A central control room security video stream real-time screening method based on AI edge computing

Abstract

The present application relates to the security and video monitoring technical field, specifically to a kind of real-time screening method of security video stream in central control room based on AI edge computing, real-time coverage and observation ability of security observation unit are evaluated to video stream using edge AI, and qualified observation area is screened;Further identify the content that can be replaced by other video stream, determine the non-replaceable area of each stream, and accordingly implement complete transmission, local area transmission or edge screening, realize the on-demand real-time optimization transmission of security video stream in central control room.The present application identifies effective observation area through calibration mapping and real-time AI analysis, excludes shielding, low definition and low value picture;Combined with security constraints and multi-channel video cooperative coverage, non-replaceable content and necessary transmission area are determined, and on-demand complete, local or screening transmission of video stream is realized, while effectively monitoring key areas, bandwidth, decoding and centralized analysis resource consumption are reduced.

Description

Technical Field

[0001] This invention relates to the field of security and video surveillance technology, and in particular to a real-time filtering method for security video streams in a central control room based on AI edge computing. Background Technology

[0002] With the continuous improvement of the intelligence level of urban rail transit and large-scale public facilities, video surveillance systems have become an important technical means for central control rooms to carry out on-site operation monitoring, safety management, and abnormal event handling. Existing security systems typically deploy a large number of cameras in areas such as station halls, platforms, entrances and exits, equipment areas, and passageways, transmitting real-time video streams to the central control room for centralized monitoring, storage, and analysis. With the development of artificial intelligence image recognition and edge computing technologies, some video analysis tasks are gradually being pushed down to edge nodes closer to the cameras. These edge nodes perform real-time identification of personnel, equipment, environmental conditions, and specific events. Meanwhile, the security monitoring tasks in the central control room are usually related to specific spatial areas, key equipment, security boundaries, and corresponding monitoring requirements. Different cameras have different coverage areas, observation angles, and observable content in the field space, and the same area may be covered by multiple cameras.

[0003] Current video filtering methods rely heavily on the overall video status, camera field of view, and spatial coverage to determine the actual observation value of different areas within the same frame. When parts of the field of view are obstructed, in low light, or have degraded recognition quality, overall filtering can easily lead to the continued transmission of invalid areas or the simultaneous exclusion of critical local content. If substitution relationships between multiple cameras are determined solely based on the degree of field-of-view overlap, it may misjudge formal overlap as valid substitution, resulting in observation gaps at critical security boundaries. Furthermore, the coarse-grained output of overall transmission or filtering can easily lead to a mismatch between link resource consumption and actual security value. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides a real-time filtering method for security video streams in a central control room based on AI edge computing. The technical solution is as follows:

[0005] A real-time filtering method for security video streams in a central control room based on AI edge computing includes the following steps:

[0006] Based on real-time security video streams, camera on-site calibration information, and rail transit site area information, a mapping relationship between the video image coordinate system and a unified on-site spatial coordinate system is established. The theoretical spatial coverage area of ​​each real-time security video stream is determined and divided into spatial evaluation units. Based on the spatial evaluation units, video frame image analysis results are obtained. According to the camera on-site calibration information and the mapping relationship, the observation capability parameters corresponding to each spatial evaluation unit are determined. The observation capability parameters include ground pixel density, line-of-sight angle, occlusion ratio, sharpness, and illumination. The observation capability parameters corresponding to each spatial evaluation unit are judged item by item against the preset basic observation validity conditions. The current effective observation subdomains are then selected and merged. The observation capability parameters corresponding to each current effective observation subdomain and the real-time security video stream affiliation information are called. According to the camera identifier, the current effective observation subdomains are associated with the corresponding real-time security video streams. The spatial boundaries, observation status, and observation capability parameters of each subdomain are assigned to the corresponding video stream association items to obtain the current effective observation characterization set of the video stream.

[0007] Based on the security task configuration data, security observation units are divided and the security observation constraints of the central control room are determined. The current effective observation subdomain in the current effective observation representation set of the video stream is spatially matched with the security observation units. The overlapping area is determined as the spatial overlap area, and the spatial range corresponding to the spatial overlap area is determined as the actual coverage range of the current effective observation subdomain to the corresponding security observation unit. At the same time, the corresponding observation capability parameters are associated to obtain the spatial coverage association relationship. The security observation constraints of the central control room are called to judge the qualification of the actual coverage range and observation capability parameters corresponding to each spatial overlap area. The spatial overlap area that meets the corresponding security monitoring requirements is determined as the qualified observation subdomain, and the qualified observation association result of the video stream containing the qualified observation subdomain is obtained.

[0008] Based on the qualified observation association results of the video streams, real-time security video streams with qualified observation subdomains are screened, and the qualified observation subdomains corresponding to each video stream are determined as the available observation range of the corresponding video stream, forming an available alternative observation set. Then, video streams with qualified observation subdomains are selected as target video streams in sequence. For the security observation unit associated with the target video stream, available observation ranges corresponding to other video streams that spatially overlap with the qualified observation subdomains corresponding to the target video stream are extracted from the available alternative observation set. The extraction results are spatially combined and associated with corresponding observation capability parameters to determine the joint observation capability. Based on the security observation constraints of the central control room, the joint alternative observation area that meets the requirements is determined. The remaining part of the qualified observation subdomain after deducting the joint alternative observation area is determined as the non-replaceable observation content, resulting in a set of non-replaceable observation content.

[0009] Based on the set of irreplaceable observation content, all qualified observation subdomains and irreplaceable observation content are summarized according to the target video stream identifier, forming qualified observation ranges and irreplaceable observation ranges respectively. The spatial inclusion relationship between the irreplaceable observation range and the qualified observation range is determined. When the irreplaceable observation range is empty, it is determined that there is no necessary transmission area. When the irreplaceable observation range is in a local area of ​​the qualified observation range, it is determined that it is a locally necessary transmission area. When the spatial range of the irreplaceable observation range is consistent with that of the qualified observation range, it is determined that it is a complete necessary transmission area. According to the mapping relationship, the irreplaceable observation ranges corresponding to the locally necessary transmission area and the complete necessary transmission area are reverse-mapped to the video screen coordinate system to obtain the corresponding necessary transmission area and associated security observation unit, thus obtaining the necessary transmission configuration result of the video stream.

[0010] Based on the necessary transmission configuration results of the video stream, configure continuous transmission state, local area transmission state, or edge filtering state to obtain the real-time filtering results of the security video stream in the central control room.

[0011] Furthermore, obtaining the currently valid observation representation set of the video stream includes:

[0012] Based on the real-time security video streams generated by the cameras connected to each edge node, the on-site calibration information of the cameras, and the on-site area information of the rail transit, the video image position, the camera installation pose, and the boundary of the on-site area are extracted. The video image position is then mapped to a unified on-site spatial coordinate system to determine the theoretical spatial coverage area of ​​each real-time security video stream.

[0013] Spatial evaluation units are formed by intersecting the theoretical spatial coverage boundary with a horizontal grid of 10m and a vertical grid of 2m. The spatial boundary and video stream affiliation of each spatial evaluation unit are recorded.

[0014] Furthermore, the acquisition of the currently valid observation subdomain also includes:

[0015] Based on the spatial evaluation unit, the output video frame image analysis results of each edge node are obtained, and the target presentation state, screen occlusion state, image clarity state and on-site imaging state are mapped to each spatial evaluation unit. According to the on-site calibration information of the camera and the mapping relationship, the corresponding observation capability parameters are sorted out. The observation capability parameters corresponding to each spatial evaluation unit are judged item by item with the preset basic observation validity conditions, and the valid spatial evaluation units are screened and merged according to the spatial adjacency relationship to obtain the current valid observation subdomain.

