A construction activity compliance analysis and early warning system based on remote sensing images

Abstract

The application relates to the technical field of remote sensing image processing and intelligent supervision of construction engineering, in particular to a construction activity compliance analysis and early warning system based on remote sensing images, which comprises the following modules: a multi-source monitoring data access module, which constructs a basic data pool for compliance monitoring; an expected state deduction module, which generates an expected projection benchmark representing a legal construction state at the moment; a spatial deviation correction module, which eliminates false deviations caused by sensor positioning jitter and locks an actual monitoring coordinate system; a multi-dimensional consistency identification module, which analyzes the residual error of a building shadow and identifies a concealed illegal feature; and a hierarchical early warning evaluation module, which quantifies a geometric compliance index, automatically triggers and outputs an illegal alarm signal; the application effectively avoids the ill-posed inversion problem with a huge amount of calculation in a traditional three-dimensional reconstruction scheme, and greatly reduces the required computing resources for system operation.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing image processing and intelligent supervision of construction projects, specifically a compliance analysis and early warning system for construction activities based on remote sensing images. Background Technology

[0002] With the rapid advancement of modern urban construction and large-scale engineering projects, the scale and complexity of construction activities supervision have significantly increased. In order to ensure that engineering construction strictly follows the digital building information model and the established construction schedule, real-time and automated compliance monitoring of the construction site has become a key link in industry supervision.

[0003] Currently, compliance analysis of construction activities mainly relies on manual on-site inspections or traditional 3D reconstruction schemes based on multi-view geometry. Manual inspections are limited by spatial span and labor costs, making it difficult to achieve large-scale, high-frequency regulatory coverage. Traditional 3D reconstruction schemes often face ill-conditioned inversion problems, have extremely high computational resource requirements, and are poorly robust under uncontrolled imaging conditions. In addition, when using remote sensing imagery for automated monitoring, positioning jitter from drones or satellite sensors often leads to systematic positional deviations between the image and the design model, resulting in false alarms. At the same time, changes in lighting at construction sites, cloud cover, and interference from numerous temporary facilities such as cranes and scaffolding make it difficult for the system to accurately identify hidden violations or distinguish permanent violations. In the absence of expensive stereo image pairs, even a single 2D image is insufficient to effectively monitor the height compliance of buildings. Therefore, how to achieve high-precision, low-computing-power automated analysis and early warning of construction activity compliance in complex and ever-changing construction environments, utilizing limited remote sensing observation resources, and through efficient data fusion and spatiotemporal alignment technologies, has become an urgent problem to be solved in this field. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a construction activity compliance analysis and early warning system based on remote sensing imagery. Specifically, the technical solution of this invention includes:

[0005] The multi-source monitoring data access module is configured to access the dynamic stream of remote sensing images of the target monitoring area, synchronized geolocation parameters, digital building information models, and construction progress plan data to build a basic data pool for compliance monitoring.

[0006] The expected state deduction module is configured to extract the building components that should be completed at the current time node based on the construction schedule data, and map the building components that should be completed onto the remote sensing image imaging plane through a virtual camera model to generate an expected projection benchmark representing the legal construction state at that moment.

[0007] The spatial deviation correction module is configured to rigidly align the expected projection reference with the structured edge features of the remote sensing image, eliminate false deviations caused by sensor positioning jitter, and lock the actual monitoring coordinate system.

[0008] The multidimensional consistency identification module is configured to simultaneously perform entity contour comparison and light and shadow topology verification under the actual monitoring coordinate system. The entity contour comparison is used to quantify the shape difference between the actual building and the design drawing, and the light and shadow topology verification is used to analyze the residual of the building shadow based on the ambient light vector to identify the concealed violation features.

[0009] The graded early warning assessment module is configured to integrate entity contour differences and shadow geometric residuals to quantify geometric compliance indicators, and automatically trigger and output a violation alarm signal when the indicators exceed a preset physical tolerance threshold.

[0010] Preferably, the expected state deduction module performs the following operations to construct an alarm baseline:

[0011] Analyze the rational polynomial coefficients or position and attitude data of remote sensing images to simulate a virtual imaging optical path consistent with that of the actual shooting lens;

[0012] Perform forward geometric projection on the effective components in the current construction phase, calculate the expected semantic attributes and depth values ​​of the projected elements, and use them as the logical base map to determine whether there is any illegal intrusion into the structure.

