the same current sub-area coverage candidate relationship meets multiple non-permitted conditions at the same time, the pending review closure candidate result is used first; if the pending review closure candidate result is not met, the risk transfer candidate result is judged; if the risk transfer candidate result is not met, the insufficient coverage candidate result is judged.

[0060] The final output of S3 is the disposal write-permit result, which includes the permit status, non-permit type, permit basis record, and upstream status chain identifier. The permit status controls whether the cleaning coverage is written into the qualified closure state. The non-permit type controls whether the current public contact surface sub-area enters the update path of pending closure, risk transfer, or insufficient coverage. The permit basis record and upstream status chain identifier are used by S4 to generate public health video monitoring results. Through the upstream status chain permit verification process of S3, the public contact surface sub-area no longer enters the qualified closure state solely based on the cleaning coverage itself, but is jointly controlled by the prior risks of the inheritable health status object, reset evidence, blind spot acceptance, and effective coverage to determine the write qualification.

[0061] S4 takes the disposal write permission result output by S3 as the processing object, reads the permission status, non-permission type, permission basis record and upstream status chain identifier in the disposal write permission result, and reads the current sub-area coverage candidate relationship, public contact surface sub-area, current temporal contact segment, action end position record, object identity continuity quantity, risk contact inheritance quantity, reset evidence credibility quantity, effective coverage quality and blind zone uncertainty quantity associated with the disposal write permission result. The aforementioned data comes from the permission basis record or the associated information corresponding to the non-permission type formed by S3. S4 does not regenerate the intermediate quantities required for permission judgment from continuous video data. The public contact surface sub-area is used as the status update object, the disposal write permission result is used as the status update admission basis, the non-permission type is used as the status branch basis, and the permission basis record and upstream status chain identifier are used as the evidence chain source of the sub-area regulatory status record. S4 does not recalculate the disposal write permission result, nor does it re-determine whether the upstream status chain is valid. Instead, it performs sub-area regulatory status evidence chain update processing based on the permission status and non-permission type formed by S3.

[0062] When the public contact surface sub-area number, effective contact area, and coordinate caliber are consistent, the sub-area monitoring status record of that public contact surface sub-area is read. An overlapping boundary area refers to an area where the effective contact areas of two adjacent public contact surface sub-areas overlap under the same coordinate caliber, or where the distance from the point of action to the boundary of both adjacent sub-areas falls within the contact confirmation distance range. Overlapping boundary areas are recorded synchronously when S1 forms a set of public contact surface sub-areas. The recorded content includes the adjacent sub-area number, overlapping range, coordinate caliber, and corresponding contact confirmation distance range. If the public contact surface sub-area number is missing, the effective contact area is missing, the effective contact area area is 0, the coordinate caliber cannot be matched, or the positioning result falls within the overlapping boundary area of ​​two adjacent public contact surface sub-areas, no status writing of any public contact surface sub-area is performed. Only the current sub-area coverage candidate relationship, the reason for the inability to locate, and the corresponding temporal contact segment are recorded in the public health video monitoring results. Before the public contact surface sub-area positioning relationship is verified and confirmed, the aforementioned records are not included in the sub-area monitoring status record of any specific public contact surface sub-area.

[0063] Each public contact area sub-zone establishes a corresponding sub-zone monitoring status record. The sub-zone monitoring status record includes at least the public contact area sub-zone number, current status, status update time, most recent qualified closure basis, reason for pending review, upstream status chain identifier, and permit basis record index. The current status is used to indicate the monitoring status of the public contact area sub-zone in the public health video monitoring results. The status update time is determined by the public monitoring timeline. The most recent qualified closure basis is used to record the corresponding disposal written into the permit result when the qualified closure status was formed. The reason for pending review is used to record the source of the closure status to be reviewed. The permit basis record index is used to point to the permit basis record formed in S3. The sub-zone monitoring status record is used as the online status update object of S4, and is not used as the correction basis for risk source intensity, contact confirmation threshold, status acceptance edge establishment threshold, effective coverage quality threshold, risk inheritance limit threshold, blind spot uncertainty limit threshold, or reset credibility threshold.