[0016] Furthermore, the determination of the security observation constraints in the central control room includes:

[0017] Based on the monitoring areas, security monitoring requirements, key equipment locations, and key security boundaries in the current security task configuration data of the central control room, the monitoring areas are spatially divided according to the distribution of key equipment locations, key security boundaries, and security monitoring requirements. Each divided area is then matched with the security monitoring requirements, and the spatial range corresponding to each divided area is used as the monitoring range of the corresponding security observation unit. Observation conditions are determined in conjunction with the corresponding security monitoring requirements to form multiple security observation units. Each security observation unit is then associated with its corresponding security monitoring requirements, and the security observation constraints of the central control room are determined based on the security monitoring requirements corresponding to each security observation unit.

[0018] Furthermore, the acquisition of the video stream qualification observation correlation results includes:

[0019] The system calls upon the currently valid observation subdomain, spatial range, and observation capability parameters from the currently valid observation representation set of the video stream, spatially matches the boundary of the currently valid observation subdomain with the boundary of the security observation unit, associates the corresponding qualified observation subdomain according to the real-time security video stream identifier, and records the real-time security video streams that have not formed qualified observation subdomains, thereby obtaining the qualified observation association results of the video stream.

[0020] Furthermore, the acquisition of the set of non-substitutable observation content includes:

[0021] When there is no joint alternative observation area that meets the requirements, the joint alternative observation area is determined to be empty. The remaining part of the qualified observation subdomain corresponding to the target video stream after deducting the joint alternative observation area is determined as the non-substitutable observation content. The qualified observation subdomain and the non-substitutable observation content are associated according to the target video stream identifier. After processing each target video stream, a set of non-substitutable observation content is obtained.

[0022] Furthermore, obtaining the necessary transmission configuration results for the video stream includes:

[0023] Real-time security video streams that do not form qualified observation subdomains are identified as areas where there is no necessary transmission, thus obtaining the necessary transmission configuration results for the video streams.

[0024] Furthermore, the acquisition of the real-time filtering results of the central control room security video stream includes:

[0025] Based on the necessary transmission area type of each real-time security video stream in the necessary transmission configuration result of the video stream, extract the video stream identifiers corresponding to the complete necessary transmission area, the partial necessary transmission area and the non-existent necessary transmission area, configure the corresponding real-time security video stream to the continuous transmission state, the partial area transmission state and the edge filtering state respectively, and associate it with the edge node to obtain the video stream output state.

[0026] The real-time security video streams in the continuous transmission state and the local area transmission state are invoked, the corresponding necessary transmission areas and associated security observation units are extracted, the real-time security video stream in the continuous transmission state is continuously transmitted to the central control room as a complete video frame, and the real-time security video stream in the local area transmission state is executed with regional-level video output according to the necessary transmission areas to obtain the video stream transmission configuration.

[0027] Furthermore, the acquisition of the real-time filtering results of the central control room security video stream also includes:

[0028] According to the video stream transmission configuration, the real-time security video stream that is in the edge filtering state in the video stream output state is called, and the corresponding real-time security video stream is stopped from continuously outputting to the central control room. At the same time, the corresponding camera continues to perform local video acquisition and AI image analysis on the edge node side.

[0029] By associating the video stream output status, the necessary transmission area, and the corresponding security observation unit with the real-time security video stream identifier, the output relationship of each edge node is organized to obtain the real-time filtering result of the security video stream in the central control room.

[0030] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0031] 1. This invention establishes a mapping between video footage and on-site space through camera calibration information, subdivides the theoretical field of view into spatial evaluation units, and determines the effective observation subdomain by combining real-time image analysis, thereby eliminating occlusion, low-resolution and low-identification value areas.

[0032] 2. The present invention further determines the qualified observation range based on the security observation unit and corresponding constraints, identifies irreplaceable observation content through the joint coverage relationship of multiple video streams, and reverse maps it to form the necessary transmission area, so that complete transmission, partial transmission and edge filtering all correspond to the actual security needs, thereby reducing the transmission of repeated images, reducing the bandwidth, decoding and centralized analysis resource consumption, while retaining the effective observation information of key areas. Attached Figure Description

[0033] Figure 1 This is a flowchart of the method of the present invention;

[0034] Figure 2 This is a flowchart illustrating the process of obtaining the currently valid observation representation set of the video stream in this invention.

[0035] Figure 3 This is a flowchart illustrating the process of obtaining the video stream qualification observation correlation results of the present invention.

[0036] Figure 4 This is a flowchart illustrating the process of obtaining the set of irreplaceable observation content in this invention.

[0037] Figure 5 This is a flowchart illustrating the process of obtaining the necessary transmission configuration results for the video stream in this invention.

[0038] Figure 6 This is a flowchart illustrating the process of obtaining real-time filtering results of security video streams in the control room according to the present invention. Detailed Implementation

[0039] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0040] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0041] Please see Figure 1 This invention provides a technical solution: a real-time screening method for security video streams in a central control room based on AI edge computing, comprising the following steps:

[0042] Based on the real-time security video streams generated by cameras connected to each edge node, camera on-site calibration information, and rail transit site area information, a mapping relationship between the video frame coordinate system and the unified on-site spatial coordinate system is established. The theoretical spatial coverage area of ​​each real-time security video stream is determined and divided into spatial evaluation units. The video frames of the corresponding real-time security video streams are analyzed by AI image analysis models deployed on each edge node. Based on the video frame image analysis results, camera on-site calibration information, and mapping relationship, the observation capability parameters of each spatial evaluation unit are determined. Valid spatial evaluation units are selected according to preset basic observation validity conditions. The valid spatial evaluation units are merged according to their spatial adjacency relationship to obtain the current valid observation subdomain. The correlation between the current valid observation subdomain and the corresponding observation capability parameters is established. Then, based on the correspondence between the real-time security video stream, the current valid observation subdomain, and the observation capability parameters, the current valid observation representation set of the video stream is obtained.

[0043] Based on the monitoring area, security monitoring requirements, key equipment locations, and key security boundaries in the current security task configuration data of the central control room, the monitoring area is spatially divided to obtain multiple security observation units. Security observation constraints for the central control room are determined according to the security monitoring requirements corresponding to each security observation unit. Based on the current effective observation representation set of the video stream, each currently effective observation subdomain is spatially matched with each security observation unit to determine the spatially overlapping area. The spatial range corresponding to the spatially overlapping area is determined as the actual coverage area of ​​the current effective observation subdomain for the corresponding security observation unit. Simultaneously, corresponding observation capability parameters are extracted. The actual coverage area and observation capability parameters are judged for compliance according to the security observation constraints of the central control room. Spatially overlapping areas that meet the requirements are determined as qualified observation subdomains, resulting in qualified video stream observation association results. Real-time security video streams that do not form qualified observation subdomains are also recorded.

[0044] Based on the video stream qualified observation association results, the available observation range of the video stream with qualified observation subdomains is set as the corresponding qualified observation subdomain, forming an available alternative observation set. Video streams with qualified observation subdomains are selected sequentially as target video streams. For the security observation unit associated with the target video stream, available observation ranges corresponding to other video streams that spatially overlap with the qualified observation subdomains corresponding to the target video stream are extracted from the available alternative observation set. The extraction results are spatially combined and the joint observation capability is determined to obtain the joint alternative observation results. Based on the security observation constraints of the central control room, the joint alternative observation area that meets the requirements is determined. If there is no joint alternative observation area that meets the requirements, it is determined to be empty. The remaining part of the qualified observation subdomain after deducting the joint alternative observation area is determined as the non-replaceable observation content. The qualified observation subdomain corresponding to the target video stream is retained as the available observation range. The qualified observation subdomain and the non-replaceable observation content are associated according to the target video stream identifier. After the target video stream judgment is completed, the set of non-replaceable observation content is obtained.

