Machine room inspection and supervision method and system based on track coverage

Abstract

The invention provides a machine room inspection and supervision method and system based on track coverage. The method comprises the following steps: acquiring real-time video streams of a plurality of monitoring cameras deployed in a machine room inspection area; analyzing each frame of image of the real-time video stream by using a pre-trained visual target detection model, identifying an inspector target in the image, and determining a detection frame and pixel coordinates of the inspector target in an image coordinate system; executing a target position mapping step, and converting pixel coordinates of the inspector target in different monitoring camera images into plane graph coordinates on a machine room electronic plane graph based on a predefined coordinate mapping relation; a multi-view track merging step is executed, space-time fusion is carried out on plane graph coordinates from different monitoring cameras, and a continuous moving track of the inspection personnel on the electronic plane graph is generated; and executing inspection supervision based on the continuous moving track, and judging whether the inspection work of the inspector meets a preset requirement or not.

Description

Technical Field

[0001] This invention belongs to the field of data center inspection and supervision technology, specifically relating to a data center inspection and supervision method and system based on trajectory coverage. Background Technology

[0002] In data center inspections, ensuring that inspectors complete their tasks according to regulations is crucial for the safe and stable operation of the data center. However, currently, there is a lack of effective means to supervise the work of inspectors within the data center, making it difficult to ensure whether they are truly carrying out their inspection duties and whether they are spending sufficient time in front of each rack. This poses a challenge to the operation and maintenance management of the data center.

[0003] Currently, the main technologies for supervising data center inspectors include: First, using AI recognition technology combined with infrared cameras to determine whether members are present in the data center, and using access control systems to record their entry and exit times to infer inspection time; second, using video playback to check whether inspectors have actually carried out inspection work; third, using GPS positioning within the data center to determine the location of inspectors; fourth, setting up patrol points and recording the locations visited by inspectors through check-in; and fifth, using indoor positioning technology based on Wi-Fi or Bluetooth to track the movement of inspectors.

[0004] However, these existing technologies all have certain limitations. Method 1 can only determine whether the inspection personnel are in the server room, but cannot monitor their specific movement path within the server room or whether they stayed in front of each rack for a sufficient amount of time. Method 2's video playback method requires a significant amount of manpower and time for review and analysis. In Method 3, GPS signals are severely attenuated or even completely ineffective indoors, underground, or in densely built-up areas, making effective positioning impossible. Method 4's patrol point check-in method can only record discrete location information and cannot form continuous, smooth movement trajectories. And Method 5, based on Wi-Fi or Bluetooth for precise indoor positioning technology, requires the pre-deployment of a large number of positioning base stations, making it extremely costly and difficult to apply on a large scale. Summary of the Invention

[0005] This invention proposes a data center inspection and supervision method and system based on trajectory coverage, which solves the problem that existing supervision technologies cannot comprehensively and efficiently supervise the movement trajectory and inspection behavior of inspection personnel in the data center.

[0006] To address the aforementioned technical problems, this invention provides a data center inspection and supervision method based on trajectory coverage, comprising the following steps:

[0007] Step S1: Acquire real-time video streams from multiple surveillance cameras in the computer room inspection area, identify the inspection personnel target in each frame of the real-time video stream, and determine the detection box and pixel coordinates of the inspection personnel target in the image coordinate system;

[0008] Step S2: Extract the minimum bounding rectangle of the detection box to obtain the spatial range of the pixel coordinates in the image coordinate system. Define the target mapping area corresponding to the spatial range in the computer room electronic floor plan. Calculate the scaling factor based on the geometric relationship between the spatial range and the target mapping area.

[0009] Step S3: Obtain the angle between the current optical axis orientation of the surveillance camera and the wall of the computer room as the rotation angle of the surveillance camera, obtain the angle between the rotation angle and the main axis direction of the electronic floor plan of the computer room as the angle difference, and convert the pixel coordinates of the inspection personnel target in different surveillance camera images into floor plan coordinates on the electronic floor plan of the computer room according to the scaling factor and the angle difference.

[0010] Step S4: Perform spatiotemporal fusion of the planar coordinates from different surveillance cameras to generate a continuous movement trajectory of the inspection personnel target on the electronic planar map. Based on the continuous movement trajectory, perform inspection and supervision to determine whether the inspection work of the inspection personnel target meets the preset requirements.

