Building facade intelligent monitoring and early warning method and system

Abstract

The invention relates to the technical field of intelligent security and protection and building monitoring technologies, and provides an intelligent monitoring and early warning method and system for a building facade, and the method comprises the steps: S1, reference map generation and region division, S2, real-time monitoring and filtering, S3, abnormality recognition and signal generation, and S4, early warning response: according to an abnormal event signal, carrying out early warning according to the abnormal event signal; by deploying a camera array with a specific pitch angle, shooting of indoor scenes is avoided, and privacy is strictly protected; a high-precision floor level digital reference image is generated by utilizing multi-view image splicing and a space calibration model based on a real physical size, pixel abnormity in the image can be accurately mapped to a specific floor even a resident unit, and the problem of fuzzy traditional monitoring and positioning is solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent security and building monitoring technology, specifically to an intelligent monitoring and early warning method and system for building facades. Background Technology

[0002] With the acceleration of urbanization, high-rise and super high-rise buildings are becoming increasingly dense, highlighting public safety issues related to their facades. The main risks include: ① Objects falling from heights: Objects falling from heights travel at high speeds and with great impact, seriously threatening pedestrian safety; ② Fire smoke: Smoke from indoor fires overflowing from windows is a key early warning sign of a fire; ③ Detachment of exterior wall components: Wall plaster, tiles, or decorations loosening and falling due to aging; ④ Hazardous exposure of personnel: Such as children climbing window sills, and insufficient safety measures for outdoor workers. These incidents are characterized by their suddenness, concealment, and high level of harm.

[0003] Currently, the prevention and control of the aforementioned risks mainly relies on the following methods, but all of them have obvious limitations: manual patrols and ground monitoring: manual patrols are inefficient and have long intervals, making it impossible to achieve all-weather coverage. Ordinary ground-mounted surveillance cameras are limited by their field of view, making it difficult to completely cover building facades over 100 meters high, resulting in numerous blind spots, and their shooting distance is far, leading to insufficient image detail.

[0004] Traditional video analytics systems: Some systems attempt to detect motion from fixed camera feeds. However, the exterior of high-rise buildings is complex and ever-changing. Strong day-night lighting variations, seasonal and weather effects (such as rain, snow, fog, and glare), and non-target movements like birds, insects, and drifting objects all generate a large amount of interference signals, resulting in a very high false alarm rate and poor practicality.

[0005] The conflict between privacy protection and monitoring needs: To achieve effective monitoring, cameras often need to be pointed at areas such as windows and balconies, but this infringes on residents' privacy and raises legal and ethical issues, making it difficult to deploy many monitoring solutions in practice.

[0006] The early warning mechanism is simplistic: existing technologies typically end once an alarm is generated in the background, relying entirely on real-time responses from monitoring personnel. If personnel miss, delay, or are unable to assess the situation, the early warning information cannot be effectively delivered to the risk site, resulting in a "last mile" response delay and blind spots.

[0007] Based on the aforementioned problems, we propose an intelligent monitoring and early warning method and system for building facades. Summary of the Invention

[0008] In response to the shortcomings of existing technologies, this invention achieves automatic, accurate, and reliable identification of various safety hazards on the facades of high-rise buildings while protecting residents' privacy, and establishes an intelligent hierarchical response mechanism to improve the timeliness and effectiveness of safety warnings.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A method for intelligent monitoring and early warning of building facades includes the following steps:

[0011] S1. Baseline map generation and area division: Collect images of the building facade, generate digital baseline maps corresponding to each floor, and divide independent monitoring areas on the digital baseline maps based on the building structure information;

[0012] S2. Real-time monitoring and filtering: Continuously acquire real-time video streams, filter environmental interference from the images, and output filtered real-time images;

[0013] S3. Anomaly identification and signal generation: The filtered real-time image is compared with the digital reference image of the corresponding monitoring area. When an anomaly feature of a preset type is identified, an anomaly event signal containing anomaly type and location information is generated.