[0013] Preferably, the spatial deviation correction module improves the accuracy of early warning through frequency domain locking technology:

[0014] In the frequency domain, the high-frequency structural components and low-frequency texture components of remote sensing images are separated.

[0015] By maximizing the correlation between high-frequency structural components and the expected projection reference gradient field, the external orientation elements of the virtual camera are corrected to prevent false alarms caused by spatial misalignment.

[0016] Preferably, the multidimensional consistency identification module uses the following logic to identify hidden violations when performing shadow geometry verification:

[0017] Obtain the solar elevation angle and azimuth angle at the moment of imaging;

[0018] Based on the digital building information model, the theoretical shadow area is deduced and topologically differs from the actual shadow area in the remote sensing image.

[0019] When the actual shadow exceeds the theoretical boundary or the expected shadow is missing, the anomaly is quantified as the shadow geometric residual, which serves as a hidden criterion for triggering an alarm.

[0020] Preferably, the multidimensional consistency identification module is also configured to perform high-level violation alerts:

[0021] In response to the monitoring state where the entity outline is occluded, the height value of the unknown physical entity is inverted by combining the projection length of the shadow geometric residual with the solar altitude angle.

[0022] The height value is compared with the height limit parameter in the digital building information model. If the height is exceeded, the compliance indicator is corrected to activate the alarm.

[0023] Preferably, the graded early warning assessment module filters out interference through an inverse difference field model:

[0024] Construct an inverse difference field to identify rigid structures that exist in remote sensing images but are missing from the design model;

[0025] Assess the geometric stability of the structure, eliminate transient dynamic noise from cranes and scaffolding, and identify permanently non-compliant entities;

[0026] The effective entity deviations are weighted and calculated to generate the final early warning trigger signal.

[0027] Preferably, the graded early warning assessment module optimizes alarm confidence through signal-to-noise ratio weighting:

[0028] Calculate the signal-to-noise ratio of local land cover texture in remote sensing images and generate a global alarm confidence map;

[0029] The confidence map is used to spatially weight the inverse difference field to suppress measurement noise caused by interference from clouds, lighting, etc., and to ensure that the output warning signal meets the preset reliability.

[0030] Preferably, the multi-source monitoring data access module performs a time axis alignment operation:

[0031] Obtain the absolute timestamp of the remote sensing image capture;

[0032] Only retrieve the set of existing components that should be completed before the specified timestamp in the construction schedule and load them into the early warning judgment engine to eliminate logical misjudgments caused by future planned construction periods.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. This system combines the construction schedule with the digital building information model and uses a dimensionality reduction strategy to project the 3D model onto the 2D imaging plane to generate a logical base map. This approach effectively avoids the ill-conditioned inversion problem with huge computational load in traditional 3D reconstruction schemes and greatly reduces the computing resources required for system operation. Even under uncontrolled imaging conditions with only a single satellite pass or a single image, it can achieve efficient and real-time automated compliance analysis.

[0035] 2. This system incorporates spatial deviation correction and frequency domain locking technologies, utilizing high-frequency structured edge features in remote sensing images to rigidly align the expected projection reference. This effectively eliminates systematic displacement deviations caused by positioning jitter from UAVs or satellite sensors, preventing false alarms due to spatial misalignment. By locking the actual monitoring coordinate system, subsequent comparisons are ensured to be performed on a pixel-level aligned basis, significantly improving the accuracy of early warnings.

[0036] 3. This system innovatively introduces a light and shadow topology verification and height inversion mechanism. By comparing theoretical shadows with actual observed shadows, it can identify concealed illegal features or shoddy construction. By combining shadow length with solar altitude angle, the system can invert the physical height of buildings from a single two-dimensional image. This quasi-three-dimensional measurement capability breaks through the limitation of traditional algorithms that can only monitor visible surfaces, effectively curbing hidden illegal construction and excessively tall illegal construction under obstruction.

[0037] 4. This system can intelligently identify and eliminate instantaneous dynamic noise such as cranes and scaffolding through the inverse difference field model and signal-to-noise ratio weighting mechanism, and lock permanent illegal entities. At the same time, it combines local texture signal-to-noise ratio to dynamically evaluate image quality and automatically suppress measurement noise caused by cloud or lighting interference. This multi-dimensional filtering and weighting logic ensures that the early warning signal meets the preset reliability, greatly reduces the false alarm rate in complex construction environments, and improves the rigor of the supervision results. Attached Figure Description

[0038] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0039] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0041] Example 1:

[0042] Please see Figure 1 A construction activity compliance analysis and early warning system based on remote sensing imagery, comprising:

[0043] The multi-source monitoring data access module is configured to access the dynamic stream of remote sensing images of the target monitoring area, synchronized geolocation parameters, digital building information models, and construction progress plan data to build a basic data pool for compliance monitoring.

