Steel structure container hospital construction site safety risk real-time monitoring method

Abstract

The application discloses a kind of steel structure container hospital construction site safety risk real-time monitoring method, belong to data acquisition and surveillance control technical field;Method obtains the spatial absolute pose obtained by background occlusion mask inverse solution extracted from field video desensitization as first state data;Extract the phase doppler frequency shift of bottom wireless communication channel state information as second state data;Using asynchronous pose assimilation model, with absolute coordinate reference correct the drift error of high-frequency radio frequency deduction sequence, generate resonance fatigue risk feature matrix;When the extracted frequency domain main peak and spatial divergence rate touch safety constraint, synchronous output is used for targeted to break metal fatigue local lifting point stress release parameter and global lifting speed limit intervention instruction for situation immunity;The application breaks through the limitation of traditional perception lag and electromagnetic blind area, realizes the efficient, accurate and intelligent prevention of resonance precursor in large-scale steel structure hoisting process.

Description

Technical Field

[0001] This invention relates to the field of data acquisition and monitoring control technology, and in particular to a method for real-time monitoring of safety risks at the construction site of a steel structure container hospital. Background Technology

[0002] With the rapid advancement of modern industrialization and digitalization in construction, the cluster hoisting and rapid assembly technology of modular buildings (especially steel-structured containers) has been widely applied. In this complex engineering process, relying on data acquisition and monitoring control systems to achieve real-time interaction and monitoring of structural safety risks at the construction site has become a core technological approach to ensure project quality and personnel safety.

[0003] However, in scenarios involving the hoisting of dense clusters of heavy steel structures, existing monitoring technologies face severe physical environmental constraints, manifested in two major perception limitations: Firstly, macroscopic monitoring suffers from significant perception lag. Secondly, traditional monitoring systems based on Building Information Modeling (BIM) or crane sensors are limited by low frame rates and complex visual obstruction conditions such as backlighting and dense fog, often only passively triggering alarms when irreversible macroscopic geometric deformation occurs in the medical ward module, failing to capture preceding high-frequency structural fatigue signals. Thirdly, microscopic monitoring suffers from high-frequency oscillation data blind spots. Furthermore, the presence of numerous giant steel structures on-site easily creates extreme electromagnetic multipath scattering environments, leading to widespread signal loss and data packet loss in conventional wireless pose sensing networks (such as RTK or ordinary IoT nodes), making it difficult to continuously acquire accurate microscopic state data. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for real-time monitoring of safety risks at the construction site of a steel structure container hospital.

[0005] The objective of this invention is achieved as follows: a method for real-time monitoring of safety risks at the construction site of a steel structure container hospital, comprising the following specific steps:

[0006] S1. Obtain the first state data representing the three-dimensional displacement characteristics of the target medical container ward. The first state data includes the spatial pose absolute deviation vector obtained by comparing the background occlusion mask matrix of the multi-view two-dimensional projection contour library based on the BIM model with the background occlusion mask matrix of the hospital ward environment video.

[0007] S2. Obtain second-state data characterizing the high-frequency vibration features of the target medical container ward. The second-state data includes the radio frequency multipath signal phase perturbation sequence extracted based on the channel state information of wireless communication at the construction site.

[0008] S3. Using a pre-built asynchronous pose assimilation model based on Kalman filtering, feature alignment and fusion are performed: the phase perturbation sequence of the radio frequency multipath signal is converted into a high-frequency relative amplitude extrapolation sequence with length dimensions; the absolute spatial pose deviation vector under the current timestamp is extracted as the absolute coordinate observation benchmark, and the long-term cumulative position drift error of the high-frequency relative amplitude extrapolation sequence is dynamically corrected using the benchmark; a resonant fatigue risk feature matrix that can simultaneously characterize the absolute spatial position, high-frequency vibration intensity, and local node spatial displacement divergence rate is generated.

[0009] S4. Extract the first frequency domain feature parameter representing the main peak value of vibration frequency and the local node spatial displacement divergence rate from the resonance fatigue risk feature matrix. Extract the center reference value of the preset container natural frequency range. When the absolute value of the difference between the first frequency domain feature parameter and the center reference value is less than the safety margin threshold and the local node spatial displacement divergence rate exceeds the normal operation baseline, generate a node connection strength degradation diagnosis command pointing to the associated spatial node.