[0045] Based on the set of irreplaceable observation content, all qualified observation subdomains and irreplaceable observation content associated with each target video stream are summarized to obtain the qualified observation range and the irreplaceable observation range respectively. Based on the spatial inclusion relationship between the irreplaceable observation range and the qualified observation range, the necessary transmission area type of each target video stream is determined as no necessary transmission area, partially necessary transmission area, or completely necessary transmission area. For target video streams with partially necessary transmission areas or completely necessary transmission areas, based on the mapping relationship between the video screen coordinate system and the unified on-site spatial coordinate system, the irreplaceable observation range is mapped back to the video screen coordinate system to obtain the corresponding necessary transmission area and associated security observation unit. Real-time security video streams that have not formed qualified observation subdomains are determined as having no necessary transmission area, and the necessary transmission configuration result of the video stream is obtained.

[0046] Based on the necessary transmission area type corresponding to each real-time security video stream in the necessary transmission configuration results, configure the video stream output status of each edge node's corresponding real-time security video stream. Configure real-time security video streams with complete necessary transmission areas to continuously transmit to the central control room. Configure real-time security video streams with partial necessary transmission areas to partial area transmission and perform area-level video output for the necessary transmission areas. Configure real-time security video streams without necessary transmission areas to edge filtering and stop continuous output to the central control room. At the same time, maintain local video acquisition and AI image analysis of the corresponding cameras on the edge node side. Based on the correspondence between the video stream output status of each real-time security video stream, the necessary transmission area, and the corresponding security observation unit, obtain the real-time filtering results of the security video stream in the central control room.

[0047] The AI ​​image analysis model is a visual analysis model deployed at edge nodes for target detection and segmentation of video frames in real-time security video streams. It includes one or more target detection and segmentation models. The target detection model obtains the target category, bounding box, and confidence score of personnel targets in the video frame. The target segmentation model obtains the outlines of personnel targets and foreground occlusions. The target detection model can employ YOLO series models, SSD models, Faster R-CNN models, RetinaNet models, or Transformer-based target detection models. The target segmentation model can employ Mask R-CNN models, YOLACT models, or YOLO series models with instance segmentation capabilities. Based on the target detection and segmentation results output by the AI ​​image analysis model, combined with the camera's on-site calibration information, the mapping relationship between the video frame coordinate system and the unified on-site spatial coordinate system, and the on-site acquisition parameters, the observation capability parameters corresponding to each spatial evaluation unit are determined.

[0048] Example 1, please refer to Figure 2 This embodiment illustrates the process of obtaining the current valid observation representation set of a video stream:

[0049] Based on the real-time security video streams generated by the cameras connected to each edge node, the on-site calibration information of the cameras, and the information of the rail transit site, the video image position, camera installation pose, and site area boundary are extracted. The video image position is mapped to a unified on-site spatial coordinate system to determine the theoretical spatial coverage area of ​​each real-time security video stream. The theoretical spatial coverage area is divided according to the preset area division rules, and the spatial boundary and video stream affiliation of each divided area are recorded to generate a spatial evaluation unit.

[0050] The system reads the camera identifier, video stream identifier, image resolution, frame rate, installation coordinates, installation height, horizontal deflection angle, pitch angle, and boundary coordinates of the field area from the edge node access. The camera installation coordinates are measured using the platform building coordinate control network. A total station is used to measure the three-dimensional coordinates of the camera lens optical center relative to the platform coordinate origin. The installation height is measured using a laser rangefinder. The horizontal deflection angle and pitch angle are read by the pan-tilt encoder and electronic tilt meter. The video image resolution and frame rate are read from the camera encoding parameters. The field area boundary is jointly determined by the platform electronic map database and total station re-measurement points. A unified field spatial coordinate system with a length of 120m and a width of 10m is established for the platform area. At least four non-collinear calibration points are set up on the platform ground. The field coordinates of each calibration point and the pixel coordinates in the same video frame are recorded point by point. Lens distortion correction is first performed on the video image. Then, a planar mapping relationship between the pixel coordinates and the field ground coordinates is established based on the calibration point correspondence. The installation coordinates of camera C01 are recorded as follows: The horizontal deflection angle is recorded as follows: The pitch angle is recorded as The mapping results of the four corners of the screen are registered in sequence as follows: , , , The theoretical space coverage area is calculated based on the area of ​​a trapezoid:

[0051] ;

[0052] in, Indicates the theoretical spatial coverage area. Indicates the length of the upper base. Indicates the length of the lower base. Indicates the distance between two parallel sides, substitute into , , ,get Spatial evaluation units are formed by intersecting a 10m horizontal and 2m vertical grid with the theoretical spatial coverage boundary. The boundary of U11 is 10.0m to 20.0m horizontally and 2.0m to 4.0m vertically, with an area of ​​[area missing]. The overlapping area is less than Edge units are not included in the spatial evaluation unit set, and the overlapping area is not less than The edge units are written according to the actual polygon boundaries to complete the spatial evaluation unit registration.

[0053] Based on the spatial evaluation unit, the image analysis results of the output video frames of each edge node are obtained. The target presentation status, screen occlusion status, image clarity status and on-site imaging status are mapped to each spatial evaluation unit. The corresponding observation capability parameters are organized according to the on-site camera calibration information and spatial mapping relationship. The observation capability parameters corresponding to each spatial evaluation unit are judged item by item with the preset basic observation validity conditions. Valid spatial evaluation units are screened and merged according to spatial adjacency relationship to obtain the current valid observation subdomain.

[0054] The video frame analysis records of spatial evaluation units U11 and U12 at the same timestamp or with a time difference of no more than 40ms are read. The video frames are sequentially decoded from a real-time bitstream of 25 frames per second by the edge nodes. The personnel target bounding boxes are output by the target detection processing unit. The target presentation status is generated according to the target bounding box confidence, target category, and number of consecutive frames. When the personnel target category confidence is not less than 0.80 and the corresponding target is detected in the same spatial evaluation unit for 3 consecutive frames, the target presentation status is written as 1; otherwise, it is written as 0. The center pixel of the target bounding box is transformed to the field coordinates according to the planar mapping relationship established in paragraph 1. The target center point in U11 is recorded as follows. Its coordinates fall within the boundaries of U11, ranging from 10.0m to 20.0m horizontally and from 2.0m to 4.0m vertically. The occlusion area is determined by the overlapping pixels of the person instance segmentation contour and the foreground occluding object contour, with the occlusion ratio according to... Calculate, where, Indicates the occlusion ratio. Indicates the number of occluded pixels. Indicates the total number of pixels in the target outline, in U11 Pixels pixels, to obtain Ground pixel density is based on:

[0055] ;

[0056] in, Represents ground pixel density, in units of , Indicates the length of the measured line pixels. This indicates the on-site length of the measurement line. In U11, a 10.0m measurement line corresponds to 860 pixels. Sharpness was obtained using the mean absolute value of the grayscale Laplacian response, and normalized to 0 to 1 based on the 5th and 95th percentile values ​​of the valid samples from the camera over 30 consecutive days. The sharpness of U11 was 0.81. Illuminance of 145 lx was obtained by sampling at 10-second intervals using a digital illuminance meter installed above the corresponding area. The line-of-sight angle... The angle between the line-of-sight vector from the camera's optical center to the unit's center point and the ground normal vector is obtained, with a pixel density of not less than 60px / m and a line-of-sight angle not exceeding [missing information]. The basic observation validity conditions are set as follows: occlusion ratio not exceeding 0.25, sharpness not less than 0.70, illuminance between 100 lx and 500 lx, and target presentation state 1. U11 meets these conditions, and U12 corresponds to the parameter 78 px / m. The values ​​of 0.18, 0.77, 138 lx and target presentation state 1 also meet the conditions. U11 and U12 share a boundary of length 2.0m, belong to the same video stream and have a timestamp difference of no more than 40ms. The two are merged into the current effective observation subdomain D01, which is 10.0m to 30.0m horizontally and 2.0m to 4.0m vertically.