[0044] The expected state deduction module is configured to extract the building components that should be completed at the current time node based on the construction schedule data, and map the building components that should be completed onto the remote sensing image imaging plane through a virtual camera model to generate an expected projection benchmark representing the legal construction state at that moment.

[0045] The spatial deviation correction module is configured to rigidly align the expected projection reference with the structured edge features of the remote sensing image, eliminate false deviations caused by sensor positioning jitter, and lock the actual monitoring coordinate system.

[0046] The multidimensional consistency identification module is configured to simultaneously perform entity contour comparison and light and shadow topology verification under the actual monitoring coordinate system. The entity contour comparison is used to quantify the shape difference between the actual building and the design drawing, and the light and shadow topology verification is used to analyze the residual of the building shadow based on the ambient light vector to identify the concealed violation features.

[0047] The graded early warning assessment module is configured to integrate entity contour differences and shadow geometric residuals to quantify geometric compliance indicators, and automatically trigger and output a violation alarm signal when the indicators exceed a preset physical tolerance threshold.

[0048] This embodiment discloses a construction activity compliance analysis and early warning system based on remote sensing imagery. The system consists of five core logic modules, which work together through a data bus to form a closed-loop automated supervision process.

[0049] Multi-source monitoring data access module: This module is the system's sensing entry point, and its core definition is: a set of interfaces for cleaning and standardizing heterogeneous spatiotemporal data; in this embodiment, this module is configured to synchronously access the following four types of data through API interfaces or data buses:

[0050] Remote sensing image dynamic stream: refers to a sequence of real-time or near real-time image data from satellites, drones, or fixed-point high-position cameras;

[0051] Synchronized geolocation parameters: These refer to the camera's external orientation elements that strictly correspond to the moment the image was captured, including latitude and longitude, elevation, and attitude angle.

[0052] Digital Building Information Modeling (BIM): refers to a vector database containing geometric and semantic information about a building throughout its entire lifecycle;

[0053] Construction schedule data: refers to time node data including Gantt charts or network diagrams, used to define the legal existence status of a specific point in time;

[0054] This module unifies the spatiotemporal benchmarks of the above data, constructs a basic data pool for compliance monitoring, and provides cleaned and standardized inputs for subsequent processing.

[0055] Expected State Deduction Module: The core task of this module is to solve the problem of what the standard is; in this embodiment, it is configured to perform virtual holographic projection operations; it filters out the current time node by analyzing the construction schedule plan. The following are the building components that are to be completed and are in the current state: load these components using a graphics engine, simulate the intrinsic and extrinsic parameters of real sensors using a virtual camera model, and project them onto a two-dimensional imaging plane consistent with the remote sensing image.

[0056] This process generates the expected projection baseline, which is an idealized depth map or semantic map that represents what the current scene should look like if the construction is fully compliant.

[0057] Spatial Deviation Correction Module: This module aims to address the false displacement caused by sensor positioning errors. In practical engineering, the positioning data (POS / RPC) of UAVs or satellites, including position and attitude system POS data or rational polynomial coefficient RPC files, often contain errors of several meters. This module utilizes structured edge features such as building edges and roads in remote sensing images to rigidly align the expected projection reference. By searching for the optimal matching position within a local window, it eliminates systematic deviations caused by sensor jitter, thereby locking the actual monitoring coordinate system and ensuring that subsequent comparisons are performed based on pixel-level alignment.

[0058] Multidimensional Consistency Authentication Module: This module is the core engine for violation detection; after coordinate system locking is completed, this module simultaneously executes two verification logics:

[0059] Entity contour comparison: used to calculate the intersection-over-union ratio (IoU) or edge distance of objects in remote sensing images of real buildings and the expected projection reference of the design drawings in terms of geometry, quantifying shape differences;

[0060] Light and shadow topology verification: This is the key innovation of this system; it introduces ambient lighting vectors to analyze the distribution of shadows in the image; by comparing theoretical shadows with actual shadows, it identifies violations such as objects being occluded but their presence being exposed by shadows, or objects existing but lacking the expected shadows; the shadow residuals are verified by applying the theoretical shadow mask... With actual observation of shadow mask Perform topological XOR operation To obtain, and use to quantify the spatial topological inconsistency between the two;

[0061] The graded early warning and assessment module is responsible for decision output. It integrates the above-mentioned differences in entity contours and shadow geometric residuals to calculate a comprehensive geometric compliance index. When the index exceeds the preset physical tolerance threshold, for example, allowing a construction error of 0.5 meters, the system automatically triggers an alarm signal and outputs the coordinates of the violation area and the estimated degree of violation.