[0010] S5. Based on the node connection strength degradation diagnosis command, generate local suspension point stress release adjustment parameters for associated spatial nodes, and simultaneously output the lifting speed limit intervention command in the global monitoring layer to characterize the decrease in the overall wind load safety margin.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0012] This invention uses a reverse ray tracing inference algorithm to inversely solve the absolute deviation vector of spatial pose using a two-dimensional background occlusion mask matrix extracted from the environmental video stream; it embeds an environmental illumination extreme value game judgment and semantic mask stripping mechanism, and replaces target surface texture recognition by calculating the occluded area of ​​the background. This not only achieves physical-level blocking of personnel privacy, but also completely eliminates the fatal interference of traditional machine vision such as strong reflection of heavy steel structures and low illumination at night, ensuring the all-weather high availability of the absolute coordinate observation benchmark.

[0013] This invention extracts the channel state information of the underlying orthogonal frequency division multiplexing subcarriers during the idle heartbeat cycle, and extracts the high-frequency phase disturbance sequence through complex domain phase dewinding technology; it cleverly transforms multipath interference that leads to communication quality degradation into a highly sensitive physical probe, which penetrates the electromagnetic shielding area composed of giant metal stacks without the need to install physical vibration measurement equipment or occupy service bandwidth, and realizes the spatiotemporal decoupling and accurate extraction of high-frequency mechanical vibration characteristics.

[0014] This invention constructs an asynchronous pose assimilation model based on Kalman filtering as the core of fusion, and uses low-frequency visual absolute displacement to correct the relative position drift of high-frequency radio frequency inference, achieving deep integration at the physical dimension level; when the risk of local node resonance fatigue is diagnosed, it simultaneously outputs microscopic local suspension point stress release parameters and macroscopic global lifting speed limit commands; while interrupting the accumulation of metal fatigue at the physical source, it actively improves the wind load safety margin of the overall structure.

[0015] This invention configures a degradation strategy for unilateral failures in heterogeneous sensor networks. When the confidence level of the visual link drops sharply due to extreme weather such as sandstorms and touches the blocking limit, the absolute update of the filtering model is actively frozen and the model is shrunk to single-source operation. At the same time, the allowable tolerance bandwidth of the structure is forcibly reduced according to the configuration parameters. This avoids control downtime caused by poor data polluting the state matrix. Instead, it replaces safety with physical deceleration and cost reduction, giving it top-level anti-interference survivability in the face of extreme and unmeasurable conditions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the overall process flow of the present invention.

[0018] Figure 2 This is a schematic diagram of the pixel-level filtering steps of the two-dimensional background occlusion mask matrix of the present invention.

[0019] Figure 3 A schematic diagram illustrating the steps of generating a continuous radio frequency sequence for the instantaneous Doppler frequency shift characteristics of this invention.

[0020] Figure 4 This is a schematic diagram illustrating the steps of dynamically correcting the long-term cumulative position drift error of the high-frequency relative amplitude extrapolation sequence according to the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figures 1 to 4This invention provides a technical solution: a method for real-time monitoring of safety risks at the construction site of a steel structure container hospital, comprising the following steps:

[0023] S1. Obtain the first state data representing the three-dimensional displacement characteristics of the target medical container ward. The first state data includes the spatial pose absolute deviation vector obtained by comparing the background occlusion mask matrix of the multi-view two-dimensional projection contour library based on the BIM model with the background occlusion mask matrix of the hospital ward environment video.

[0024] S2. Obtain second-state data characterizing the high-frequency vibration features of the target medical container ward. The second-state data includes the radio frequency multipath signal phase perturbation sequence extracted based on the channel state information of wireless communication at the construction site.