[0064] When the disposal write permission result meets the permission conditions, the public contact surface sub-area corresponding to the current sub-area coverage candidate relationship is updated to a qualified closed state. Before the update, the current state and state update time of the public contact surface sub-area are read, and it is determined whether the end time of the time-series contact segment corresponding to the current disposal write permission result is later than the most recent state update time. If the end time of the time-series contact segment corresponding to the current disposal write permission result is later than the most recent state update time, the current state of the public contact surface sub-area is updated to a qualified closed state, and the current sub-area coverage candidate relationship, inheritable sanitary status object identifier, current time-series contact segment, effective coverage quality, risk contact inheritance amount, reset evidence credibility amount, blind zone uncertainty amount, object identity continuity amount, and upstream state chain identifier are written into the permission basis record. If the end time of the time-series contact segment corresponding to the current disposal write permission result is not later than the most recent state update time, the newer sub-area regulatory status record is not covered, and the disposal write permission result is retained as a historical permission basis record. This process ensures that the qualified closed state is only triggered by the disposal write permission result that has passed the S3 upstream state chain permission verification, avoiding the direct closure of the public contact surface sub-area based solely on the cleaning coverage trajectory.

[0065] When the write-permit result does not meet the permitting conditions, a differentiated status update is performed according to the non-permit type output in S3. If the non-permit type is a candidate result for closure pending review, the corresponding public contact surface sub-area is updated to the closure pending review state, and the corresponding reasons are written in the reasons for pending review, such as the continuous quantity of object identity not reaching the acceptance threshold, the uncertainty quantity of blind zone reaching the limit threshold, the coordinate caliber not corresponding, the effective contact area missing, or the key input quantity missing. If the non-permit type is a candidate result for risk transfer, the corresponding public contact surface sub-area is updated to the risk transfer state, and the risk contact segment that has not been cut off, the corresponding risk source area, the risk contact inheritance quantity, and the upstream status chain identifier are written in the sub-area regulatory status record. If the non-permit type is a candidate result for insufficient coverage, the corresponding public contact surface sub-area is updated to the insufficient coverage state, and the effective coverage quality, the effective coverage quality threshold, the current time-series contact segment, and the reason for insufficient coverage are written in the sub-area regulatory status record. All of the above status updates are based on the non-permit type in S3 and the non-permit type is not regenerated in S4.

[0066] When multiple actions are written into the permit and apply to the same public contact surface sub-area within the same monitoring time window, conflict resolution is performed according to the public monitoring time axis. The monitoring time window is measured in minutes, ranging from 1 minute to 30 minutes, and is determined by the usage frequency of the public contact surface sub-area, cleaning operation cycle, camera frame rate, and historical review records. If historical review records are insufficient, the shortest inspection interval or the shortest cleaning operation interval in the scene hygiene management procedure is used to determine the initial monitoring time window. When the initial monitoring time window is less than 1 minute, it is taken as 1 minute; when the initial monitoring time window is greater than 30 minutes... Clocks are set in 30-minute intervals. For hospital waiting areas, pharmacy windows, and ticket retrieval screens, 1 to 10 minutes are used. For school cafeterias and mother-and-baby rooms, 5 to 20 minutes are used. For public restrooms and areas near trash cans, 10 to 30 minutes are used. The monitoring time window is only used for resolving state conflicts in the same public contact area in S4 and does not replace the upstream state chain backtracking process in S3. The monitoring time window is not lower than the time resolution of the public monitoring time axis. When the monitoring time window is greater than the preset backtracking time of S3 in the same scenario, the preset backtracking time of S3 is used as the upper limit of the monitoring time window.