[0011] Preferably, step S2, which calculates the scaling factor based on the geometric relationship between the spatial range and the target mapping area, includes the following steps: calculating the ratio of the width of the target mapping area to the width of the spatial range, and the ratio of the height of the target mapping area to the height of the spatial range, to obtain the horizontal scaling factor and the vertical scaling factor.

[0012] Preferably, step S3, which converts the pixel coordinates of the inspection personnel target in different surveillance camera images into planar coordinates on the electronic floor plan of the computer room, includes the following steps: constructing a rotation transformation matrix based on the angle difference, constructing a scaling transformation relationship based on the scaling factor, applying the rotation transformation matrix and the scaling transformation relationship to the pixel coordinates in sequence to obtain preliminary mapped coordinates; calculating the translation vector of the preliminary mapped coordinates relative to the target mapped area, and using the translation vector to perform offset calibration on the preliminary mapped coordinates to obtain the final planar coordinates.

[0013] Preferably, step S1 includes the following steps:

[0014] Step S11: Obtain real-time video streams from multiple surveillance cameras within the computer room inspection area;

[0015] Step S12: Use a target detection model to identify the inspection personnel target from the real-time video stream;

[0016] Step S13: Use a tracking algorithm to perform inter-frame correlation on the inspection personnel target, and output the movement trajectory of the inspection personnel target and the corresponding detection box.

[0017] Preferably, the spatiotemporal fusion of planar coordinates from different surveillance cameras in step S4 includes the following steps:

[0018] Step S41: When the inspection personnel target enters the field of view of the monitoring camera, extract the feature vector according to the detection frame and generate a global identity ID for the inspection personnel target;

[0019] Step S42: When the inspection personnel target appears simultaneously in the overlapping area of ​​the field of view of multiple surveillance cameras, calculate the feature similarity between the detection boxes output by different surveillance cameras. If the feature similarity is greater than the set threshold, determine that the inspection personnel target in the field of view of different surveillance cameras is the same inspection personnel and inherit the global identity ID.

[0020] Step S43: In the area where the field of view intersects, calculate the weight of the monitoring camera based on the distance between the inspection personnel target and the different monitoring cameras. Use the weight to perform a weighted average calculation on the planar coordinates from different monitoring cameras to obtain the optimal coordinate point. Combine the optimal coordinate points at all times to obtain the continuous movement trajectory of the inspection personnel target on the electronic planar map.

[0021] Preferably, in step S43, in the field-of-view intersection area, when the inspection personnel target completely switches from the field of view of the first monitoring camera to the field of view of the second monitoring camera, the first planar coordinates of the inspection personnel target at the last visible moment in the first monitoring camera, and the second planar coordinates of the inspection personnel target in the second monitoring camera at the same moment are recorded. A linear interpolation algorithm is used to generate intermediate trajectory points between the first planar coordinates and the second planar coordinates, and a moving average filter is applied to smooth the continuous movement trajectory.

[0022] Preferably, in step S43, the weighted average of the planar coordinates from different surveillance cameras is calculated using the weights to obtain the expression for the optimal coordinate point:

[0023] ;

[0024] ;

[0025] In the formula, , These are the x and y coordinates of the optimal coordinate point, respectively. The total number of surveillance cameras when the target of the inspection personnel appears on multiple surveillance cameras at the same time; For the first The weight of each surveillance camera; , For the first The coordinates of a floor plan of a surveillance camera; For the inspection personnel's target and the first The distance between the surveillance cameras; It is a local minimum.

[0026] Preferably, step S4, determining whether the inspection work of the inspection personnel target meets the preset requirements, includes at least the following steps: determining whether the coordinates of the plan are located in the non-inspection area set on the electronic plan; if the coordinates of the plan are located in the non-inspection area, calculating the continuous stay time of the inspection personnel target in the non-inspection area; when the continuous stay time exceeds the set time threshold, triggering an illegal intrusion alarm and capturing image evidence.

[0027] Preferably, step S4, determining whether the inspection work of the inspection personnel target meets the preset requirements, includes at least the following steps: after the inspection task is completed, reading the coordinate extreme values ​​of the continuous movement trajectory in each inspection area, calculating the actual inspection route length, calculating the ratio of the actual inspection route length to the planned inspection route length, obtaining the trajectory coverage rate of each inspection area, and generating an alarm for non-compliance of inspection completion if the trajectory coverage rate is lower than the preset coverage rate threshold.