[0062] Through the synergy of the above modules, this system avoids the problem of ill-conditioned inversion in traditional 3D reconstruction schemes; by using the expected projection benchmark as the logical base map, compliance verification is performed directly on the 2D image, which greatly reduces the computing power requirement. At the same time, the sensor error is eliminated through spatial deviation correction, enabling high-precision compliance analysis to be achieved even under uncontrolled imaging conditions, such as a single satellite pass-through.

[0063] Example 2:

[0064] The expected state simulation module performs the following operations to build an alarm baseline:

[0065] Analyze the rational polynomial coefficients or position and attitude data of remote sensing images to simulate a virtual imaging optical path consistent with that of the actual shooting lens;

[0066] Perform forward geometric projection on the effective components in the current construction phase, calculate the expected semantic attributes and depth values ​​of the projected elements, and use them as the logical base map to determine whether there is any illegal intrusion into the structure.

[0067] This embodiment further defines the specific operation process for the expected state deduction module to construct the alarm benchmark, aiming to ensure a high degree of consistency between the virtual projection and the real image in terms of geometric and optical properties;

[0068] The core of this step lies in establishing the mapping relationship between the virtual environment and the physical environment; the module parses the rational polynomial coefficients (RPC) in the remote sensing image metadata for satellite imagery or position and attitude data (POS) for UAV imagery; based on the above parameters, the projection matrix of the virtual camera is constructed. :

[0069]

[0070] in, The intrinsic parameter matrix, which includes focal length and principal point coordinates, is derived from the sensor calibration report.

[0071] : Rotation matrix, calculated from the attitude angles of the POS data;

[0072] Translation vector, obtained by converting GPS / BeiDou positioning coordinates;

[0073] Through this operation, the system simulates a virtual imaging optical path in the digital twin space that is completely identical to that of the actual shot.

[0074] The module loads valid components for the current construction phase, excluding components to be built in the future, and uses the graphics processing unit (GPU) to perform forward geometric projection.

[0075] For each 3D vertex in the BIM model Calculate its projected coordinates on the image plane. :

[0076]

[0077] in, The scale factor in the projection transformation. The homogeneous coordinates of the image plane;

[0078] During this process, the module not only calculates the pixel position, but also calculates two key attributes for each projection unit: the expected semantic attribute and the depth value;

[0079] Expected semantic attributes: Does this pixel belong to a wall, floor, or window?

[0080] Depth value: The distance of this pixel from the camera;

[0081] The generated image containing the above information is defined as a logical base map; this base map serves as the absolute benchmark for determining whether an unauthorized intrusion structure exists. Any area marked as empty in the base map but containing an entity in the actual image is considered a potential violation.

[0082] By parsing RPC / POS data and performing precise forward geometric projection, this embodiment transforms the complex 3D BIM model into a 2D semantic map that perfectly matches the perspective of remote sensing images. This dimensionality reduction strategy avoids the huge uncertainty in recovering 3D structures from 2D images, transforming compliance judgment into a deterministic image difference problem, and significantly improving the robustness of the system.

[0083] Example 3:

[0084] The spatial deviation correction module improves early warning accuracy through frequency domain locking technology:

[0085] In the frequency domain, the high-frequency structural components and low-frequency texture components of remote sensing images are separated.

[0086] By maximizing the correlation between high-frequency structural components and the expected projection reference gradient field, the external orientation elements of the virtual camera are corrected to prevent false alarms caused by spatial misalignment.

[0087] This embodiment details the specific implementation of the spatial deviation correction module using frequency domain locking technology to improve the accuracy of early warning, aiming to solve the interference of illumination changes and texture noise on registration accuracy;

[0088] Traditional grayscale-based registration is susceptible to lighting conditions, such as cloud shadows; this embodiment introduces a frequency domain separation mechanism.