[0025] S3. Using a pre-built asynchronous pose assimilation model based on Kalman filtering, feature alignment and fusion are performed: the phase perturbation sequence of the radio frequency multipath signal is converted into a high-frequency relative amplitude extrapolation sequence with length dimensions; the absolute spatial pose deviation vector under the current timestamp is extracted as the absolute coordinate observation benchmark, and the long-term cumulative position drift error of the high-frequency relative amplitude extrapolation sequence is dynamically corrected using the benchmark; a resonant fatigue risk feature matrix that can simultaneously characterize the absolute spatial position, high-frequency vibration intensity, and local node spatial displacement divergence rate is generated.

[0026] S4. Extract the first frequency domain feature parameter representing the main peak value of vibration frequency and the local node spatial displacement divergence rate from the resonance fatigue risk feature matrix. Extract the center reference value of the preset container natural frequency range. When it is determined that the absolute value of the difference between the first frequency domain feature parameter and the center reference value is less than the safety margin threshold (in this embodiment, the priority range of the safety margin threshold is defined as between 0.1 and 0.5, and the exact priority value is set to 0.2) and the local node spatial displacement divergence rate exceeds the normal operation baseline (in this embodiment, the priority range of the normal operation baseline is defined as between 10% and 25%, and the priority value is set to 15%), generate a node connection strength degradation diagnosis command pointing to the associated spatial node.

[0027] S5. Based on the node connection strength degradation diagnosis command, generate local suspension point stress release adjustment parameters for associated spatial nodes, and simultaneously output the lifting speed limit intervention command in the global monitoring layer to characterize the decrease in the overall wind load safety margin.

[0028] The steps for obtaining the first state data in S1 are as follows: obtain the two-dimensional background occlusion mask matrix after edge desensitization processing, retaining only the rigid body shadow contour; calculate the cross-union ratio error between the two-dimensional background occlusion mask matrix and the multi-view two-dimensional projection contour library through the inverse ray tracing inference algorithm, and inversely solve the absolute deviation vector of spatial pose.

[0029] The pixel-level filtering steps for the two-dimensional background occlusion mask matrix are as follows: Acquire the video stream of the hospital ward environment and extract the pixel matrix of a single-frame color image at the current timestamp; for any coordinate pixel in the single-frame color image pixel matrix, extract the red channel pixel value, green channel pixel value, and blue channel pixel value respectively; obtain three preset human eye physiological perception weight constants (in this embodiment, the human eye physiological perception weight constants are, preferentially, 0.299 for red, 0.587 for green, and 0.114 for blue); perform a product operation on the extracted red channel pixel value and the red weight constant to obtain the first product; similarly, calculate the second product of the green channel pixel value and the green weight constant, and the second product of the blue channel pixel value and the blue weight constant. The third product of the constant; the first, second and third products are added together, and the integer part of the result is extracted as the unique brightness channel value of the coordinate pixel; after traversing the entire matrix and performing weighted synthesis, the extracted full brightness channel values ​​are compared with the preset semantic segmentation binarization threshold one by one; when the brightness channel value is determined to be greater than the semantic segmentation binarization threshold, it represents dynamic background, strong reflection or human features, and the pixel status bit at the corresponding coordinate position is forcibly set to 0; when the brightness channel value is determined to be less than or equal to the semantic segmentation binarization threshold, it represents the rigid shadow outline on the backlit side, and the pixel status bit at the corresponding coordinate position is set to 1; after traversing the entire matrix, a two-dimensional background occlusion mask matrix consisting of status bits 0 and 1, retaining only the rigid shadow features, is output.

[0030] Specifically, the drastic changes in lighting caused by the alternation of day and night in the hospital ward are addressed by an ambient lighting extreme value game judgment mechanism for dynamic threshold calibration before obtaining the preset semantic segmentation binarization threshold. The global mean of the brightness channel values ​​of the entire matrix in a single frame image is extracted as the background lighting representation of the current operating condition. It is then determined whether the background lighting representation is lower than a preset extremely dark environment boundary threshold. If the determination is yes, nighttime low-light compensation logic is triggered, dynamically lowering the semantic segmentation binarization threshold to an optimal value within the first preset threshold range to prevent rigid shadow features from being incorrectly filtered out in dark fields. If the determination is no, it is determined that the current operating condition is normal or highly reflective, and the semantic segmentation binarization threshold is locked to an optimal value within the second preset threshold range. In the above-mentioned judgment and calibration logic for the brightness channel thresholds, all parameter extraction is configured based on a standard 8-bit grayscale depth system.