[0028] This invention also provides a data center inspection and supervision system based on trajectory coverage, which is implemented based on the above-mentioned data center inspection and supervision method based on trajectory coverage, including: a video stream acquisition module, a task management module, an access control linkage module, a target detection and tracking module, a multi-view trajectory fusion module, an inspection and supervision module, an alarm module, and an interaction and visualization module;

[0029] The video stream acquisition module acquires real-time video streams from multiple surveillance cameras within the computer room inspection area.

[0030] The task management module is responsible for creating, distributing, and storing inspection work orders, and defining planned inspection routes and key check-in areas.

[0031] The access control linkage module: connects to the computer room access control system to receive the facial recognition check-in information of the inspection personnel; when the inspection personnel pass the verification, a task start signal is triggered, and the identity of the inspection personnel is bound to the current inspection task document;

[0032] The target detection and tracking module identifies the patrol personnel in each frame of the real-time video stream, outputs the pixel coordinates of the detection box, generates a global identity ID for the patrol personnel, uses a multi-target tracking algorithm to track multiple targets, and maintains the consistency of the global identity ID.

[0033] The multi-view trajectory fusion module: delineates the physical coverage area of ​​each camera in the electronic planar map, calculates the scaling factor from the image coordinate system to the planar map coordinates, performs coordinate rotation correction based on the rotation angle difference, converts pixel coordinates into unified planar map coordinates, fuses multi-camera data, and generates a continuous global trajectory.

[0034] The inspection and supervision module monitors the matching degree between the movement trajectory of the inspection personnel and the preset standard inspection path and key checkpoints, and monitors the dwell time and trajectory coverage of the inspection personnel.

[0035] The alarm module triggers an alarm when the inspection personnel's dwell time exceeds a set time threshold or the trajectory coverage is lower than a preset coverage threshold.

[0036] The interaction and visualization module loads a vector map of the computer room and draws the movement trajectory points and connections of the inspection personnel in real time.

[0037] The beneficial effects of the present invention include at least the following:

[0038] 1. By calculating the minimum bounding rectangle and spatial range, the pixel coordinates in the camera can be accurately mapped to the actual physical space of the equipment room. The introduction of the scaling factor enables the system to dynamically adjust the mapping relationship according to the actual coverage of the camera, adapting to the installation position and angle of different cameras. Through precise geometric relationship calculation, high positioning accuracy is ensured, avoiding positioning deviations caused by differences in camera position and angle. At the same time, the angle difference between the camera rotation angle and the planar coordinate system is considered, and the accuracy of coordinate transformation is ensured through angle correction.

[0039] 2. It can process image data from different cameras and transform the pixel coordinates from different viewpoints into the same planar coordinate system, enabling multiple cameras to work together. This method has low requirements for the installation position and angle of the cameras, has strong adaptability, and can work stably in complex computer room environments.

[0040] 3. Through spatiotemporal fusion technology, the continuous movement trajectory of inspection personnel can be generated, which solves the problem that existing technologies can only record discrete points. The continuous trajectory can clearly show the movement path of the inspection personnel, including whether they have passed all the necessary inspection points and whether they have stayed in front of each cabinet for a sufficient amount of time, thus achieving comprehensive supervision. Based on the generated continuous trajectory, the system can automatically determine whether the inspection work meets the preset requirements, reducing manual intervention and improving the efficiency and accuracy of supervision. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of the inspection task distribution process according to an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram of the inspection trajectory fusion process based on multi-view vision in an embodiment of the present invention;

[0044] Figure 4 This is a schematic diagram of the system structure according to an embodiment of the present invention. Detailed Implementation

[0045] 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 protection scope of the present invention.

[0046] like Figure 1 As shown, this embodiment of the invention provides a data center inspection and supervision method based on trajectory coverage, including the following steps:

[0047] Step S1: Acquire real-time video streams from multiple surveillance cameras in the computer room inspection area, identify the inspection personnel target in each frame of the real-time video stream, and determine the detection box and pixel coordinates of the inspection personnel target in the image coordinate system.

[0048] Specifically, to ensure high-precision target detection even in complex lighting conditions and equipment obstructions within the computer room, this embodiment of the invention preferably employs a high-definition network camera that supports an AI open platform. To achieve continuous trajectory tracking across cameras, the fields of view of adjacent cameras must maintain at least 15%-20% overlap. The downward angle of rotation is controlled between 15° and 45°.