[0014] S4. Warning Response: Based on the abnormal event signal, perform the following warning response operations:

[0015] S41. Send a first control command to the remote monitoring terminal to trigger an audible and visual alarm and start a preset time limit for timing;

[0016] S42. If a confirmation signal is received from the remote monitoring terminal before the timeout, the alarm is deactivated;

[0017] S43. If the confirmation signal is not received when the timer expires, a second control command is sent to the local warning indicator light on the floor corresponding to the abnormal location to activate the indicator light for on-site warning.

[0018] Preferably, in step S1, generating a digital baseline map and dividing the monitoring area includes:

[0019] By combining multiple images from tilted perspectives through image stitching and geometric correction, a panoramic view of the building facade is generated. The geometric correction employs a perspective transformation model, and the calculation formula is as follows:

[0020] ,

[0021] In the formula, ( , ) represents the pixel coordinates of the image before correction of the tilt. , () represents the pixel coordinates of the corrected panoramic image. These are the homogeneous coordinate coefficients in the matrix. ~ The perspective transformation parameters are obtained by matching at least four pairs of image feature points.

[0022] Based on the building's physical dimensions, a mapping relationship is established from the pixel coordinates of the panoramic image to the actual floor and resident unit locations. The mapping formula is as follows:

[0023] ,

[0024] In the formula, This refers to the actual building height. This is the actual horizontal distance. The vertical pixel-to-physical size conversion factor. The horizontal pixel-to-physical size conversion factor, ( , () represents the pixel coordinates corresponding to the origin of the panoramic image. , These are the actual physical coordinates corresponding to the origin.

[0025] This is used to generate the digital baseline map aligned by floor, and to complete the division and labeling of each independent monitoring area.

[0026] Preferably, the environmental interference filtering in step S2 includes:

[0027] Based on a preset background model, pixel regions that conform to the characteristics of transient interference are separated and filtered out from the real-time image to obtain an interference-filtered image; and based on real-time ambient light data obtained from the environmental perception module or meteorological data interface, the interference-filtered image is subjected to illumination normalization processing to output the filtered real-time image.

[0028] Preferably, the background modeling and foreground detection are implemented using an adaptive Gaussian mixture model; the probability density function and parameter update formula for each pixel in the model are as follows:

[0029] ,

[0030] In the formula, Let be the pixel grayscale value at time t. The number of Gaussian models, Let be the weight of the i-th Gaussian component at time t. Let be the mean of the i-th Gaussian component at time t. Let be the variance of the i-th Gaussian component at time t. It is a Gaussian probability density function;

[0031] Parameter update rules:

[0032] ,

[0033] In the formula, The learning rate is the weight. The learning rate is the mean and variance; the weights of unmatched Gaussian components are updated to... The mean and variance remain unchanged;

[0034] The illumination normalization process is implemented using an algorithm based on Retinex theory; the multi-scale Retinex calculation formula is as follows:

[0035] ,

[0036] middle, The normalized image grayscale values. For scale quantity, Let be the weight of the s-th scale. To filter out the grayscale values ​​of the input image to remove interference, For convolution operations, Let Gaussian kernel function be defined as follows:

[0037] ,

[0038] In the formula, Let be the Gaussian kernel standard deviation of the s-th scale.

[0039] Preferably, in step S3, identifying the preset type of abnormal features includes:

[0040] The difference between the filtered real-time image and the digital reference image is calculated using the sum of squared gray-level differences method, as shown in the following formula:

[0041] ,

[0042] In the formula, for Location difference value, For window The number of pixels within, For filtered real-time images The grayscale value of the location, Let (i,j) be the grayscale value of the digital reference image at position (i,j); when When the value is greater than a preset threshold T, it is marked as a difference pixel, and then the significant change region is extracted;

[0043] Analyze the movement trajectory, appearance, and temporal changes of the areas of significant change, and classify them into at least one of the following abnormal types: objects thrown from heights, fire smoke, building structure detachment, or dangerous exposure of personnel.

[0044] A building facade intelligent monitoring and early warning system, used to implement the aforementioned monitoring and early warning method, includes an image acquisition unit, a data processing and control center, and an early warning execution unit interconnected and operating collaboratively via a communication network.