[0089] For remote sensing images and expected projection benchmark Performing Fast Fourier Transform (FFT) or Laplace pyramid decomposition respectively, two types of components are separated:

[0090] High-frequency structural components ( ): This component includes rigid geometric features such as building edges and corners; it is insensitive to changes in lighting and reflects the topological structure of the scene.

[0091] Low-frequency texture components ( ): This component includes color fill and large-area illumination distribution; it is susceptible to cloud cover and seasonality and is considered noise in the registration process.

[0092] This step performs a correlation maximization operation; the system constructs an optimized objective function, calculating the cross-correlation matrix only for high-frequency structural components:

[0093]

[0094] in, : Represents the calculated translational deviation; : indicates cross-correlation operation; where, Represents the pixel coordinates in a remotely sensed image; This represents the translation component within the preset search window; using the calculated... The external orientation elements of a virtual camera are mainly translation vectors. The projection reference is shifted on the image plane until it completely coincides with the edge features of the remote sensing image.

[0095] By searching for the optimal matching position within a local window, systematic biases caused by sensor jitter are eliminated, thereby locking the actual monitoring coordinate system and ensuring that subsequent comparisons are performed based on pixel-level alignment; the corrected external orientation element translation vector The calculation formula is:

[0096]

[0097] Considering the potential for transient attitude jitter in the sensor during data acquisition, the system, while correcting the translation vector, calculates the sensitivity of the high-frequency structural component gradient to the virtual camera's attitude angle using the Jacobian matrix, and collaboratively updates the rotation matrix using the Gauss-Newton iterative method. ;

[0098] in, The rotation matrix of the current virtual camera The inverse matrix; This is the back projection function, used to determine the ground resolution (GSD) and corresponding depth value of the current pixel. Shift pixels on a two-dimensional image plane Mapped to a physical displacement vector in a three-dimensional world coordinate system; specifically, the back projection function The parsing expression is defined as:

[0099]

[0100] in, The ground resolution of the image in the reference plane. For the corresponding reference flight altitude, The downward viewing angle during payload imaging; by introducing depth values The function performs real-time scaling of the mapping ratio, which linearly compensates the two-dimensional deviation in pixel space into a three-dimensional displacement vector in physical space, eliminating projection scale distortion caused by terrain undulations or differences in building height.

[0101] By employing frequency domain locking technology, this embodiment cleverly removes low-frequency textures that are susceptible to environmental interference, and uses only stable high-frequency structural features for alignment. This effectively prevents spatial misalignment caused by sensor positioning jitter or changes in lighting, thereby avoiding false positive alarms caused by inaccurate alignment and ensuring high accuracy of early warning.

[0102] Example 4:

[0103] When performing shadow geometry verification, the multidimensional consistency identification module uses the following logic to identify hidden violations:

[0104] Obtain the solar elevation angle and azimuth angle at the moment of imaging;

[0105] Based on the digital building information model, the theoretical shadow area is deduced and topologically differs from the actual shadow area in the remote sensing image.

[0106] When the actual shadow exceeds the theoretical boundary or the expected shadow is missing, the anomaly is quantified as the shadow geometric residual, which serves as a hidden criterion for triggering an alarm.

[0107] This embodiment specifically illustrates the logic of the multidimensional consistency identification module performing shadow geometry verification, aiming to solve the problem of hidden violation identification under visual occlusion;

[0108] The module obtains the local solar altitude angle at the time of imaging from the metadata or astronomical calendar database of the remote sensing image. ) and solar azimuth ( );

[0109] Based on digital building information modeling (BIM), a ray tracing algorithm is used to simulate sunlight and generate theoretical shadow areas. This is the shape of the shadow that should appear on the ground assuming the construction is fully compliant.

[0110] Extracting actual shadow regions from remote sensing images using image segmentation algorithms The module performs topological difference operations. When extracting the actual shadow region, the module introduces a spectral feature verification mechanism, simultaneously calculating the Normalized Differential Shadow Index (NDSI) of the candidate region or analyzing the low reflectance characteristics of the near-infrared (NIR) band. If the radiance value of the candidate region in the infrared band is higher than the preset shadow confidence threshold, it is determined to be an interfering object such as dark asphalt pavement or dark building components and is removed to ensure... Only includes realistic shadows caused by physical occlusion.

[0111]

[0112] in, The expression represents the XOR operation, and this formula achieves the topological difference between the actual shadow region and the theoretical shadow region through the topological XOR operation.