[0049] Meanwhile, intelligent access control terminals with facial recognition capabilities are deployed at the main entrance and exit of the computer room. This device not only controls physical access but also serves as a trigger for inspection tasks. When inspection personnel pass through by scanning their faces, the access control system needs to push structured data containing personnel ID, name, captured photo, and timestamp in real time; for inspection robots, they are equipped with positioning modules and identification tags.

[0050] like Figure 2 As shown, the inspection process in this embodiment of the invention begins with task-driven and precise binding of personnel identity, and the specific steps are as follows:

[0051] S:01: First, create an inspection task ticket. This ticket contains a unique task ID, the identity information of the designated inspection personnel, the planned inspection area, the planned inspection route drawn based on the computer room floor plan, and the planned start and end times of the task.

[0052] S102: The facial photos and other identification information of the inspection personnel associated with the task ticket are sent to the computer room access control system in advance to complete the authorization of this inspection task.

[0053] S103: When inspection personnel arrive at the computer room entrance, they authenticate their identity by facial recognition. Once the access control system verifies their identity, it not only opens the door to allow entry but also sends a task start signal to the business application server. This signal binds the inspection personnel's identity to a specific task ticket and activates the entire trajectory recording system, initiating subsequent trajectory tracking and supervision.

[0054] The core of this invention lies in utilizing existing surveillance cameras to generate and visualize the continuous trajectories of inspection personnel through computer vision technology. The latest object detection algorithm, YOLO11, is employed for personnel detection, and to adapt to the specific environment of the computer room, incremental training of YOLO11 is performed. Approximately 5000 images of the computer room are collected, covering different lighting conditions, viewing angles, and degrees of occlusion. Inspection personnel wearing standard anti-static work clothes and potentially misdetected objects are labeled on these images to construct an incremental training dataset. During training, Mosaic and Mixup data augmentation strategies are enabled to simulate complex scenarios in the computer room, such as dense personnel or overlapping with server racks, thereby improving the robustness of the model.

[0055] The incrementally trained YOLO11 is used to identify the inspection personnel target in each frame of a real-time video stream, and the detection box and pixel coordinates of the inspection personnel target in the image coordinate system are determined.

[0056] Step S2: Extract the minimum bounding rectangle of the detection box to obtain the spatial range of pixel coordinates in the image coordinate system. Define the target mapping area corresponding to the spatial range in the computer room electronic floor plan. Calculate the scaling factor based on the geometric relationship between the spatial range and the target mapping area.

[0057] Step S3: Obtain the angle between the current optical axis orientation of the surveillance camera and the wall of the computer room as the rotation angle of the surveillance camera. Obtain the angle between the rotation angle and the main axis direction of the electronic floor plan of the computer room as the angle difference. Based on the scaling factor and the angle difference, convert the pixel coordinates of the inspection personnel's target in different surveillance camera images into floor plan coordinates on the electronic floor plan of the computer room.

[0058] Specifically, steps S2 and S3 fuse the scattered visual observation data into a unified trajectory based on the computer room floor plan. For example... Figure 3 As shown, it includes the following steps:

[0059] S1: Analyze the set of detection boxes generated by each surveillance camera within a certain period of time, and calculate the minimum bounding rectangle of the point set in the image coordinate system. This rectangle represents the range of image pixels that the surveillance camera can effectively detect human activity under the current rotation angle and focal length.

[0060] S2: Import the clear server room floor plan generated based on accurate CAD drawings into the system. During the configuration phase, the administrator defines a corresponding logical area polygon on the electronic floor plan based on the actual physical installation location and coverage of each camera, which is recorded as the target mapping area.

[0061] S3: Calculate the anisotropic scaling factor based on the geometric size ratio of the minimum bounding rectangle and the target mapping region to solve the image distortion problem caused by camera tilt shooting.

[0062] Horizontal scaling factor: ;

[0063] Vertical scaling factor: ;

[0064] In the formula, , These are the horizontal scaling factor and the vertical scaling factor, respectively. , These represent the width and height of the image's pixel range, respectively. , These represent the width and height of the target mapping region, respectively.