[0045] The image acquisition unit consists of an array of tilt cameras deployed around the building and set at a preset tilt angle. It is used to acquire image data of the building's exterior facade and transmit the image data to the data processing and control center through the communication network.

[0046] The data processing and control center is used to receive and process image data from the image acquisition unit, generate and send control commands to the early warning execution unit;

[0047] The early warning execution unit includes a remote monitoring terminal and local early warning indicator lights deployed on each floor. The remote monitoring terminal is used to receive and respond to alarms and control commands sent by the data processing and control center, and the local early warning indicator lights are used to receive and execute control commands from the data processing and control center.

[0048] Preferably, the data processing and control center further includes an interface for accessing an environmental sensing module or a meteorological data interface to obtain real-time environmental data to assist in the filtering of environmental interference.

[0049] Preferably, the communication network includes a data transmission network for connecting the image acquisition unit and the data processing and control center, and a building control network for connecting the local early warning indicator lights on each floor of the data processing and control center and the early warning execution unit.

[0050] This invention provides a method and system for intelligent monitoring and early warning of building facades. It has the following beneficial effects:

[0051] By deploying camera arrays with specific pitch angles (such as 40-50°), indoor scenes are physically avoided, thus strictly protecting privacy. By using multi-view image stitching and a spatial calibration model based on real physical dimensions, a high-precision floor-level digital reference map is generated, which can accurately map pixel anomalies in the image to specific floors or even residential units, solving the problem of ambiguous positioning in traditional monitoring.

[0052] By combining adaptive background modeling (such as GMM) with illumination normalization processing (such as Retinex algorithm), a two-level filtering mechanism can effectively distinguish and filter out interference caused by birds, cloud shadows, drastic changes in day and night illumination, seasonal weather effects, etc., which greatly improves the consistency of image analysis, significantly reduces the false alarm rate of the system, and can work stably under different times and weather conditions.

[0053] By integrating multi-dimensional features such as motion trajectory analysis (for projectiles and personnel), appearance and texture analysis (for smoke and detachment), and temporal change analysis (for smoke diffusion and static detachment), classification rules are constructed, which can accurately distinguish and identify four core safety hazards: objects thrown from heights, smoke from initial fires, detachment / cracks in external walls, and dangerous exposure of personnel. The system is comprehensive and highly intelligent.

[0054] The system employs a dual response mechanism: firstly, the remote monitoring center conducts professional review and handling; simultaneously, the system has a built-in timer, and if no manual response is received within the specified time, it automatically determines that on-site intervention is required and activates the local warning lights (such as flashing red) on the abnormal floors. This ensures that even if the monitoring center misses the warning, the warning information can reach the risk area directly, reminding on-site personnel to pay attention or evacuate, thus forming a complete warning closed loop. Attached Figure Description

[0055] Figure 1 This is a flowchart of the intelligent monitoring and early warning process of the present invention;

[0056] Figure 2 This is a flowchart of the anomaly identification and signal generation process of the present invention;

[0057] Figure 3 This is a flowchart of the hardware connection process of the system of the present invention; Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] See attached document Figures 1-2 As shown, a method for intelligent monitoring and early warning of building facades includes the following steps:

[0060] S1. Baseline map generation and area division: Collect images of the building facade, generate digital baseline maps corresponding to each floor, and divide independent monitoring areas on the digital baseline maps based on the building structure information;

[0061] S2. Real-time monitoring and filtering: Continuously acquire real-time video streams, filter environmental interference from the images, and output filtered real-time images;

[0062] S3. Anomaly identification and signal generation: The filtered real-time image is compared with the digital reference image of the corresponding monitoring area. When an anomaly feature of a preset type is identified, an anomaly event signal containing anomaly type and location information is generated.

[0063] S4. Warning Response: Based on the abnormal event signal, perform the following warning response operations:

[0064] S41. Send a first control command to the remote monitoring terminal to trigger an audible and visual alarm and start a preset time limit for timing;

[0065] S42. If a confirmation signal is received from the remote monitoring terminal before the timeout, the alarm is deactivated;

[0066] S43. If the confirmation signal is not received when the timer expires, a second control command is sent to the local warning indicator light on the floor corresponding to the abnormal location to activate the indicator light for on-site warning.