[0031] The extreme dark environment boundary threshold is determined by statistical deduction based on a large amount of historical benchmark data of passive auxiliary lighting conditions at night. In this embodiment, its priority range is between 30 and 50, with a priority value of 40. The first preset threshold interval represents the fault-tolerant extraction boundary under low illumination conditions. In this embodiment, the priority range is between 15 and 30, with a priority value of 20. The second preset threshold interval represents the anti-interference isolation boundary under normal industrial lighting conditions. In this embodiment, the priority range is between 80 and 120, with a priority value of 100. The embodiment is not limited to the above specific values. In practical applications, those skilled in the art can scale and adaptively calibrate the above intervals proportionally according to the auxiliary lighting intensity around the medical container.

[0032] All BIM contour library calibration parameters and virtual camera initial perspective matrix parameters are predefined and stored in a structured external data carrier; the configuration file is read and the content is mapped into the working memory; when processing the visual flow, a binary semantic mask is forcibly applied to strip away all dynamic background and personnel features, and then the minimum value optimization solver matches the rigid body pose with the smallest intersection-union error in the memory dictionary, outputting the pose deviation vector; by calculating the occluded area of ​​the background instead of directly identifying the target surface texture, not only is physical-level privacy of personnel at the hospital construction site blocked, but also the traditional machine vision's fatal interferences such as heavy steel structure reflection and low nighttime illumination are completely immune, ensuring the high availability of macroscopic absolute coordinate observation data.

[0033] The steps for obtaining the second state data in S2 are as follows: extract the complex matrix of the underlying orthogonal frequency division multiplexing subcarriers of the distributed nodes at the hospital construction site within the idle heartbeat period of the wireless network; perform signal phase continuity processing on the complex matrix of the underlying orthogonal frequency division multiplexing subcarriers, extract the instantaneous Doppler frequency shift features that characterize the change of electromagnetic wave transmission path, and generate a radio frequency multipath signal phase perturbation sequence.

[0034] The steps for generating a continuous radio frequency sequence from instantaneous Doppler frequency shift features are as follows: Extract the real and imaginary parts of each discrete carrier signal from the complex matrix of the underlying orthogonal frequency division multiplexing subcarriers; perform an arctangent function operation on the ratio of the imaginary to the real part to extract the original folded phase sequence truncated between negative and positive pi; calculate the absolute value of the phase difference between two adjacent time sampling points in the original folded phase sequence; set the following judgment condition: when the absolute value of the phase difference is greater than the constant pi, it is determined that phase winding has occurred, and an algebraic compensation operation of adding or subtracting twice the constant pi is performed on the phase values ​​of the current and subsequent sampling points; continuously iterate the detection and compensation operations until the absolute value of all adjacent phase differences is less than the constant pi, and output a continuous radio frequency multipath signal phase disturbance sequence that eliminates physical jumps and fully characterizes the small changes in multipath transmission distance.

[0035] Instead of parsing any network communication service payload, it calls the gateway's underlying API to directly read the channel state information data in the physical medium access control layer register. Because the huge metal surface of the target medical container ward module constitutes a strong multipath reflector, its own minute high-frequency mechanical vibrations cause extremely slight physical changes in the reflection path length. Using complex domain phase dewinding technology, the Doppler frequency shift characteristics caused by metal vibration are extracted as the data source for the physical high-frequency probe. Multipath interference that degrades communication quality is converted into a highly sensitive vibration sensing signal. No physical vibration measurement equipment needs to be installed on the hoisting module, it does not occupy effective communication bandwidth, penetrates the physical boundary of electromagnetic shielding, and achieves extremely rapid acquisition of the microscopic state.