[0057] Based on the currently effective observation subdomains, the corresponding observation capability parameters and real-time security video stream affiliation information of each subdomain are called. The currently effective observation subdomains are associated with the corresponding real-time security video streams according to the camera identifiers. The spatial boundaries, observation status, and observation capability parameters of each subdomain are assigned to the corresponding video stream association items. The correspondence between each association item is sorted out according to the real-time security video stream identifiers to obtain the current effective observation representation set of the video stream. Among them, the observation status includes target presentation status, unobstructed status, clear status, and imaging status, which correspond to the aforementioned target presentation status, screen obstruction status, image clear status, and on-site imaging status, respectively.

[0058] The spatial boundary, observation status, observation capability parameters, video stream affiliation information, and timestamp of the currently valid observation subdomain D01 are retrieved. D01 is associated with camera C01 and video stream VS-C01. A conservative aggregation rule is applied to similar parameters of U11 and U12. Ground pixel density is calculated as follows:

[0059] ;

[0060] in, Indicates the lowest ground pixel density within the D01 range. , Representing the ground pixel density of U11 and U12 respectively, substituting 86px / m and 78px / m yields... The angle of the line of sight is as follows:

[0061] ;

[0062] in, This represents the maximum line-of-sight angle of D01, and we obtain... The occlusion ratio is as follows:

[0063] ;

[0064] get Clarity according to Record, obtain Illuminance according to Record, obtain ,in, Indicates the maximum occlusion ratio. Indicates the minimum normalized sharpness. The minimum illumination is represented by the logical AND rule for target presentation status, unobstructed status, clear status, and imaging status. When the corresponding status of two units is 1, the corresponding status of D01 is written as 1. When the status of any unit is 0, the corresponding status of D01 is written as 0. In this example, all statuses are 1. Thus, the target presentation status, unobstructed status, clear status, and imaging status corresponding to the current effective observation subdomain D01 are obtained respectively, and are associated and recorded as the observation status corresponding to D01. The boundary, parameters, status, video stream identifier, camera identifier, and timestamp of D01 are written into the VS-C01 association item. An index is built according to the video stream identifier, observation subdomain identifier, and timestamp for subsequent spatial matching steps to obtain the current effective observation representation set of the video stream.

[0065] Example 2, please refer to Figure 3 Based on the current valid observation representation set of the video stream, this embodiment further determines the actually usable qualified observation range of each video stream for the current security task by combining the current security task configuration in the central control room:

[0066] Based on the monitoring areas, security monitoring requirements, key equipment locations, and key security boundaries in the current security task configuration data of the central control room, the monitoring areas are spatially divided according to the distribution of key equipment locations, key security boundaries, and security monitoring requirements. Each of the divided areas is matched with a security monitoring requirement, and the spatial range corresponding to each divided area is used as the monitoring range of the corresponding security observation unit. The observation conditions are determined in conjunction with the corresponding security monitoring requirements. The monitoring range and observation conditions of each area are judged to form multiple security observation units, which are then associated with the corresponding security monitoring requirements to obtain the security observation constraints of the central control room.

[0067] The system reads the monitoring area R01, escalator E01 location, platform screen door safety boundary, and monitoring requirements from the central control room security task configuration. R01 and the platform screen door safety boundary are retrieved from the electronic map database. The coordinates of escalator E01 are obtained by comparing the installation coordinates in the equipment log with the coordinates measured on-site using a total station. The R01 boundary is 15.0m to 45.0m laterally and 0.0m to 8.0m longitudinally. The center coordinates of E01 are... A01 is formed by extending 5.0m laterally to both sides and 2.0m longitudinally to both sides of the center of escalator E01. The lateral starting point of A01 is as follows: Calculate, lateral endpoint according to Calculate, where, This represents the transverse coordinates of the center of E01. Substituting 28.0m, we get transverse coordinates from 23.0m to 33.0m. The longitudinal starting point is... Calculation, longitudinal endpoint according to Calculate, where, Let E01 be the center longitudinal coordinate. Substituting 3.0m, we get a longitudinal range of 1.0m to 5.0m. The resulting spatial range of A01 is 23.0m to 33.0m laterally and 1.0m to 5.0m longitudinally. This spatial range is the monitoring range corresponding to A01. The monitoring requirements for A01 are set as follows: personnel target presentation status 1, pixel density not less than 70px / m, occlusion ratio not higher than 0.20, clarity not less than 0.75, and illuminance not less than 120lx. The longitudinal coordinate of the platform screen door safety boundary is 0.5m. Combining this with the boundary of the platform's passable area, A02 is defined as 15.0m to 45.0m laterally and 0.0m to 2.0m longitudinally. In this embodiment, we use... The key security boundary for spatial division is the platform screen door security boundary read from the electronic map database. Based on the platform screen door security boundary and the boundary of the platform's passable area, A02 is formed. The spatial range of A02 is 15.0m to 45.0m horizontally and 0.0m to 2.0m vertically. This spatial range is the monitoring range corresponding to A02. The monitoring requirements for A02 are set as follows: target presentation status 1, pixel density not less than 60px / m, occlusion ratio not higher than 0.25, clarity not less than 0.70, and illuminance not less than 100lx. The security observation constraints in the central control room are written according to the security observation unit identifier, spatial boundary, associated equipment or security boundary, monitoring conditions, and task version number.

[0068] Based on the security observation constraints in the central control room, the system calls upon the current effective observation subdomains, spatial ranges, and observation capability parameters from the current effective observation representation set of the video stream. It then performs spatial matching between the boundaries of the current effective observation subdomains and the boundaries of the security observation units, determines the overlapping positions and spatial correspondences of the two types of regions, identifies the overlapping regions as spatially overlapping regions, and determines the spatial range corresponding to the spatially overlapping regions as the actual coverage range of the current effective observation subdomains for the corresponding security observation units. Simultaneously, it associates the corresponding observation capability parameters to obtain the spatial coverage correlation relationship.