[0113] The system focuses on the following two abnormal situations and quantifies them as shadow geometric residuals:

[0114] Actual shadows exceed theoretical boundaries: This means that there are entities not shown on the design drawings that are blocking the light, such as illegally added layers;

[0115] Missing Expectations: This means that the building structure that should have been on the design drawings was not built on site, such as due to shoddy workmanship.

[0116] This embodiment utilizes shadow geometry verification, overcoming the limitation of traditional visual algorithms that can only detect visible surfaces; even if the illegal structure is obscured by clouds or tower cranes, the shadows it casts on the ground are often difficult to hide; by using shadows as an implicit criterion, the system can identify unseen violations, significantly improving the penetration and comprehensiveness of supervision.

[0117] Example 5:

[0118] The multidimensional consistency identification module is also configured to perform high-level violation alerts:

[0119] In response to the monitoring state where the entity outline is occluded, the height value of the unknown physical entity is inverted by combining the projection length of the shadow geometric residual with the solar altitude angle.

[0120] The height value is compared with the height limit parameter in the digital building information model. If the height is exceeded, the compliance indicator is corrected to activate the alarm.

[0121] This embodiment is a refinement of the previous embodiment, specifically illustrating how to use shadow inversion technology to perform high violation warnings;

[0122] When the system detects that the entity outline is occluded and the height cannot be directly measured, this module responds to this state by utilizing the projected length of the shadow geometric residual. Combined with the current solar altitude angle ( The height of an unknown physical entity is inverted based on the principles of trigonometric geometry. :

[0123]

[0124] In order to improve the inversion accuracy on non-flat terrain, This is obtained by real-time retrieval of digital elevation model (DEM) data at the shadow projection point in the digital building information model, thereby compensating for the shadow stretching error caused by ground slope.

[0125] The height value of the unknown physical entity obtained through inversion;

[0126] : The measured length of the shadow that exceeds the theoretical boundary;

[0127] : Solar elevation angle at the moment of imaging;

[0128] The base elevation of the shadow projection surface is usually the ground level or the top elevation of the podium building;

[0129] The inversion obtained The height limit parameters for this area in the digital building information model (BIM) Compare them;

[0130] like ,in If the measurement error tolerance is exceeded, it is determined to be an excessive height violation; at this time, the system corrects the compliance indicators, forcibly activates the alarm, and marks the estimated violation height value.

[0131] This embodiment achieves quasi-three-dimensional measurement based on a single two-dimensional image; through a height violation early warning mechanism, the system can accurately invert the building height based solely on the shadow information of a single image without the need for expensive stereo image pairs or lidar data; this greatly expands the monitoring dimensions of the system and effectively curbs common illegal construction activities.

[0132] Example 6:

[0133] The graded early warning assessment module filters out interference through an inverse difference field model:

[0134] Construct an inverse difference field to identify rigid structures that exist in remote sensing images but are missing from the design model;

[0135] Assess the geometric stability of the structure, eliminate transient dynamic noise from cranes and scaffolding, and identify permanently non-compliant entities;

[0136] The effective entity deviations are weighted and calculated to generate the final early warning trigger signal.

[0137] This embodiment details how the graded early warning assessment module uses the inverse difference field model to filter out non-violation interference and ensure the permanent nature of the alarm.

[0138] Module construction of inverse difference field This field recorded a set of pixels that existed in remote sensing images but were missing in BIM projection;

[0139] In this set, the system uses edge gradient orientation histogram (HOG) or deep learning features to identify rigid structures with regular geometric shapes, such as straight lines and right angles, and excludes natural features such as vegetation and water bodies.

[0140] To distinguish between cranes under construction and illegally added floors, the module assesses the geometric stability of the structure; if the system receives multiple consecutive frames of images, it calculates the optical flow vector of the structure; if it is a single frame of image, it identifies specific objects such as tower crane booms and scaffolding nets based on a semantic database.

[0141] Areas identified as transient dynamic noise such as cranes, scaffolding, and vehicles are eliminated; the system only locks permanent non-compliant entities such as concrete walls and steel structure frames; specifically, the system uses a pre-set semantic feature library to perform morphological filtering on objects in the inverse difference field, and by identifying the topological connectivity differences between cantilever structures and tower crane trusses, temporary facilities with discrete grid features are marked as logical noise, while solid structures with continuous closed contours are locked as non-compliant entities to be evaluated;

[0142] Specifically, the logical operator for generating the early warning trigger signal is to determine whether the accumulated weighted area exceeds the physical tolerance threshold, and its calculation formula is as follows:

[0143]

[0144] in, Represents the pixel index in the image coordinate system; The set of pixels representing permanently non-compliant entities identified; The physical area corresponding to a single pixel; This is a preset position weight matrix; This is a preset threshold for the area of ​​allowable construction error; for example, small deviations in the core area are given a higher weight than large deviations in the edge area.