[0065] S4: During the inspection, the system obtains the rotation angle of each monitoring camera relative to the wall in real time through the camera backend. This parameter reflects the angle between the camera's current optical axis orientation and the physical coordinate system of the equipment room. For the pixel coordinates in each frame of the image output in step S1, rotation, scaling, and translation transformations are applied sequentially to convert them into planar coordinates. Specifically, the following steps are included:

[0066] Construct a rotation transformation matrix based on the angle difference, construct another rotation transformation matrix based on the scaling factor, and apply the rotation transformation matrix and scaling transformation relationship to the pixel coordinates sequentially to obtain the initial mapped coordinates:

[0067] ;

[0068] In the formula, , Initial mapping coordinates; , These are pixel coordinates.

[0069] Calculate the translation vector of the initial mapping coordinates relative to the target mapping region, and use the translation vector to calibrate the initial mapping coordinates by offset to obtain the final planar coordinates.

[0070] Through the above steps, the originally isolated video pixels with different perspectives are uniformly transformed into planar coordinate points with actual physical meaning.

[0071] Step S4: Perform spatiotemporal fusion of the plan coordinates from different surveillance cameras to generate a continuous movement trajectory of the inspection personnel on the electronic plan. Based on the continuous movement trajectory, perform inspection and supervision to determine whether the inspection work of the inspection personnel meets the preset requirements.

[0072] Specifically, when the server room is large, inspection personnel will span the field of view of multiple cameras. This embodiment of the invention uses the deep learning-based target tracking algorithm BoT-SORT to associate all single-frame detection boxes into a continuous time-series trajectory, maintaining the consistency of the inspection personnel's identity. The steps include:

[0073] Step S41: When the inspection personnel enter the field of view of the surveillance camera, extract the pedestrian re-identification (ReID) feature vector of the inspection personnel based on the detection frame, and generate a global identity ID for the inspection personnel.

[0074] Step S42: When the inspection personnel target appears in the overlapping area of ​​the field of view of multiple surveillance cameras at the same time, calculate the feature similarity between the detection boxes output by different surveillance cameras. If the feature similarity is greater than the set threshold, it is determined that the inspection personnel target in the field of view of different surveillance cameras is the same inspection personnel, and the global identity ID is inherited.

[0075] Step S43: In the overlapping field of view area, when the inspection personnel's target completely switches from the field of view of the first surveillance camera to the field of view of the second surveillance camera, record the first planar coordinates of the inspection personnel's target at the last visible moment in the first surveillance camera, and the second planar coordinates of the inspection personnel's target in the second surveillance camera at the same moment. Use a linear interpolation algorithm to generate an intermediate trajectory point between the first and second planar coordinates, and calculate the weights based on the distance of the inspection personnel's target from different surveillance cameras. Use the weights to calculate a weighted average of the planar coordinates from different surveillance cameras to obtain the optimal coordinate point.

[0076] ;

[0077] ;

[0078] In the formula, , These are the x and y coordinates of the optimal coordinate point, respectively. The total number of surveillance cameras when the target of the inspection personnel appears on multiple surveillance cameras at the same time; For the first The weight of each surveillance camera; , For the first The coordinates of a floor plan of a surveillance camera; For the inspection personnel's target and the first The distance between the surveillance cameras; It is a local minimum.

[0079] S44: For blind spots, linear interpolation is used to fill in the breakpoints, and a moving average filtering algorithm is used to smooth the trajectory, generating a continuous and natural movement curve. The optimal coordinates at all times are collected to obtain the continuous movement trajectory of the inspection personnel's target on the electronic plan view.

[0080] like Figure 2 As shown, during the inspection process, it is determined in real time whether the coordinates of the inspection personnel on the floor plan are located in the non-inspection area set on the electronic floor plan. If the coordinates on the floor plan are located in the non-inspection area, the continuous dwell time of the inspection personnel in the non-inspection area is calculated. When the continuous dwell time exceeds the set time threshold, an illegal intrusion alarm is triggered and image evidence is captured.

[0081] For critical equipment, a minimum dwell time threshold has been set. The total time that inspection personnel spent in this area ,like Furthermore, since the personnel have left the area, an alarm for perfunctory inspection is generated and recorded as an abnormal event.

[0082] At the same time, a restricted area is set up. Once the coordinates of the patrol personnel on the floor plan fall into the restricted area polygon, the sound and light alarm will be triggered immediately, and an alarm will pop up in the management terminal. At the same time, the current video frame will be captured and saved as evidence.