[0067] In step S1, generating a digital baseline map and dividing the monitoring area includes:

[0068] By combining multiple images from tilted perspectives through image stitching and geometric correction, a panoramic view of the building facade is generated. The geometric correction employs a perspective transformation model, and the calculation formula is as follows:

[0069] ,

[0070] In the formula, ( , ) represents the pixel coordinates of the image before correction of the tilt. , () represents the pixel coordinates of the corrected panoramic image. These are the homogeneous coordinate coefficients in the matrix. ~ The perspective transformation parameters are obtained by matching at least four pairs of image feature points.

[0071] Based on the building's physical dimensions, a mapping relationship is established from the pixel coordinates of the panoramic image to the actual floor and resident unit locations. The mapping formula is as follows:

[0072] ,

[0073] In the formula, This refers to the actual building height (unit: m). This represents the actual horizontal distance (unit: m). Vertical pixel-to-physical size conversion factor (unit: m / pixel). The horizontal pixel-to-physical size conversion factor (unit: m / pixel). , () represents the pixel coordinates corresponding to the origin of the panoramic image. , These are the actual physical coordinates corresponding to the origin.

[0074] This is used to generate the digital baseline map aligned by floor, and to complete the division and labeling of each independent monitoring area.

[0075] The environmental interference filtering in step S2 includes:

[0076] Based on a preset background model, pixel regions that conform to the characteristics of transient interference are separated and filtered out from the real-time image to obtain an interference-filtered image; and based on real-time ambient light data obtained from the environmental perception module or meteorological data interface, the interference-filtered image is subjected to illumination normalization processing to output the filtered real-time image.

[0077] The background modeling and foreground detection are achieved using an adaptive Gaussian mixture model; the probability density function and parameter update formula for each pixel in the model are as follows:

[0078] ,

[0079] In the formula, Let be the pixel grayscale value at time t. This represents the number of Gaussian models (usually 3 to 5). Let be the weight of the i-th Gaussian component at time t. Let be the mean of the i-th Gaussian component at time t. Let be the variance of the i-th Gaussian component at time t. It is a Gaussian probability density function;

[0080] Parameter update rules (when pixel) When matching the i-th Gaussian component):

[0081] ,

[0082] In the formula, The weighted learning rate (usually taken as 0.01~0.05). The learning rate is the mean and variance (typically set to 0.001~0.01); the weights of unmatched Gaussian components are updated as follows: The mean and variance remain unchanged;

[0083] The illumination normalization process is implemented using an algorithm based on Retinex theory; the multi-scale Retinex calculation formula is as follows:

[0084] ,

[0085] middle, The normalized image grayscale values. For scale quantity, The weights for the s-th scale (satisfying) =1), To filter out the grayscale values ​​of the input image to remove interference, For convolution operations, Let Gaussian kernel function be defined as follows:

[0086] ,

[0087] In the formula, Let be the Gaussian kernel standard deviation at the s-th scale (usually taken as 15, 80, 250, etc.).

[0088] In step S3, identifying the abnormal features of the preset type includes:

[0089] The difference between the filtered real-time image and the digital reference image is calculated using the sum of squared gray-level differences method, as shown in the following formula:

[0090] ,

[0091] In the formula, for Location difference value, For window The number of pixels within, For filtered real-time images The grayscale value of the location, Let (i,j) be the grayscale value of the digital reference image at position (i,j); when When the value exceeds a preset threshold T (usually 50~200, calibrated according to the actual scene), it is marked as a difference pixel, and then the significant change region is extracted.

[0092] Analyze the movement trajectory, appearance, and temporal changes of the areas of significant change, and classify them into at least one of the following abnormal types: objects thrown from heights, fire smoke, building structure detachment, or dangerous exposure of personnel.