[0069] Read the boundary, observation parameters, status, and timestamp of D01 in the current valid observation representation set of the video stream. Read the boundary of security observation unit A01. The horizontal range of D01 is 10.0m to 30.0m, and the vertical range is 2.0m to 4.0m. The horizontal range of A01 is 23.0m to 33.0m, and the vertical range is 1.0m to 5.0m. The horizontal starting point of the spatially overlapping area is... Calculate, where, Indicates the horizontal starting point of the overlapping area. Indicates the horizontal starting point of D01. Indicates the lateral starting point of A01, which is 23.0m. The lateral ending point is... The calculated length is 30.0m, with the longitudinal starting point as follows: The calculated value is 2.0m. For the longitudinal endpoint, according to... The calculated value is 4.0m, where, Indicates the lateral endpoint of the overlapping area. , These represent the longitudinal start and end points of the overlapping region, respectively. and When overlapping areas are formed, this example meets the condition, and the overlapping areas are calculated as follows: Calculation, obtained ,in, This represents the overlapping area, which is registered as O01. O01 corresponds to a spatial range of 23.0m to 30.0m horizontally and 2.0m to 4.0m vertically. This spatial range is the actual coverage area of ​​the current effective observation subdomain D01 over the security observation unit A01, inheriting the pixel density of 78px / m and the line-of-sight angle of D01. The occlusion ratio is 0.18, the resolution is 0.77, the illuminance is 138lx, the target presentation status is 1 and the corresponding timestamp are associated with VS-C01, D01 and A01. D01 and A02 only touch at the vertical coordinate of 2.0m, with an overlap height of 0m and an overlap area of ​​0m². They are not written into the spatial overlapping area record to obtain the spatial coverage association relationship.

[0070] Based on the spatial coverage relationship, the security observation constraints of the central control room are invoked to make a qualification judgment on the actual coverage range and observation capability parameters of each spatial overlapping area. The spatial overlapping area that meets the monitoring requirements of the security observation unit is determined as a qualified observation subdomain. The corresponding qualified observation subdomain is associated with the real-time security video stream identifier. The real-time security video stream that does not form a qualified observation subdomain is recorded to obtain the qualified observation association result of the video stream.

[0071] The system retrieves the actual coverage area of ​​O01 in the spatial coverage association, along with its associated observation capability parameters, observation status, and A01 monitoring requirements. It then performs item-by-item judgments on pixel density, occlusion ratio, sharpness, illumination, and target presentation status. The pixel density judgment value is determined according to... when ,otherwise ,in, Indicates the pixel density determination value. Indicates the pixel density of O01. This represents the lower limit of pixel density for A01. Substituting 78px / m and 70px / m yields... The occlusion determination value is based on when ,otherwise ,in, Indicates the occlusion ratio of O01. This represents the upper limit of the occlusion ratio of A01. Substituting 0.18 and 0.20, we get... Sharpness judgment value according to when ,otherwise Substituting 0.77 and 0.75, we get... Illuminance determination value according to when ,otherwise Substituting 138lx and 120lx, we get Target presentation status judgment value The target status is equal to O01, which is 1 in this example. The pass / fail flag is set accordingly. Calculate, where, This indicates a qualified spatial overlapping area, with a value of 0 or 1. In this example... Update O01 to the qualified observation subdomain G01. The pixel density of the spatially overlapping region corresponding to VS-C02 is 64px / m. Substituting 64 into the equation results in a value less than 70. Its qualification mark If a qualified observation subdomain is not formed, and VS-C03 does not form a currently valid observation subdomain, it is registered as a video stream that has not formed a qualified observation subdomain. The qualified observation association results of the video stream are obtained by summarizing the video stream identifier, security observation unit identifier, subdomain identifier, timestamp, and task version number.

[0072] Example 3, please refer to Figure 4 Based on the qualified observation correlation results of the video stream, this embodiment determines the observation content in each target video stream that cannot be replaced by other video streams by comparing the qualified observation range of different video streams within the same security observation unit:

[0073] Based on the video stream qualified observation association results, real-time security video streams with qualified observation subdomains are screened, and the qualified observation subdomains corresponding to each video stream are determined as the available observation range of the corresponding video stream. The correspondence between video streams and available observation ranges is organized according to the security observation unit affiliation relationship. Then, video streams with qualified observation subdomains are selected as target video streams in sequence, and the security observation units corresponding to the target video streams are associated to generate a set of available alternative observations.

[0074] Read the qualified observation association results from the video stream, and screen out VS-C01 and VS-C04 with qualified observation subdomains. Define the usable observation range as the horizontal 23.0m to 30.0m and vertical 2.0m to 4.0m of the G01 boundary corresponding to VS-C01. The area is then calculated as follows: Calculate, where, The area of ​​G01 is defined as follows: the usable observation range is defined as 25.0m to 33.0m laterally and 1.5m to 4.5m longitudinally corresponding to the G04 boundary of VS-C04. The area is calculated as follows: Calculate, where, The area of ​​G04 is represented by the calibration data, video frame analysis data, and observation capability parameters of VS-C04, which are obtained using the same acquisition, mapping, and judgment methods as VS-C01. The observation unit identifiers of G01 and G04 are read, both of which are A01. The timestamps of both are read, and the same alternative observation period is written only when the time difference does not exceed 40ms. VS-C01, G01, A01, 14.0m², pixel density 78px / m, occlusion ratio 0.18, resolution 0.77, illumination 138lx, and timestamp are written to the first target record. VS-C04, G04, A01, 24.0m², corresponding capability parameters, and timestamp are written to the second target record. VS-C02 and VS-C03 do not form qualified observation subdomains and are not written to the available alternative observation set. The processing order is determined by the ascending order of the video stream identifier, and the available alternative observation set is generated.

[0075] Based on the available alternative observation set, for the security observation unit associated with the target video stream, the available observation range corresponding to other video streams that spatially overlap with the qualified observation subdomains corresponding to the target video stream is extracted. The extracted ranges are combined according to the spatial overlap position and associated with the corresponding observation capability parameters to determine the joint observation capability corresponding to the combined area. The observation status of the combined area is judged according to the security observation constraints of the central control room to obtain the joint alternative observation area.

[0076] Set VS-C01 as the target video stream, read its G01 boundary, associated A01, and timestamp. Read the G04 boundary corresponding to VS-C04 within the same A01 and the same alternative observation period. The horizontal range of G01 is 23.0m to 30.0m, and the vertical range is 2.0m to 4.0m. The horizontal range of G04 is 25.0m to 33.0m, and the vertical range is 1.5m to 4.5m. The horizontal starting point of the overlapping area is... The horizontal endpoint is The vertical starting point is The longitudinal endpoint is We obtain J01, and the area of ​​J01 is calculated according to... Calculate, where, Indicates the area of ​​J01. , These represent the lateral start and lateral end points of J01, respectively. , Representing the longitudinal start point and longitudinal end point of J01 respectively, we obtain... The joint alternative observation capability does not use an arithmetic average of parameters from different video streams. Instead, it reads the VS-C04 observation capability parameters that actually perform the alternative function in J01. The VS-C04 pixel density of 82px / m is obtained by dividing the pixel length of the calibration measurement line by the field length. The occlusion ratio of 0.14 is obtained by dividing the number of pixels in the occlusion mask by the number of pixels in the target outline. The sharpness of 0.80 is obtained by the normalized Laplacian response. The illuminance of 151lx is obtained by the digital illuminance meter. The target presentation state is 1. These are checked against the corresponding conditions of A01. If all conditions are met, J01 is registered as a joint alternative observation area. If the same spatial location is covered by multiple alternative video streams, the joint alternative capability is formed by taking the maximum value of pixel density, the minimum value of occlusion ratio, the maximum value of sharpness, the maximum value of illuminance, and the logical OR value of target presentation state. The premise is that the timestamp difference between the video streams does not exceed 40ms.

[0077] Based on the joint alternative observation area, when there is no joint alternative observation area that meets the security observation constraints of the central control room, the corresponding joint alternative observation area is set to empty. The remaining spatial range after deducting the joint alternative observation area from the qualified observation subdomain is determined. The remaining spatial range is determined as the non-replaceable observation content. The qualified observation subdomain corresponding to the target video stream is retained as the available observation range. The non-replaceable observation content is recorded as the non-replaceable range associated with the qualified observation subdomain. After completing the corresponding processing of each target video stream, the set of non-replaceable observation content is obtained.