[0083] After the inspection task is completed, post-processing analysis is performed to generate a quantitative report. The extreme coordinate values ​​of the continuous movement trajectory of the inspection personnel within each inspection area are read to calculate the actual inspection route length. The extreme coordinate values ​​of the planned inspection route, set based on the electronic floor plan, are also read to calculate the planned inspection route length. The ratio of the actual inspection route length to the planned inspection route length is calculated to obtain the trajectory coverage rate of each inspection area. If the trajectory coverage rate is lower than a preset coverage threshold, an alarm indicating that the inspection completion rate is not up to standard is generated.

[0084] like Figure 4 As shown, this embodiment of the invention also provides a data center inspection and supervision system based on trajectory coverage, which is implemented based on the above-mentioned data center inspection and supervision method based on trajectory coverage. It includes: a video stream acquisition module, a task management module, an access control linkage module, a target detection and tracking module, a multi-view trajectory fusion module, an inspection and supervision module, an alarm module, and an interaction and visualization module.

[0085] Video stream acquisition module: Acquires real-time video streams from multiple surveillance cameras within the computer room inspection area.

[0086] Task Management Module: Responsible for creating, distributing, and storing inspection work orders, and defining planned inspection routes and key check-in areas.

[0087] Access control linkage module: Connects to the computer room access control system to receive personnel's facial recognition check-in information. When a person passes verification, a task start signal is triggered, and the person's identity is bound to the current inspection task document.

[0088] Target detection and tracking module: Identifies patrol personnel in each frame of the real-time video stream, outputs the pixel coordinates of the detection box, generates a global identity ID for the patrol personnel, and uses a multi-target tracking algorithm to track multiple targets while maintaining the consistency of the global identity ID.

[0089] Multi-view trajectory fusion module: Delineates the physical coverage area of ​​each camera on the electronic planar map, calculates the scaling factor from the image coordinate system to the planar map coordinates, performs coordinate rotation correction based on the rotation angle difference, converts pixel coordinates into unified planar map coordinates, fuses data from multiple cameras, and generates a continuous global trajectory.

[0090] Inspection and Supervision Module: Based on the preset standard inspection path and key checkpoints, monitor the matching degree between the movement trajectory of the inspection personnel and the preset path, and monitor the dwell time and trajectory coverage of the inspection personnel.

[0091] Alarm module: When the inspection personnel stay for more than the set time threshold, or the trajectory coverage is lower than the preset coverage threshold, an alarm is triggered.

[0092] Interaction and visualization module: Load the data center vector map on the web or APP, draw personnel movement trajectory points and connections in real time, and support clicking on any point on the trajectory to retrieve the corresponding monitoring video recording at that time and location, realizing trajectory-video linkage.

[0093] The method and system of the present invention have the following advantages:

[0094] (1) It significantly reduced deployment costs and hardware dependence, and improved system applicability.

[0095] Without requiring the deployment of expensive dedicated positioning hardware such as GPS modules, Bluetooth beacons, or Wi-Fi positioning base stations, this solution creatively utilizes existing, widely deployed video surveillance systems as the data source. This significantly reduces the hardware costs and installation complexity of the system, enabling rapid deployment and application in complex environments such as indoors, underground, and factories where GPS signals are unavailable, greatly expanding the applicability of the inspection trajectory management system.

[0096] (2) It generates continuous, intuitive and high-precision inspection trajectories, which improves the intuitiveness and authenticity of supervision.

[0097] Compared to the discrete location information generated by traditional patrol point check-ins, the solution in this invention continuously maps the personnel positions detected in video frames onto a planar map, generating a smooth and continuous trajectory line that highly matches the actual movement path of the patrol personnel. This allows managers to clearly grasp the complete movement route, coverage area, and dwell time of the patrol personnel at a glance, greatly enhancing the intuitiveness of supervision and the ability to accurately reproduce the patrol process.

[0098] (3) It has achieved deep integration of video system and location information, and improved the linkage efficiency and intelligence level of security management.

[0099] This fundamentally breaks down the information silos between video surveillance systems and location management systems. By directly linking trajectory points with video frames, managers can directly trigger and replay live video from any location and at any time point on the floor plan. This makes tracing abnormal events more efficient and convenient than ever before, achieving an intelligent upgrade from "seeing" to "understanding." For example, when an alarm occurs at a certain location, it is possible to quickly see whose trajectory passed through that area at that time.