[0093] See attached document Figure 3 As shown, an intelligent monitoring and early warning system for building facades, used to implement the aforementioned monitoring and early warning method, includes an image acquisition unit, a data processing and control center, and an early warning execution unit interconnected and operating collaboratively via a communication network.

[0094] The image acquisition unit consists of an array of tilt cameras deployed around the building and set at a preset tilt angle. It is used to acquire image data of the building's exterior facade and transmit the image data to the data processing and control center through the communication network.

[0095] The data processing and control center is used to receive and process image data from the image acquisition unit, generate and send control commands to the early warning execution unit;

[0096] The early warning execution unit includes a remote monitoring terminal and local early warning indicator lights deployed on each floor. The remote monitoring terminal is used to receive and respond to alarms and control commands sent by the data processing and control center, and the local early warning indicator lights are used to receive and execute control commands from the data processing and control center.

[0097] The data processing and control center also includes an interface for accessing an environmental sensing module or meteorological data to obtain real-time environmental data to assist in the filtering of environmental interference.

[0098] The communication network includes a data transmission network for connecting the image acquisition unit and the data processing and control center, and a building control network for connecting the local early warning indicator lights on each floor of the data processing and control center and the early warning execution unit.

[0099] Example 1: Image Acquisition Unit Deployment: High-definition network cameras are deployed at at least three high points in different directions around the high-rise building to be monitored, such as the surrounding plaza and adjacent rooftops. All cameras are fixed with universal brackets, and their tilt angles are uniformly adjusted so that the lens's principal optical axis forms an angle of 40° to 50° with the normal to the building's facade. This angle has been calculated to be sufficient to ensure that the camera's field of view completely covers the target facade from top to bottom. The formula for calculating the field of view coverage height is as follows:

[0100] ,

[0101] In the formula, The height of the building that the camera can cover (in meters). The horizontal distance between the camera and the building facade (unit: m). The elevation angle (in rad) between the lens's principal optical axis and the horizontal direction. The angle of depression between the lens's principal optical axis and the horizontal direction (unit: rad).

[0102] At the same time, due to its tilted perspective, the image only shows external structures such as windows and balconies, and cannot reveal the interior view, thus protecting the privacy of residents on a physical level.

[0103] Data Processing and Control Center Deployment: This center can be deployed locally or in the cloud. In the case of local deployment, a high-performance server is set up in the building monitoring room. This server integrates an image processing module and a system control module. The image processing module is responsible for running complex computer vision algorithms, and the system control module is responsible for logical judgment and command issuance. The server is interconnected with all cameras and terminal devices through a network switch.

[0104] Deployment of the early warning execution unit:

[0105] Remote monitoring terminal: Deployed in the security center, it consists of one or more workstations equipped with dedicated client software and has audible and visual alarm capabilities.

[0106] Local warning indicator: Install a smart LED indicator with a network address in the elevator lobby or public corridor on each floor of the building. Its color and flashing pattern can be remotely programmed and controlled.

[0107] Network Topology: All cameras are connected to the data transmission network switch in the equipment room via a fiber optic network. The data processing and control center server and remote monitoring terminals are also connected to this switch. The server is also connected to the warning indicator lights on each floor through a separate building control network gateway (such as a TCP / IP-based IoT gateway), forming two physically or logically isolated data channels.

[0108] Example 2: Establishment and regional division of digital baseline maps. During the initial system startup, the following steps are performed to establish monitoring baselines:

[0109] Initialize image acquisition: During a typical time of day with uniform lighting and no abnormal conditions (e.g., a weekday morning), control all tilt cameras to simultaneously capture a set of high-resolution photos covering the entire facade of the building; the fields of view of these photos overlap.

[0110] Panoramic image compositing: The image processing module in the server automatically runs the following process:

[0111] a. Feature matching: The Scale Invariant Feature Transform (SIFT) algorithm is used to extract a large number of local feature points from each photo.

[0112] b. Image registration and stitching: Based on the matched feature points, the spatial transformation relationship (such as homography matrix) between adjacent photos is calculated. All photos are projected onto a unified two-dimensional plane, and the stitching seams are eliminated through image fusion technology. Finally, a complete orthophoto panoramic image of the building facade without perspective distortion is generated.