[0078] Read the qualified observation subdomain G01 and the joint alternative observation area J01 of the target video stream VS-C01. G01 has an area of ​​14.0 m², and J01 has an area of ​​10.0 m². The horizontal boundary of J01 (25.0 m to 30.0 m) and the vertical boundary (2.0 m to 4.0 m) are completely within the horizontal boundary of G01 (23.0 m to 30.0 m) and the vertical boundary (2.0 m to 4.0 m). The area of ​​the non-alternative observation content is calculated as follows: Calculate, where, This represents the area of ​​the non-replaceable observation content. Indicates the area of ​​the qualified observation subdomain. Indicate the area of ​​the joint alternative observation region, substitute into , get G01 minus J01 retains N01. The boundary of N01 is 23.0m to 25.0m laterally and 2.0m to 4.0m longitudinally. The area verification is as follows: N01 is registered as an irreplaceable observation content of VS-C01, while G01 is retained as the complete usable observation range of VS-C01. The association between G01 and N01 is established according to VS-C01, A01 and the corresponding timestamp. If there is no joint alternative observation area that meets the corresponding security observation constraints and time synchronization conditions for the target video stream, the joint alternative observation area is set to empty and the complete spatial range of the qualified observation subdomain corresponding to the target video stream is registered as an irreplaceable observation content. The set of irreplaceable observation content is obtained by summarizing the target video stream identifier, security observation unit identifier, alternative video stream identifier, qualified observation subdomain, irreplaceable observation content and timestamp.

[0079] Example 4, please refer to Figure 5 In this embodiment, based on the set of irreplaceable observation content, the qualified observation range and the irreplaceable observation range of the target video stream are compared to determine whether the target video stream needs to be transmitted and whether the complete picture or a partial picture needs to be transmitted.

[0080] Based on the set of irreplaceable observation content, all qualified observation subdomains and irreplaceable observation content are summarized according to the target video stream identifier. Qualified observation subdomains associated with the same target video stream are grouped according to spatial boundaries, and the corresponding irreplaceable observation content is grouped according to its subdomain. The corresponding positions of the two types of areas in the unified field spatial coordinate system are determined, and qualified observation range and irreplaceable observation range are formed respectively, thus obtaining the correspondence of observation ranges.

[0081] The qualified observation subdomain records and the set of irreplaceable observation content are read according to the target video stream identifier. VS-C01 corresponds to G01 boundaries of 23.0m to 30.0m horizontally and 2.0m to 4.0m vertically, and corresponds to N01 boundaries of 23.0m to 25.0m horizontally and 2.0m to 4.0m vertically. Qualified observation subdomains with common boundaries within the same video stream, the same security observation unit, and the same time window are merged. When the length of the common boundary is greater than 0m, they are treated as contiguous subdomains. They are added directly only when there is no area overlap between subdomains. When there is area overlap, the overlapping area is calculated first and then duplicates are removed. The area of ​​the qualified observation range is calculated according to... Calculate, where, This represents the area after merging qualified observation subdomains with no area overlap. Indicates the first Area of ​​each qualified observation subdomain Indicates the number of subdomains participating in the merge, in VS-C01 , ,get The area of ​​the irreplaceable observation range is calculated according to Calculate, where, This represents the area after merging non-overlapping and non-substitutable regions. Indicates the first An irreplaceable area area Indicates the number of non-substitutable regions in VS-C01. , ,get Write F01, Q01, VS-C01, A01, G01, N01, timestamp, and task version number into the observation range correspondence record.

[0082] Based on the correspondence of observation ranges, the qualified observation range and the irreplaceable observation range corresponding to each target video stream are extracted. The spatial inclusion relationship between the irreplaceable observation range and the qualified observation range is determined. When the irreplaceable observation range is empty, there is no necessary transmission area. When it is in the local area of ​​the qualified observation range, it corresponds to a local necessary transmission area. When the spatial ranges of the two are consistent, it corresponds to a complete necessary transmission area, thus obtaining the necessary transmission area type.

[0083] Read the qualified observation range F01 and the non-replaceable observation range Q01 of VS-C01. The area of ​​F01 is 14.0 m², and the area of ​​Q01 is 4.0 m². The lateral boundary of F01 is 23.0 m to 30.0 m, and the longitudinal boundary is 2.0 m to 4.0 m. The lateral boundary of Q01 is 23.0 m to 25.0 m, and the longitudinal boundary is 2.0 m to 4.0 m. Check the boundaries of Q01 item by item. The lateral starting point of Q01 (23.0 m) is not less than the lateral starting point of F01 (23.0 m). The lateral ending point of Q01 (25.0 m) is not greater than the lateral ending point of F01 (30.0 m). The longitudinal starting point of Q01 (2.0 m) is not less than the longitudinal starting point of F01 (2.0 m). The longitudinal ending point of Q01 (4.0 m) is not greater than the longitudinal ending point of F01 (4.0 m). Confirm that Q01 is located inside F01. The area ratio is as follows: Calculate, where, This indicates the proportion of the area of ​​the non-replaceable observation range to the area of ​​the qualified observation range. This represents the area of ​​the non-replaceable observation range. This represents the area of ​​the qualified observation range, and we obtain... ,when When there is no necessary transmission area, And when the irreplaceable range is completely within the qualified observation range, it corresponds to the local necessary transmission area, when Furthermore, when the boundaries of the two are consistent, they correspond to a complete necessary transmission region. In this example, VS-C01 is registered as a locally necessary transmission region. The non-substitutable range and the qualified observation range of VS-C02 are both 8.0m² and have consistent boundaries, so it is registered as a complete necessary transmission region. VS-C03 has not formed a qualified observation subdomain and the non-substitutable range is empty, so it is registered as a region without necessary transmission regions, thus obtaining the necessary transmission region type.

[0084] Based on the necessary transmission area type, the mapping relationship between the video frame coordinate system and the unified on-site spatial coordinate system is invoked. The irreplaceable observation ranges corresponding to the local necessary transmission area and the complete necessary transmission area are back-mapped to the video frame coordinate system, associated with the corresponding security observation unit, and real-time security video streams that have not formed a qualified observation subdomain are registered as having no necessary transmission area. The necessary transmission areas and area types of each video stream are sorted out to obtain the necessary transmission configuration results of the video stream.

[0085] The calibration mapping relationship between the VS-C01 video frame coordinate system and the unified on-site spatial coordinate system is read. This mapping relationship is established by the pixel coordinates of the ground calibration points and the on-site coordinates in paragraph 1. The four spatial corner points of N01 are reverse-mapped to the video frame coordinates respectively. When the reverse mapping result has decimal pixel coordinates, the left and top boundaries are rounded down, and the right and bottom boundaries are rounded up. When the mapped coordinates are less than 0, they are taken as 0. When the mapped coordinates are greater than the upper limit of the screen width or height, the upper limit of the boundary is taken. The pixel coordinate envelope boundary of the four corner points of N01 is the upper left. Bottom right The width of the screen area is based on Calculate, where, This indicates the width of the region, in pixels. This represents the pixel value at the left horizontal coordinate. Representing the pixel value at the right horizontal coordinate, we obtain... Pixels, screen area height by Calculate, where, Indicates the height of the region, in pixels. This represents the pixel value at the top vertical coordinate. Representing the pixel value of the lower vertical coordinate, we obtain Pixels, the number of pixels in the image area by Calculation, obtained Pixels, the region boundaries are aligned outwards to the minimum coding block size of 16 pixels set by the video encoder, while the upper left boundary remains the same. The lower right boundary has been adjusted to The adjusted area has a width of 80 pixels, a height of 160 pixels, and a pixel count of 12800 pixels. The aligned area is then associated with VS-C01, A01, and necessary local transmission area types, and the entire VS-C02 image is captured. to Associate A02 with the complete necessary transmission area type, set the VS-C03 necessary transmission area to empty and register it as having no necessary transmission area, and obtain the necessary transmission configuration result for the video stream.