[0067] When resolving conflicts, the order of writing the disposal permit results is first determined by the end time of the corresponding time-series contact segment. Then, the writing priority is determined by the status type. If the non-permission type corresponding to the later-formed disposal permit result is a risk transfer candidate result or a pending closure candidate result, the corresponding public contact sub-area is updated to a risk transfer status or a pending closure status, and is written to the sub-area regulatory status record first. If the non-permission type corresponding to the later-formed disposal permit result is a coverage insufficiency candidate result, and a qualified closure status already exists within the same regulatory time window, and no new risk transfer status or pending closure status exists, the qualified closure status is retained, and the coverage insufficiency status is retained as a historical status record. The coverage insufficiency status only indicates that the corresponding disposal has not reached the effective coverage quality threshold and does not invalidate the already formed qualified closure status. If the disposal permit results of multiple permit statuses all meet the qualified closure conditions, the disposal with the later end time of the time-series contact segment is used first. The write permission result is used as the most recent qualified closure basis. When the end time of the timing contact segment of multiple disposal write permission results is the same or they are in the same common regulatory timeline frame, the disposal write permission result with higher effective coverage quality is used as the most recent qualified closure basis. When the end time of the timing contact segment and the effective coverage quality are the same, the system compares the upstream state chain identifier in the permission basis record. The upstream state chain identifier must include at least the actual existence of the current sub-area coverage candidate relationship, state succession edge identifier, risk inheritance relationship identifier, reset confirmation relationship identifier, or pending confirmation succession relationship identifier corresponding to the backtracking entry, and must be able to point to the corresponding timing contact segment. If the upstream state chain identifier is deemed to meet the integrity requirements, when multiple disposal write permission results meet the integrity requirements, the disposal write permission result with the later state update time is used as the most recent qualified closure basis. This conflict resolution process is used to avoid the same common contact surface sub-area being repeatedly written with contradictory states in a short period of time.

[0068] In this implementation, S1 to S4 are executed continuously and cyclically according to the public monitoring timeline. The edge continuously acquires public health video monitoring data, using video frames or monitoring time slices formed by merging consecutive video frames as trigger units. For each monitoring time slice, S1 generates a unified monitoring observation result. S2 incrementally writes new temporal contact segments, state succession edges, risk inheritance relationships, reset confirmation relationships, pending confirmation succession relationships, and sub-area coverage candidate relationships based on the existing health status inheritance graph. S3 only performs upstream state chain permission verification on newly added or changed current sub-area coverage candidate relationships. S4 updates the corresponding public contact surface sub-area's sub-area monitoring status record based on the new disposal write permission result. When an old state transitions to a new state, the public contact surface sub-area sequence number, monitoring time slice, and status update time are used as the writing order. Newer risk transfer states or pending review closure states can cover earlier qualified closure states. Earlier historical permission records are retained as an evidence chain and do not cover the current state. After a public contact surface sub-area enters a qualified closure state, the system continues to read the regional contact candidate results output by S2 after the qualified closure state is formed, corresponding to the same public contact surface sub-area. These regional contact candidate results serve as candidate inputs for subsequent state changes. If the contact confirmation result does not reach the contact confirmation threshold, it is recorded as a non-triggered record in the public health video monitoring result, without changing the current sub-area monitoring status. When the contact confirmation result reaches the contact confirmation threshold, the upstream state chain of the object corresponding to the regional contact candidate result is read, and the sub-area monitoring status evidence chain update process is re-executed according to the new disposal and written into the permission result. This incremental update mechanism enables the public contact surface sub-area to continue responding to subsequent contact events after forming a qualified closure state, avoiding the long-term coverage of subsequent risks by a single qualified closure.

[0069] For the pending closed state, the sub-area monitoring status record includes the inheritable health status object identifier that triggered the state, the current temporal contact segment, the blind zone uncertainty, the continuous quantity of object identity, and the reason for pending review. For the risk transfer state, the sub-area monitoring status record includes the risk contact segment, the risk source area, the risk contact inheritance quantity, the credibility of reset evidence, the upstream state chain identifier, and the risk inheritance relationship that has not been severed. For the insufficient coverage state, the sub-area monitoring status record includes the current sub-area coverage candidate relationship, effective coverage quality, effective coverage quality threshold, coverage duration, and effective contact area. The aforementioned records are included in the public health video monitoring results to characterize the state source of the current public contact surface sub-area. The sub-area monitoring status record, permit basis record, historical permit basis record, historical status record, reason for inability to locate, non-triggered record, and public health video monitoring results are only used as online monitoring results and online update basis, and do not directly modify the risk source intensity, threshold, weight, or sample weight in S1 to S3. The offline update of risk source intensity, threshold, weight, and sample weight is only performed based on manual review records, offline confirmation records, or scenario health management procedures.