[0100] (4) It improves the system’s flexibility and scalability.

[0101] Based on its deep learning model-driven object detection capabilities, the system can easily adapt to new inspector uniforms, newly added inspection equipment, or changing environmental backgrounds through incremental training. Furthermore, through multi-camera trajectory fusion technology, the system can flexibly scale to the management of large-scale, complex areas without altering its core architecture. This flexibility and scalability are unmatched by traditional technologies that rely on fixed hardware deployments.

[0102] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described; only preferred embodiments of the present invention are illustrated. The descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. As long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0103] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this invention should be determined by the appended claims.

Claims

1. A method for inspecting and supervising computer rooms based on trajectory coverage, characterized in that, Includes the following steps: Step S1: Acquire real-time video streams from multiple surveillance cameras in the computer room inspection area, identify the inspection personnel target in each frame of the real-time video stream, and determine the detection box and pixel coordinates of the inspection personnel target in the image coordinate system; Step S2: Extract the minimum bounding rectangle of the detection box to obtain the spatial range of the pixel coordinates in the image coordinate system. Define the target mapping area corresponding to the spatial range in the computer room electronic floor plan. Calculate the scaling factor based on the geometric relationship between the spatial range and the target mapping area. Step S3: Obtain the angle between the current optical axis orientation of the surveillance camera and the wall of the computer room as the rotation angle of the surveillance camera, obtain the angle between the rotation angle and the main axis direction of the electronic floor plan of the computer room as the angle difference, and convert the pixel coordinates of the inspection personnel target in different surveillance camera images into floor plan coordinates on the electronic floor plan of the computer room according to the scaling factor and the angle difference. Step S4: Perform spatiotemporal fusion of the planar coordinates from different surveillance cameras to generate a continuous movement trajectory of the inspection personnel target on the electronic planar map. Based on the continuous movement trajectory, perform inspection and supervision to determine whether the inspection work of the inspection personnel target meets the preset requirements.

2. The method for inspecting and supervising a computer room based on trajectory coverage according to claim 1, characterized in that: Step S2 calculates the scaling factor based on the geometric relationship between the spatial range and the target mapping area, including the following steps: calculating the ratio of the width of the target mapping area to the width of the spatial range, and the ratio of the height of the target mapping area to the height of the spatial range, to obtain the horizontal scaling factor and the vertical scaling factor.

3. The method for inspecting and supervising a computer room based on trajectory coverage according to claim 1, characterized in that: Step S3 converts the pixel coordinates of the inspection personnel target in different surveillance camera images into planar coordinates on the electronic floor plan of the computer room, including the following steps: constructing a rotation transformation matrix based on the angle difference, constructing a scaling transformation relationship based on the scaling factor, applying the rotation transformation matrix and the scaling transformation relationship to the pixel coordinates in sequence to obtain preliminary mapped coordinates; calculating the translation vector of the preliminary mapped coordinates relative to the target mapped area, and using the translation vector to perform offset calibration on the preliminary mapped coordinates to obtain the final planar coordinates.

4. The method for inspecting and supervising a computer room based on trajectory coverage according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: Obtain real-time video streams from multiple surveillance cameras within the computer room inspection area; Step S12: Use a target detection model to identify the inspection personnel target from the real-time video stream; Step S13: Use a tracking algorithm to perform inter-frame correlation on the inspection personnel target, and output the movement trajectory of the inspection personnel target and the corresponding detection box.

5. The method for inspecting and supervising a computer room based on trajectory coverage according to claim 1, characterized in that: Step S4, which involves spatiotemporal fusion of planar coordinates from different surveillance cameras, includes the following steps: Step S41: When the inspection personnel target enters the field of view of the monitoring camera, extract the feature vector according to the detection frame and generate a global identity ID for the inspection personnel target; Step S42: When the inspection personnel target appears simultaneously in the overlapping area of ​​the field of view of multiple surveillance cameras, calculate the feature similarity between the detection boxes output by different surveillance cameras. If the feature similarity is greater than the set threshold, determine that the inspection personnel target in the field of view of different surveillance cameras is the same inspection personnel and inherit the global identity ID. Step S43: In the area where the field of view intersects, calculate the weight of the monitoring camera based on the distance between the inspection personnel target and the different monitoring cameras. Use the weight to perform a weighted average calculation on the planar coordinates from different monitoring cameras to obtain the optimal coordinate point. Combine the optimal coordinate points at all times to obtain the continuous movement trajectory of the inspection personnel target on the electronic planar map.