[0145] By using the inverse difference field model and geometric stability assessment, this embodiment effectively solves the problems of complex construction site environments and numerous temporary facilities. It can intelligently distinguish between legal temporary facilities not shown on the drawings and illegal permanent buildings, greatly reducing the false alarm rate caused by tower cranes, scaffolding, etc., and focusing on genuine illegal construction activities.

[0146] Example 7:

[0147] The graded early warning assessment module optimizes alarm confidence through signal-to-noise ratio weighting:

[0148] Calculate the signal-to-noise ratio of local land cover texture in remote sensing images and generate a global alarm confidence map;

[0149] The confidence map is used to spatially weight the inverse difference field to suppress measurement noise caused by interference from clouds, lighting, etc., and to ensure that the output warning signal meets the preset reliability.

[0150] This embodiment is a further optimization of the early warning assessment module, introducing a signal-to-noise ratio weighting mechanism to quantify the credibility of the alarm;

[0151] Due to cloud cover, smoke, or overexposure / underexposure, the quality of remote sensing images varies greatly across different areas; the module calculates the signal-to-noise ratio (SNR) of local land cover texture in the remote sensing image and generates a global alarm confidence map of the same size as the image. );

[0152] For any pixel Among them, confidence level The specific parsing expression is:

[0153]

[0154] in, The preset texture sensitivity weight coefficient has a value range of [value range missing]. ; For pixels Local variance of surface texture within the neighborhood. This represents the local contrast within that neighborhood. This represents the maximum local variance of the entire image. This represents the maximum local contrast of the entire image, used for normalizing various parameters.

[0155]

[0156] in, For a preset minimal arithmetic compensation constant, such as This is used to prevent computational anomalies where the denominator is zero when the overall confidence level is extremely low; if the denominator Below the preset quality threshold If the image quality is insufficient, the system will stop calculating and output a status flag indicating insufficient image quality.

[0157] Through this operation, the system automatically suppresses measurement noise caused by interference from clouds, lighting, etc.; only violations detected in areas with clear images and reliable textures will make a high contribution to the alarm, thereby ensuring that the output warning signal meets the preset reliability.

[0158] The introduction of a signal-to-noise ratio weighting mechanism gives the system self-awareness; it can dynamically assess the impact of image quality on the judgment result and avoid blindly issuing alarms when visibility is poor, such as when covered by thin clouds; this mechanism significantly improves the system's adaptability under complex weather conditions and the credibility of alarm results.

[0159] Example 8:

[0160] The multi-source monitoring data access module performs time axis alignment operation:

[0161] Obtain the absolute timestamp of the remote sensing image capture;

[0162] Only retrieve the set of existing components that should be completed before the specified timestamp in the construction schedule and load them into the early warning judgment engine to eliminate logical misjudgments caused by future planned construction periods.

[0163] This embodiment addresses the time synchronization problem in the multi-source monitoring data access module by proposing a time axis alignment operation to eliminate logical misjudgments caused by spatiotemporal misalignment.

[0164] The module parses remote sensing image metadata to accurately obtain the absolute timestamp of image capture. Accurate to the minute level;

[0165] The module queries the construction schedule database and performs a time filtering operation:

[0166]

[0167] in, : Set of valid components;

[0168] : No. Each building component;

[0169] The planned completion time for this component;

[0170] The module only retrieves data at timestamps. The existing set of components that should have been completed previously is loaded into the early warning and judgment engine; this means that for projects planned for the future... Even if a component is present in the BIM model, the system will temporarily hide it if construction has just begun.

[0171] By strictly aligning the timeline, this embodiment resolves the phase difference between the static BIM model and the dynamic construction process. It eliminates false alarms of missing components due to the project not being completed on time, and also eliminates misjudgments of unknown entities due to early construction. When combined with the logic of advance schedule, this ensures that the compliance analysis is based on the state that should be presented at this moment, rather than the ideal final state, which greatly improves the rigor of logical judgment.