[0086] Example 5, please refer to Figure 6Based on the necessary transmission configuration results of the video streams, this embodiment controls the actual output status of each real-time security video stream to form the final real-time filtering result of the central control room security video streams:

[0087] Based on the necessary transmission area type of each real-time security video stream in the necessary transmission configuration result, extract the video stream identifiers corresponding to the complete necessary transmission area, the partial necessary transmission area and the non-necessary transmission area. Determine the output category of each video stream according to the necessary transmission area type, configure the corresponding video stream to continuous transmission state, partial area transmission state and edge filtering state respectively, and associate it with the corresponding edge node to obtain the video stream output state.

[0088] Read the region type from the necessary transmission configuration results of the video stream. A complete necessary transmission region corresponds to continuous transmission status, a partial necessary transmission region corresponds to a partial region transmission status, and no necessary transmission region corresponds to an edge filtering status. Configure VS-C02 to continuous transmission status, VS-C01 to partial region transmission status, and VS-C03 to edge filtering status. Read the video stream and edge node binding relationship from the network access ledger. VS-C01 and VS-C02 belong to EN01, and VS-C03 belongs to EN02. The number of output states is as follows: Statistics show that, This indicates the total number of video streams participating in the screening. Indicates the number of video streams in continuous transmission. Indicates the number of video streams in the local area transmission status. This indicates the number of video streams in the edge filtering state, all in channels. In this example... road, road, The road, obtained The number of channels is compared with the number of input video streams (3 channels). If they are equal, VS-C02 is associated with EN01, the complete screen boundary, A02, and the continuous transmission status. VS-C01 is associated with EN01 and the screen boundary after coded block alignment. to The video stream output status is obtained by associating A01 and local area transmission status with VS-C03, EN02, empty screen boundary, empty security observation unit and edge filtering status.

[0089] Based on the video stream output status, call the real-time security video streams that are in continuous transmission status and local area transmission status, extract the corresponding necessary transmission areas and associated security observation units, determine the output range of each video stream, continuously transmit the video stream in continuous transmission status to the central control room as a complete video screen, and perform regional-level video output on the video stream in local area transmission status according to the necessary transmission areas to obtain the video stream transmission configuration.

[0090] Invoke VS-C02 in continuous transmission mode and VS-C01 in local area transmission mode, read the full-frame resolution of 1920 pixels by 1080 pixels and the frame rate of 25 frames per second output by the VS-C02 encoder, and calculate the number of pixels per frame of the full-frame output. Calculate, where, This indicates the number of pixels in a single frame of a complete image. Indicates the width of the screen. This represents the image height, with all units in pixels. Pixels, read the VS-C01 region after coded block alignment, width 80 pixels, height 160 pixels, local region single frame pixel count according to Calculate, where, This represents the number of pixels in a single frame within a local region. Pixels, local area pixel ratio by Calculate, where, This represents the proportion of pixels in a local area to the total number of pixels in the entire image. This ratio is used to characterize the pixel scale of a region and is not used as the actual bitrate ratio. The actual transmission bitrate is obtained from the number of bytes output by the encoder. Within a 1-second statistical window, the number of output bytes is multiplied by 8 to obtain the bitrate. The number of pixels processed per second for a complete image is calculated as follows: Calculate, where, This represents the number of pixels processed per second for the entire image, in units of... , Represents frame rate, in units of ,get The number of pixels processed per second in a local area is calculated according to... Calculate, where, This represents the number of pixels processed per second in a local area. Configure the continuous transmission of the complete VS-C02 screen and the VS-C01 screen area. to The area transmission configuration is written into the video stream transmission configuration of EN01.

[0091] Based on the video stream transmission configuration, the real-time security video stream in the edge filtering state of the video stream output status is called, the corresponding video stream is stopped from continuously outputting to the central control room, and the corresponding camera is kept performing local video acquisition and AI image analysis on the edge node side. According to the video stream identifier associated with the output status, necessary transmission area and corresponding security observation unit, the output relationship of each edge node is sorted out to obtain the real-time filtering result of the security video stream in the central control room.

[0092] The system reads the VS-C03 and its associated edge node EN02 corresponding to the edge filtering status. It reads the original image resolution of VS-C03 (1920 x 1080 pixels, 25 frames per second). The control room video frame output flag is written to 0, and the edge node's local video acquisition flag and image analysis flag are written to 1. The video frame is written to the circular buffer via the EN02 decoding module. The circular buffer duration is set to 60 seconds, and the buffer capacity is calculated to be 1500 frames based on 25 frames per second and 60 seconds. The target bounding box, occlusion mask, sharpness, and illuminance records are generated by the edge node's image analysis unit and digital illuminance meter interface. For the 08:30:15 video frame, the target presentation status is read as 1, occlusion ratio as 0.17, sharpness as 0.79, and illuminance as 141 lx. The frame analysis record is associated with VS-C03, EN02, timestamp, and spatial evaluation unit identifier. The number of frames processed locally by the edge node is calculated according to... Calculate, where, This indicates the number of video frames entering the edge analysis link within the statistical period, expressed in frames. Indicates video frame rate, in units of , This represents the statistical duration, in seconds (s). Substitute... , get When an edge node re-forms a qualified observation subdomain, the region type and output status are regenerated according to paragraphs 6 to 13. When communication between the edge node and the central control room is restored and the output status is continuous transmission or local area transmission, output is started according to the latest status configuration. When an edge node is abnormal, the last valid output status, timestamp, and cache index are written into the fault record. The results are sorted by edge node identifier, video stream identifier, output status, screen output range, necessary transmission region type, security observation unit identifier, and timestamp to form the real-time filtering results of the security video stream in the central control room.