[0113] Spatial calibration and region division: This is crucial for achieving precise positioning. The operator marks at least two reference points on the panoramic image with known actual distances (e.g., building corners) and inputs their actual distances. The system then automatically calculates the ratio of image pixels to actual dimensions. Subsequently:

[0114] a. Floor division: Based on the known building design floor height (e.g., 3.0 meters), the system automatically draws the horizontal dividing lines of each floor on the panoramic view, starting from the ground baseline.

[0115] b. Residential unit division: Within each floor area, based on the window or structural partition information provided in the building floor plan, vertical dividing lines are drawn proportionally to divide the exterior wall area of ​​each unit into an independent rectangular area.

[0116] c. Baseline map generation: The panoramic image is cropped according to the floor division lines to obtain a series of digital baseline map files aligned by floor; at the same time, a unique area identifier is generated for each divided rectangular area (such as F12_A representing unit A on the 12th floor), and its boundary coordinates are stored in the database; at this point, every physical part of the building facade has a unique "digital ID card" and a corresponding normal state baseline in the image space.

[0117] Example 3: Real-time filtering process for environmental interference. During real-time monitoring, the system intelligently filters each frame of video image to eliminate environmental interference.

[0118] Dynamic interference filtering: The system establishes an adaptive background model for each monitored area; this model can learn and memorize the visual features of the area under normal conditions (such as static walls and fixed windows);

[0119] When a new image frame is input, the system compares it with the background model to quickly identify "foreground" pixels that do not belong to the background (i.e., the parts that have changed). For these foreground regions, the system further analyzes their size, duration, and motion pattern. For example, a small area that appears for less than 1 second and has an erratic motion trajectory will be identified as a transient disturbance such as a bird or insect, rather than a safety hazard. The system will remove the pixel information corresponding to these disturbances from the image frame in which they are located and fill them with background information to obtain a "clean" image with filtered interference. Preferably, an adaptive Gaussian mixture model (GMM) can be used to achieve the above background modeling and foreground analysis.

[0120] Illumination Variation Normalization: Illumination at different times and under different weather conditions can greatly alter the color and brightness of building surfaces. To eliminate this effect, the system connects to external meteorological data services or ambient light sensors to obtain real-time information such as illumination intensity and color temperature. Based on this data, the system uses an illumination normalization algorithm to process the "interference-filtered image." The core of this algorithm is to decompose image information into reflection and illumination components and attempt to weaken the influence of the illumination component, so that the processed image mainly reflects the inherent color and texture of the object. For example, the same white wall photographed under warm light at dusk and strong light at noon will have very similar image appearances after normalization processing. Preferably, an algorithm based on Retinex theory can be used to implement this process.

[0121] Example 4: Rules for identifying and classifying abnormal features. The system compares the filtered real-time image with the digital reference image of the corresponding area and identifies anomalies according to preset rules:

[0122] Difference detection: First, the pixel-level differences between the real-time image and the baseline image are calculated. Then, through thresholding and morphological processing, all regions that have undergone significant changes are identified.

[0123] Feature analysis and classification: For each region of variation, extract its multi-dimensional features and apply classification rules:

[0124] High-altitude littering: Identified as a small object whose trajectory shows obvious and continuous downward acceleration, and the trajectory starts at the outer edge of a window or balcony;

[0125] Fire smoke: It is identified as a grayish-white area with blurred texture and diffused edges; its core feature is dynamic diffusion, that is, the area and concentration of the area will rapidly increase and spread over time.

[0126] Peeling / cracks in the wall: Identified as a newly added area with a relatively clear outline and fixed position, whose color and texture are different from the intact surrounding area; the key feature is static persistence, that is, once it appears, its shape and position remain basically unchanged in subsequent frames;

[0127] Personnel exposure to hazards: The identified area outline conforms to human morphological characteristics (such as head-to-shoulder ratio), and its location remains within the pre-defined hazardous geographical range, such as outside windowsills or balcony railings;

[0128] Once any rule is matched, the system generates a structured exception event signal, which includes the type, location (e.g., F12_A), timestamp, and confidence level.