6. The method for inspecting and supervising a computer room based on trajectory coverage according to claim 5, characterized in that: In step S43, in the field-of-view intersection area, when the inspection personnel target completely switches from the field of view of the first monitoring camera to the field of view of the second monitoring camera, the first planar coordinates of the inspection personnel target at the last visible moment in the first monitoring camera and the second planar coordinates in the second monitoring camera at the same moment are recorded. A linear interpolation algorithm is used to generate intermediate trajectory points between the first planar coordinates and the second planar coordinates, and a moving average filter is applied to smooth the continuous movement trajectory.

7. The method for inspecting and supervising a computer room based on trajectory coverage according to claim 5, characterized in that: In step S43, the weighted average of the planar coordinates from different surveillance cameras is calculated using the weights, and the expression for the optimal coordinate point is: ； ； In the formula, , These are the x and y coordinates of the optimal coordinate point, respectively. The total number of surveillance cameras when the target of the inspection personnel appears on multiple surveillance cameras at the same time; For the first The weight of each surveillance camera; , For the first The coordinates of a floor plan of a surveillance camera; For the inspection personnel's target and the first The distance between the surveillance cameras; It is a local minimum.

8. The method for inspecting and supervising a computer room based on trajectory coverage according to claim 1, characterized in that: Step S4, determining whether the inspection work of the inspection personnel target meets the preset requirements, includes at least the following steps: determining whether the coordinates of the plan are located in the non-inspection area set on the electronic plan; if the coordinates of the plan are located in the non-inspection area, calculating the continuous stay time of the inspection personnel target in the non-inspection area; when the continuous stay time exceeds the set time threshold, triggering an illegal intrusion alarm and capturing image evidence.

9. The method for inspecting and supervising a computer room based on trajectory coverage according to claim 1, characterized in that: Step S4 determines whether the inspection work of the inspection personnel meets the preset requirements, including at least the following steps: after the inspection task is completed, read the extreme values ​​of the coordinates of the continuous movement trajectory in each inspection area, calculate the actual inspection route length, calculate the ratio of the actual inspection route length to the planned inspection route length, obtain the trajectory coverage rate of each inspection area, and if the trajectory coverage rate is lower than the preset coverage rate threshold, generate an alarm for non-compliance of inspection completion.

10. A data center inspection and supervision system based on trajectory coverage, implemented based on the data center inspection and supervision method based on trajectory coverage as described in any one of claims 1-9, characterized in that, include: The system includes a video stream acquisition module, a task management module, an access control linkage module, a target detection and tracking module, a multi-view trajectory fusion module, an inspection and supervision module, an alarm module, and an interaction and visualization module. The video stream acquisition module acquires real-time video streams from multiple surveillance cameras within the computer room inspection area. The task management module is responsible for creating, distributing, and storing inspection work orders, and defining planned inspection routes and key check-in areas. The access control linkage module: connects to the computer room access control system and receives facial recognition check-in information from inspection personnel; When the inspector passes the verification, the task start signal is triggered, and the inspector's identity is bound to the current inspection task document; The target detection and tracking module identifies the patrol personnel in each frame of the real-time video stream, outputs the pixel coordinates of the detection box, generates a global identity ID for the patrol personnel, uses a multi-target tracking algorithm to track multiple targets, and maintains the consistency of the global identity ID. The multi-view trajectory fusion module: delineates the physical coverage area of ​​each camera in the electronic planar map, calculates the scaling factor from the image coordinate system to the planar map coordinates, performs coordinate rotation correction based on the rotation angle difference, converts pixel coordinates into unified planar map coordinates, fuses multi-camera data, and generates a continuous global trajectory. The inspection and supervision module monitors the matching degree between the movement trajectory of the inspection personnel and the preset standard inspection path and key checkpoints, and monitors the dwell time and trajectory coverage of the inspection personnel. The alarm module triggers an alarm when the inspection personnel's dwell time exceeds a set time threshold or the trajectory coverage is lower than a preset coverage threshold. The interaction and visualization module loads a vector map of the computer room and draws the movement trajectory points and connections of the inspection personnel in real time.

Images