[0093] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A real-time filtering method for security video streams in a central control room based on AI edge computing, characterized in that, Includes the following steps: Based on real-time security video streams, camera on-site calibration information, and rail transit site area information, a mapping relationship between the video image coordinate system and a unified on-site spatial coordinate system is established. The theoretical spatial coverage area of ​​each real-time security video stream is determined and divided into spatial evaluation units. Based on the spatial evaluation units, video frame image analysis results are obtained. According to the camera on-site calibration information and the mapping relationship, the observation capability parameters corresponding to each spatial evaluation unit are determined. The observation capability parameters include ground pixel density, line-of-sight angle, occlusion ratio, sharpness, and illumination. The observation capability parameters corresponding to each spatial evaluation unit are judged item by item against the preset basic observation validity conditions. The current effective observation subdomains are then selected and merged. The observation capability parameters corresponding to each current effective observation subdomain and the real-time security video stream affiliation information are called. According to the camera identifier, the current effective observation subdomains are associated with the corresponding real-time security video streams. The spatial boundaries, observation status, and observation capability parameters of each subdomain are assigned to the corresponding video stream association items to obtain the current effective observation characterization set of the video stream. Based on the security task configuration data, security observation units are divided and the security observation constraints of the central control room are determined. The current effective observation subdomain in the current effective observation representation set of the video stream is spatially matched with the security observation units. The overlapping area is determined as the spatial overlap area, and the spatial range corresponding to the spatial overlap area is determined as the actual coverage range of the current effective observation subdomain to the corresponding security observation unit. At the same time, the corresponding observation capability parameters are associated to obtain the spatial coverage association relationship. The security observation constraints of the central control room are called to judge the qualification of the actual coverage range and observation capability parameters corresponding to each spatial overlap area. The spatial overlap area that meets the corresponding security monitoring requirements is determined as the qualified observation subdomain, and the qualified observation association result of the video stream containing the qualified observation subdomain is obtained. Based on the qualified observation association results of the video streams, real-time security video streams with qualified observation subdomains are screened, and the qualified observation subdomains corresponding to each video stream are determined as the available observation range of the corresponding video stream, forming an available alternative observation set. Then, video streams with qualified observation subdomains are selected as target video streams in sequence. For the security observation unit associated with the target video stream, available observation ranges corresponding to other video streams that spatially overlap with the qualified observation subdomains corresponding to the target video stream are extracted from the available alternative observation set. The extraction results are spatially combined and associated with corresponding observation capability parameters to determine the joint observation capability. Based on the security observation constraints of the central control room, the joint alternative observation area that meets the requirements is determined. The remaining part of the qualified observation subdomain after deducting the joint alternative observation area is determined as the non-replaceable observation content, resulting in a set of non-replaceable observation content. Based on the set of irreplaceable observation content, all qualified observation subdomains and irreplaceable observation content are summarized according to the target video stream identifier, forming qualified observation ranges and irreplaceable observation ranges respectively. The spatial inclusion relationship between the irreplaceable observation range and the qualified observation range is determined. When the irreplaceable observation range is empty, it is determined that there is no necessary transmission area. When the irreplaceable observation range is in a local area of ​​the qualified observation range, it is determined that it is a locally necessary transmission area. When the spatial range of the irreplaceable observation range is consistent with that of the qualified observation range, it is determined that it is a complete necessary transmission area. According to the mapping relationship, the irreplaceable observation ranges corresponding to the locally necessary transmission area and the complete necessary transmission area are reverse-mapped to the video screen coordinate system to obtain the corresponding necessary transmission area and associated security observation unit, thus obtaining the necessary transmission configuration result of the video stream. Based on the necessary transmission configuration results of the video stream, configure continuous transmission state, local area transmission state, or edge filtering state to obtain the real-time filtering results of the security video stream in the central control room.

2. The real-time filtering method for central control room security video streams based on AI edge computing according to claim 1, characterized in that, The acquisition of the current valid observation representation set of the video stream includes: Based on the real-time security video streams generated by the cameras connected to each edge node, the on-site calibration information of the cameras, and the on-site area information of the rail transit, the video image position, the camera installation pose, and the boundary of the on-site area are extracted. The video image position is then mapped to a unified on-site spatial coordinate system to determine the theoretical spatial coverage area of ​​each real-time security video stream. Spatial evaluation units are formed by intersecting the theoretical spatial coverage boundary with a horizontal grid of 10m and a vertical grid of 2m. The spatial boundary and video stream affiliation of each spatial evaluation unit are recorded.

3. The real-time filtering method for central control room security video streams based on AI edge computing according to claim 2, characterized in that, The acquisition of the currently valid observation subdomain also includes: Based on the spatial evaluation unit, the output video frame image analysis results of each edge node are obtained, and the target presentation state, screen occlusion state, image clarity state and on-site imaging state are mapped to each spatial evaluation unit. According to the on-site calibration information of the camera and the mapping relationship, the corresponding observation capability parameters are sorted out. The observation capability parameters corresponding to each spatial evaluation unit are judged item by item with the preset basic observation validity conditions, and the valid spatial evaluation units are screened and merged according to the spatial adjacency relationship to obtain the current valid observation subdomain.

4. The real-time filtering method for central control room security video streams based on AI edge computing according to claim 1, characterized in that, The determination of the security observation constraints in the central control room includes: Based on the monitoring areas, security monitoring requirements, key equipment locations, and key security boundaries in the current security task configuration data of the central control room, the monitoring areas are spatially divided according to the distribution of key equipment locations, key security boundaries, and security monitoring requirements. Each divided area is then matched with the security monitoring requirements, and the spatial range corresponding to each divided area is used as the monitoring range of the corresponding security observation unit. Observation conditions are determined in conjunction with the corresponding security monitoring requirements to form multiple security observation units. Each security observation unit is then associated with its corresponding security monitoring requirements, and the security observation constraints of the central control room are determined based on the security monitoring requirements corresponding to each security observation unit.

5. The real-time filtering method for security video streams in a central control room based on AI edge computing according to claim 4, characterized in that, The acquisition of the video stream qualified observation correlation results includes: The system calls upon the currently valid observation subdomain, spatial range, and observation capability parameters from the currently valid observation representation set of the video stream, spatially matches the boundary of the currently valid observation subdomain with the boundary of the security observation unit, associates the corresponding qualified observation subdomain according to the real-time security video stream identifier, and records the real-time security video streams that have not formed qualified observation subdomains, thereby obtaining the qualified observation association results of the video stream.

6. The real-time filtering method for security video streams in a central control room based on AI edge computing according to claim 1, characterized in that, The acquisition of the set of irreplaceable observation content includes: When there is no joint alternative observation area that meets the requirements, the joint alternative observation area is determined to be empty. The remaining part of the qualified observation subdomain corresponding to the target video stream after deducting the joint alternative observation area is determined as the non-substitutable observation content. The qualified observation subdomain and the non-substitutable observation content are associated according to the target video stream identifier. After processing each target video stream, a set of non-substitutable observation content is obtained.

7. The real-time filtering method for security video streams in a central control room based on AI edge computing according to claim 1, characterized in that, The acquisition of the necessary transmission configuration results for the video stream includes: Real-time security video streams that do not form qualified observation subdomains are identified as areas where there is no necessary transmission, thus obtaining the necessary transmission configuration results for the video streams.

8. The real-time filtering method for security video streams in a central control room based on AI edge computing according to claim 1, characterized in that, The acquisition of the real-time filtering results of the security video stream in the central control room includes: Based on the necessary transmission area type of each real-time security video stream in the necessary transmission configuration result of the video stream, extract the video stream identifiers corresponding to the complete necessary transmission area, the partial necessary transmission area and the non-existent necessary transmission area, configure the corresponding real-time security video stream to the continuous transmission state, the partial area transmission state and the edge filtering state respectively, and associate it with the edge node to obtain the video stream output state. The real-time security video streams in the continuous transmission state and the local area transmission state are invoked, the corresponding necessary transmission areas and associated security observation units are extracted, the real-time security video stream in the continuous transmission state is continuously transmitted to the central control room as a complete video frame, and the real-time security video stream in the local area transmission state is executed with regional-level video output according to the necessary transmission areas to obtain the video stream transmission configuration.

9. The real-time filtering method for security video streams in a central control room based on AI edge computing according to claim 8, characterized in that, The acquisition of the real-time filtering results of the security video stream in the central control room also includes: According to the video stream transmission configuration, the real-time security video stream that is in the edge filtering state in the video stream output state is called, and the corresponding real-time security video stream is stopped from continuously outputting to the central control room. At the same time, the corresponding camera continues to perform local video acquisition and AI image analysis on the edge node side. By associating the video stream output status, the necessary transmission area, and the corresponding security observation unit with the real-time security video stream identifier, the output relationship of each edge node is organized to obtain the real-time filtering result of the security video stream in the central control room.

Images