[0129] Example 5: Tiered early warning response process. After receiving an abnormal event signal, the system control module initiates the following automated process:

[0130] Level 1 Alarm and Timer: Immediately send an alarm to the remote monitoring terminal, pop up a details window and sound an alarm; at the same time, start a 120-second countdown timer for the event.

[0131] Manual intervention path: If the monitor checks and confirms (or determines it to be a false alarm) within 120 seconds, the system will deactivate the remote alarm and the process will end.

[0132] Automatic upgrade path: If there is no response within 120 seconds, the system determines the event as "unhandled" and automatically performs the upgrade operation.

[0133] Level 2 On-site Alert: The system analyzes the location information F12_A in the abnormal event signal and determines that the floor is "12th floor". Subsequently, a control command is generated and accurately sent to the local warning indicator on the 12th floor through the building control network, ordering it to start in red flashing mode. This indicator immediately provides intuitive on-site risk warning to personnel in the public area of ​​the 12th floor.

[0134] System Reset: Once the abnormal state is confirmed to be eliminated on-site, a reset command can be sent to the system via a remote terminal or on-site reset button to turn off the indicator lights and restore the system to normal monitoring.

[0135] The generation of the digital reference map refers to the process of forming a panoramic orthographic projection of the building facade by performing feature matching, geometric correction and image fusion on multiple tilted images with overlapping fields of view, and spatially calibrating it based on known physical dimensions.

[0136] The transient interference characteristics refer to the pixel region exhibiting one or more of the following characteristics in a continuous image sequence: physical size smaller than a preset threshold, duration shorter than a preset duration, and motion trajectory not conforming to the laws of gravity or linear motion; these characteristics are used to distinguish it from substantial abnormal motions such as objects thrown from heights or people falling.

[0137] The classification and identification are based on the following rules: if the area of ​​significant change is a small moving target that conforms to a free fall trajectory, it is classified as a high-altitude object thrown; if it is a grayish-white area with blurred texture and diffusion, it is classified as fire smoke; if it is a newly added area with a fixed outline and static position, it is classified as building structure detachment; if it is a moving target with human outline features and located in a preset danger zone, it is classified as personnel exposure to danger.

[0138] Background modeling and foreground detection, as well as illumination normalization, can be performed using any known algorithm capable of achieving the corresponding functions; adaptive Gaussian mixture model (GMM) and Retinex-based algorithms are two preferred implementation methods.

[0139] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent monitoring and early warning of building facades, characterized in that, Includes the following steps: S1. Baseline map generation and area division: Collect images of the building facade, generate digital baseline maps corresponding to each floor, and divide independent monitoring areas on the digital baseline maps based on the building structure information; S2. Real-time monitoring and filtering: Continuously acquire real-time video streams, filter environmental interference from the images, and output filtered real-time images; S3. Anomaly identification and signal generation: The filtered real-time image is compared with the digital reference image of the corresponding monitoring area. When an anomaly feature of a preset type is identified, an anomaly event signal containing anomaly type and location information is generated. S4. Warning Response: Based on the abnormal event signal, perform the following warning response operations: S41. Send a first control command to the remote monitoring terminal to trigger an audible and visual alarm and start a preset time limit for timing; S42. If a confirmation signal is received from the remote monitoring terminal before the timeout, the alarm is deactivated; S43. If the confirmation signal is not received when the timer expires, a second control command is sent to the local warning indicator light on the floor corresponding to the abnormal location to activate the indicator light for on-site warning.