[0070] The output of public health video surveillance results includes at least the public contact area sub-zone number, the current status of the sub-zone, the identifier of the inheritable health status object that triggered the status, the associated temporal contact segment, the upstream status chain identifier, the disposal written into the permit result, the permit basis record, the non-permit type, and the status update time. For qualified closed status, the output of public health video surveillance results includes the disposal written into the permit result of the most recent permit status, the effective coverage quality, the upstream status chain identifier, and the permit basis record. For risk transfer status, the output of public health video surveillance results includes the uninterrupted risk contact segment, the risk source area, the risk contact inheritance amount, and the risk transfer basis. For closed status pending review, the output of public health video surveillance results includes the reason for pending review, the blind spot uncertainty amount, the continuous amount of object identity, and the corresponding temporal contact segment. For insufficient coverage status, the output of public health video surveillance results includes the effective coverage quality, the effective coverage quality threshold, and the basis for insufficient coverage.

[0071] The validity period for a qualified closure status is determined by the scene hygiene management procedures, the type of public contact surface sub-area, personnel flow, and historical review records, and is measured in minutes. For high-frequency touch sub-areas such as buttons, screens, and door handles, the validity period is 5 to 30 minutes; for continuous touch sub-areas such as desktops, handrails, and window sills, the validity period is 15 to 60 minutes; and for sub-areas near public restrooms, trash cans, and dish recycling areas, the validity period is 5 to 20 minutes. When a qualified closure status reaches its validity period and no new permitting status is formed, the public contact surface sub-area will not continue to maintain the qualified closure status and will be transferred to a pending review closure status or a pending update status, with the reason for natural expiration recorded in the sub-area supervision status record. If a new risk inheritance relationship, pending confirmation of inheritance relationship, or non-permitted status is recorded as a permitting result during the validity period, the qualified closure status will be terminated early according to the status update rules of S4.

[0072] S4 ultimately generates the public health video surveillance results and completes the update of the evidence chain for the sub-area surveillance status of the public contact area. Through this process, the status update of the public contact area is no longer determined solely by the cleaning coverage trajectory, but is jointly controlled by the disposal record written into the permitting result, the non-permitting type, the permitting basis record, and the upstream status chain identifier formed in S3. The permitting status triggers the qualified closure status update, and the non-permitting type triggers the pending review closure status, risk transfer status, or insufficient coverage status update, thereby enabling the public health video surveillance results to retain the evidence chain between cleaning coverage, prior risks, reset evidence, and blind spot uncertainty.

[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating public health video surveillance results, characterized in that, Includes the following steps: Public health video surveillance data is acquired, and unified surveillance observation processing is adopted. The continuous video data in the public health video surveillance data is processed by the public surveillance time axis, and the action end of the inheritable health status object is converted into the action end position record under the scene surveillance coordinate system. Then, the action end position record is used to generate regional contact candidate results and unified surveillance observation results. The unified regulatory observation results are processed by state inheritance fragmentation, which transforms the regional contact candidate results of inheritable health status objects into temporal contact fragments, and constructs a health status inheritance graph containing sub-region coverage candidate relationships according to the inheritance path between temporal contact fragments. Using the current sub-area coverage candidate relationship as the backtracking entry point, perform upstream state chain permission verification on the sanitary state inheritance graph to prevent risk contact segments that have not been cut off by the reset confirmation relationship from being written into the qualified closed state of clean coverage, and generate disposal write permission results. The sub-regional regulatory status evidence chain update process is performed on the disposal and written into the licensing result. The sub-regional coverage candidate relationship that has passed the licensing verification is updated to a qualified closed state and written into the licensing basis record, generating a public health video supervision result carrying the upstream status chain identifier.