[0172] It should be noted that 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A remote sensing image-based construction activity compliance analysis and early warning system, characterized in that, include: The multi-source monitoring data access module is configured to access the dynamic stream of remote sensing images of the target monitoring area, synchronized geolocation parameters, digital building information models, and construction progress plan data to build a basic data pool for compliance monitoring. The expected state deduction module is configured to extract the building components that should be completed at the current time node based on the construction schedule data, and map the building components that should be completed onto the remote sensing image imaging plane through a virtual camera model to generate an expected projection benchmark representing the legal construction state at that moment. The spatial deviation correction module is configured to rigidly align the expected projection reference with the structured edge features of the remote sensing image, eliminate false deviations caused by sensor positioning jitter, and lock the actual monitoring coordinate system. The multidimensional consistency identification module is configured to simultaneously perform entity contour comparison and light and shadow topology verification under the actual monitoring coordinate system. The entity contour comparison is used to quantify the shape difference between the actual building and the design drawing, and the light and shadow topology verification is used to analyze the residual of the building shadow based on the ambient light vector to identify the concealed violation features. The graded early warning assessment module is configured to integrate entity contour differences and shadow geometric residuals to quantify geometric compliance indicators, and automatically trigger and output a violation alarm signal when the indicators exceed a preset physical tolerance threshold.

2. The remote sensing image-based construction activity compliance analysis and early warning system according to claim 1, characterized in that, The expected state deduction module performs the following operations to construct an alarm baseline: Analyze the rational polynomial coefficients or position and attitude data of remote sensing images to simulate a virtual imaging optical path consistent with that of the actual shooting lens; Perform forward geometric projection on the effective components in the current construction phase, calculate the expected semantic attributes and depth values ​​of the projected elements, and use them as the logical base map to determine whether there is any illegal intrusion into the structure. 3.The remote sensing image-based construction activity compliance analysis and early warning system according to claim 1, characterized in that, The spatial deviation correction module improves the accuracy of early warning through frequency domain locking technology. In the frequency domain, the high-frequency structural components and low-frequency texture components of remote sensing images are separated. By maximizing the correlation between high-frequency structural components and the expected projection reference gradient field, the external orientation elements of the virtual camera are corrected to prevent false alarms caused by spatial misalignment.

4. The remote sensing image-based construction activity compliance analysis and early warning system according to claim 1, wherein, When performing shadow geometry verification, the multidimensional consistency identification module uses the following logic to identify hidden violations: Obtain the solar elevation angle and azimuth angle at the moment of imaging; Based on the digital building information model, the theoretical shadow area is deduced and topologically differs from the actual shadow area in the remote sensing image. When the actual shadow exceeds the theoretical boundary or the expected shadow is missing, the anomaly is quantified as the shadow geometric residual, which serves as a hidden criterion for triggering an alarm.

5. The remote sensing image-based construction activity compliance analysis and early warning system according to claim 4, characterized in that, The multidimensional consistency authentication module is also configured to perform high violation warnings: In response to the monitoring state where the entity outline is occluded, the height value of the unknown physical entity is inverted by combining the projection length of the shadow geometric residual with the solar altitude angle. The height value is compared with the height limit parameter in the digital building information model. If the height is exceeded, the compliance indicator is corrected to activate the alarm.

6. The remote sensing image-based construction activity compliance analysis and early warning system according to claim 1, wherein, The graded early warning assessment module filters out interference using an inverse difference field model: Construct an inverse difference field to identify rigid structures that exist in remote sensing images but are missing from the design model; Assess the geometric stability of the structure, eliminate transient dynamic noise from cranes and scaffolding, and identify permanently non-compliant entities; The effective entity deviations are weighted and calculated to generate the final early warning trigger signal.

7. The remote sensing image-based construction activity compliance analysis and early warning system according to claim 6, characterized in that, The graded early warning assessment module optimizes the alarm confidence level through signal-to-noise ratio weighting: Calculate the signal-to-noise ratio of local land cover texture in remote sensing images and generate a global alarm confidence map; The confidence map is used to spatially weight the inverse difference field to suppress measurement noise caused by cloud and light interference, ensuring that the output warning signal meets the preset reliability.

8. The remote sensing image-based construction activity compliance analysis and early warning system according to claim 1, wherein, The multi-source monitoring data access module performs a time axis alignment operation: Obtain the absolute timestamp of the remote sensing image capture; Only retrieve the set of existing components that should be completed before the specified timestamp in the construction schedule and load them into the early warning judgment engine to eliminate logical misjudgments caused by future planned construction periods.

Images