2. The monitoring and early warning method according to claim 1, characterized in that, In step S1, generating a digital baseline map and dividing the monitoring area includes: By combining multiple images from tilted perspectives through image stitching and geometric correction, a panoramic view of the building facade is generated. The geometric correction employs a perspective transformation model, and the calculation formula is as follows: ， In the formula, ( , ) represents the pixel coordinates of the image before correction of the tilt. , () represents the pixel coordinates of the corrected panoramic image. These are the homogeneous coordinate coefficients in the matrix. ~ The perspective transformation parameters are obtained by matching at least four pairs of image feature points. Based on the building's physical dimensions, a mapping relationship is established from the pixel coordinates of the panoramic image to the actual floor and resident unit locations. The mapping formula is as follows: ， In the formula, This refers to the actual building height. This is the actual horizontal distance. The vertical pixel-to-physical size conversion factor. The horizontal pixel-to-physical size conversion factor, ( , () represents the pixel coordinates corresponding to the origin of the panoramic image. , These are the actual physical coordinates corresponding to the origin. This is used to generate the digital baseline map aligned by floor, and to complete the division and labeling of each independent monitoring area.

3. The monitoring and early warning method according to claim 1, characterized in that, The environmental interference filtering in step S2 includes: Based on a preset background model, pixel regions that conform to the characteristics of transient interference are separated and filtered out from the real-time image to obtain an interference-filtered image; and based on real-time ambient light data obtained from the environmental perception module or meteorological data interface, the interference-filtered image is subjected to illumination normalization processing to output the filtered real-time image.

4. The monitoring and early warning method according to claim 3, characterized in that, The background modeling and foreground detection are achieved using an adaptive Gaussian mixture model; the probability density function and parameter update formula for each pixel in the model are as follows: ， In the formula, Let be the pixel grayscale value at time t. The number of Gaussian models, Let be the weight of the i-th Gaussian component at time t. Let be the mean of the i-th Gaussian component at time t. Let be the variance of the i-th Gaussian component at time t. It is a Gaussian probability density function; Parameter update rules: ， In the formula, The learning rate is the weight. The learning rate is the mean and variance; the weights of unmatched Gaussian components are updated to... The mean and variance remain unchanged; The illumination normalization process is implemented using an algorithm based on Retinex theory; the multi-scale Retinex calculation formula is as follows: ， middle, The normalized image grayscale values. For scale quantity, Let be the weight of the s-th scale. To filter out the grayscale values ​​of the input image to remove interference, For convolution operations, Let Gaussian kernel function be defined as follows: ， In the formula, Let be the Gaussian kernel standard deviation of the s-th scale.

5. The monitoring and early warning method according to claim 1, characterized in that, In step S3, identifying the abnormal features of the preset type includes: The difference between the filtered real-time image and the digital reference image is calculated using the sum of squared gray-level differences method, as shown in the following formula: ， In the formula, for Location difference value, For window The number of pixels within, For filtered real-time images The grayscale value of the location, Let (i,j) be the grayscale value of the digital reference image at position (i,j); when When the value is greater than a preset threshold T, it is marked as a difference pixel, and then the significant change region is extracted; Analyze the movement trajectory, appearance, and temporal changes of the areas of significant change, and classify them into at least one of the following abnormal types: objects thrown from heights, fire smoke, building structure detachment, or dangerous exposure of personnel.

6. A building facade intelligent monitoring and early warning system, used to implement the method as described in any one of claims 1 to 5, characterized in that, This includes image acquisition units, a data processing and control center, and an early warning execution unit that are interconnected and operate collaboratively via a communication network. The image acquisition unit consists of an array of tilt cameras deployed around the building and set at a preset tilt angle. It is used to acquire image data of the building's exterior facade and transmit the image data to the data processing and control center through the communication network. The data processing and control center is used to receive and process image data from the image acquisition unit, generate and send control commands to the early warning execution unit; The early warning execution unit includes a remote monitoring terminal and local early warning indicator lights deployed on each floor. The remote monitoring terminal is used to receive and respond to alarms and control commands sent by the data processing and control center, and the local early warning indicator lights are used to receive and execute control commands from the data processing and control center.

7. The system according to claim 6, characterized in that, The data processing and control center also includes an interface for accessing an environmental sensing module or meteorological data to obtain real-time environmental data to assist in the filtering of environmental interference.

8. The system according to claim 6, characterized in that, The communication network includes a data transmission network for connecting the image acquisition unit and the data processing and control center, and a building control network for connecting the local early warning indicator lights on each floor of the data processing and control center and the early warning execution unit.

Images