2. The method for generating public health video surveillance results according to claim 1, characterized in that, The specific operation of the unified processing of regulatory observation is as follows: Continuous video data, camera calibration parameters, and scene area configuration records are read from public health video monitoring data. A public monitoring timeline is formed based on the acquisition time of the continuous video data. The observation segments of the same inheritable health status object in different video frames are configured with the same time caliber. A scene monitoring coordinate system is established based on the camera calibration parameters. When the public contact surface sub-area is a non-planar surface, local contact surface coordinates are established. The position record of the action end of the inheritable health status object is converted to the coordinate caliber corresponding to the regional contact candidate result. For public contact surface sub-areas where the correspondence between the local contact surface coordinates and the scene monitoring coordinate system cannot be established, the corresponding regional contact candidate result will not be included in the licensing basis for the disposal and writing of the licensing result.

3. The method for generating public health video surveillance results according to claim 2, characterized in that, The specific operation for generating regional contact candidate results is as follows: Based on the scene area configuration record, the effective contact area of ​​the public contact surface sub-area is determined, and the position record of the action end of the inheritable hygiene status object is converted to the coordinate aperture corresponding to the effective contact area; when the action end position record enters the effective contact area, a regional contact candidate result is formed; when the distance between the action end position record and the boundary of the effective contact area falls within the contact confirmation distance range, a regional contact candidate result is formed, and the contact confirmation distance range is determined by the scene area configuration record or coordinate calibration error; when the public contact surface sub-area lacks effective contact area calibration, the effective contact area area is zero, or the coordinates of a local contact surface are missing, the public contact surface sub-area is transferred to the generation path of the candidate result to be reviewed for closure.

4. The method for generating public health video surveillance results according to claim 1, characterized in that, The specific operation of state inheritance fragmentation processing is as follows: taking the regional contact candidate results in the unified supervision and observation results as the processing object, reading the position maintenance status of the action end of the inheritable health state object in continuous video frames, and performing contact confirmation by combining the proximity of the action end to the boundary of the supervision area and the action recognition result corresponding to the regional contact candidate results; the action recognition result serves as the auxiliary input for contact confirmation. When contact confirmation meets the stable and continuous condition in consecutive video frames, adjacent candidate frame segments are merged into a temporal contact segment; each temporal contact segment is bound to the corresponding inheritable sanitary state object, segment start and end time, segment region and contact confirmation result, and single frame target box is not used as the basis for state inheritance.

5. The method for generating public health video surveillance results according to claim 4, characterized in that, The specific operations for constructing the health status inheritance graph are as follows: using temporal contact segments as graph nodes, state inheritance edges are established based on the temporal reachability, spatial reachability, and appearance continuity of the same inheritable health status object between adjacent temporal contact segments; when the inheritance path extends from the temporal contact segment corresponding to the risk source area to the subsequent sub-area coverage candidate segment, and there is no reset confirmation segment that meets the reset credibility condition in the inheritance path, the risk inheritance relationship of the risk contact segment to the subsequent sub-area coverage candidate segment is retained along the inheritance path; when the inheritance path passes through the temporal contact segment corresponding to the reset confirmation area, and the temporal contact segment meets the reset credibility condition, and the reset disposal object is successfully inherited, the risk inheritance relationship before the reset confirmation relationship is cut off.

6. The method for generating public health video surveillance results according to claim 5, characterized in that, The specific operation for the acceptance of the reset object after reset is as follows: When the reset confirmation relationship corresponds to a reset action of the type of continued use of the original object, the object with the same inheritable sanitary status before and after the reset is used as the basis for acceptance when it enters the subsequent cleaning coverage segment; when the reset confirmation relationship corresponds to a reset action of the type of replacement of the new object, the object of the new object under the same operation process is used as the basis for acceptance when it enters the subsequent cleaning coverage segment; the same operation process is confirmed for consistency based on the time sequence, location accessibility and tool flow relationship between the reset confirmation segment and the subsequent cleaning coverage segment. When the personnel trajectory can be stably identified in continuous video data, the continuity of the personnel trajectory is used as an auxiliary confirmation basis for the same operation process; when the same operation process cannot be confirmed, the corresponding reset confirmation relationship is not included in the upstream risk inheritance cutoff process.

7. The method for generating public health video surveillance results according to claim 6, characterized in that, The specific operation of the upstream state chain permission verification process is as follows: taking the temporal contact segment corresponding to the current sub-region coverage candidate relationship as the backtracking entry point, read the upstream state chain of the same inheritable health state object forward along the state inheritance edge of the health state inheritance graph; during the backtracking process, when a reset confirmation relationship that meets the reset trust condition is encountered, stop including the risk contact segment before the reset confirmation relationship into the current disposal and write it into the permission verification. When encountering an interrupted relationship of the object to be handled, stop accepting the temporal contact segment before the interruption into the current sub-region coverage candidate relationship; when the continuity of the object identity does not meet the acceptance condition after passing through the blind zone, transfer the current sub-region coverage candidate relationship to the candidate result to be reviewed for closure.

8. The method for generating public health video surveillance results according to claim 7, characterized in that, The specific operation for generating the write permission result is as follows: read the continuous quantity of object identity, effective coverage quality, blind zone uncertainty, risk contact inheritance, and reset evidence credibility corresponding to the current sub-region coverage candidate relationship; when the continuous quantity of object identity meets the acceptance condition, the effective coverage quality meets the coverage quality condition, and the blind zone uncertainty is lower than the limit condition, proceed to risk permission judgment; when the risk contact inheritance is lower than the limit condition, generate the write permission result with permission status; when the risk contact inheritance reaches the limit condition and the reset evidence credibility reaches the reset credibility condition, generate the write permission result with permission status; otherwise, generate the write permission result with non-permission status.

9. The method for generating public health video surveillance results according to claim 8, characterized in that, The specific operation for generating non-permitted types in the written permit results is as follows: when the uncertainty of the blind zone reaches the limit condition or the continuity of the object identity does not reach the acceptance condition, the current sub-region coverage candidate relationship is transferred to the candidate result to be reviewed and closed. After the continuity of object identity meets the acceptance condition and the uncertainty of blind zone is lower than the limit condition, when the risk contact inheritance reaches the limit condition, the confidence of reset evidence does not reach the reset confidence condition, and the contact confirmation result corresponding to the current sub-region coverage candidate relationship reaches the contact confirmation threshold, the current sub-region coverage candidate relationship will be transferred to the risk transfer candidate result. If the effective coverage quality does not meet the coverage quality conditions and the candidate results for closed-loop review and risk transfer are not included in the candidate results for risk transfer, the current sub-region coverage candidate relationship will be transferred to the candidate results for insufficient coverage.

10. The method for generating public health video surveillance results according to claim 1, characterized in that, The specific operation for updating the evidence chain of sub-area regulatory status is as follows: read the disposal write-permission result and locate its corresponding public contact surface sub-area; when the disposal write-permission result is in the permitted state, update the public contact surface sub-area corresponding to the current sub-area coverage candidate relationship to the qualified closed state, and write the permit basis record and the upstream state chain identifier into the sub-area regulatory status record of the public contact surface sub-area; when the disposal write-permission result is in the non-permission state, update the public contact surface sub-area to the corresponding pending review closed state, risk transfer state, or insufficient coverage state according to the non-permission type; the write-back of public health video supervision results is only used as the online update basis for the sub-area regulatory status record, and not as the update basis for offline parameters. Offline parameters include risk source intensity, contact confirmation threshold, state acceptance edge establishment threshold, effective coverage quality threshold, risk inheritance restriction threshold, blind zone uncertainty restriction threshold, reset credibility threshold, fusion weight, and sample weight.

Images