[0036] The steps for dynamically correcting the long-term cumulative position drift error of the high-frequency relative amplitude extrapolation sequence in S3 are as follows: Based on the difference between the first sampling frequency for obtaining the absolute spatial pose deviation vector and the second sampling frequency for obtaining the phase perturbation sequence of the radio frequency multipath signal, a sliding time alignment window is set (in this embodiment, the value of the sliding time alignment window is set to 30 milliseconds; satisfying the hard boundary constraint of a total delay of less than 50 milliseconds for complex hoisting tasks at the hospital site and the optimal technical balance between the visual and radio frequency sampling rate range); within the sliding time alignment window, spline interpolation is performed on the high-frequency relative amplitude extrapolation sequence with length dimensions to extract the radio frequency compensation parameters after timestamp alignment synchronized with the timestamp of the absolute spatial pose deviation vector; based on the current on-site environment video, the image contrast information entropy is calculated, and the Kalman measurement noise covariance scalar is dynamically adjusted according to the image contrast information entropy; the measurement noise covariance scalar, the absolute spatial pose deviation vector, and the radio frequency compensation parameters are substituted into the Kalman gain equation to output the resonance fatigue risk feature matrix after completing the spatial reference alignment.

[0037] Within the sliding time-aligned window, spline interpolation employs the following cubic polynomial construction logic: Multiple discrete input timestamps of the high-frequency relative amplitude extrapolation sequence within the sliding time-aligned window are obtained, along with a corresponding set of relative amplitude observations; a system of cubic polynomial equations is constructed between adjacent discrete input timestamps; the following boundary constraints are set for the cubic polynomial equation system: the function output value of the cubic polynomial equation system at each discrete input timestamp is required to be equal to the corresponding relative amplitude observation value, and the first derivative (velocity) and second derivative (acceleration) of the entire cubic polynomial equation system at all intersection points must remain absolutely continuous; the equation system is solved to obtain the interpolation coefficient set; the target absolute timestamp corresponding to the spatial pose absolute deviation vector is obtained, and the target absolute timestamp is substituted into the cubic polynomial equation system with the interpolation coefficient set for algebraic operations, outputting RF compensation parameters that are absolutely synchronized with the visual reference timestamp.

[0038] Specifically, the process of solving the cubic polynomial equation system is not an abstract mathematical theoretical deduction, but rather is executed by edge-aware computing nodes deployed on the field side according to objective instruction flow. Specifically, after receiving the boundary constraints generated in the preceding sequence, an independent physical memory address mapping space is allocated in the local static random access memory, and the cubic polynomial equation system is transformed into the corresponding multidimensional coefficient determinant matrix and stored in the address space; the Gaussian elimination assembly instruction set is called to perform concurrent iterative calculations on the matrix data stream in the memory address space; the underlying bus status is monitored in real time, and when the hardware register flag outputs a convergence status indicator, the set of deterministic values ​​extracted from the specified output address buffer block is the final set of interpolation coefficients.

[0039] Among them, the convergence status flag represents the degree of closed-loop completion of the matrix operation logic. The state transition value is strictly represented as a discrete binary set: when the value is 0, it represents the abnormal termination of the operation caused by the divergence of the underlying data or memory overflow; when the value is 1, it represents the successful convergence of the physical operation and triggers the coefficient extraction action.

[0040] The dynamic adjustment of the Kalman measurement noise covariance scalar based on the image contrast information entropy also includes the following sub-steps: extracting the historical image contrast information entropy sequence within the sliding time alignment window, calculating the temporal information entropy dispersion feature of the historical image contrast information entropy sequence; obtaining the radio frequency compensation parameters synchronized with the previous timestamp and the spatial pose absolute deviation vector, extracting the data pairs of the corresponding discrete points on the time axis; calculating the rate of change difference between each data pair; obtaining the absolute value of all rate of change differences, generating a sequence of absolute difference values; performing logarithmic operations on each element in the absolute difference value sequence and then summing them to calculate the trajectory separation feature index, which characterizes the degree of deviation between the radio frequency compensation parameters and the spatial pose absolute deviation vector in the same manifold space; mapping the temporal information entropy dispersion feature and the trajectory separation feature index to a preset normalized dimensionless interval; when it is determined that the normalized temporal information entropy dispersion feature is greater than the steady-state threshold parameter, and the normalized trajectory separation feature index is greater than the covariance threshold parameter (in this embodiment, steady-state... The threshold parameter and the covariant threshold parameter together constitute the logical gating boundary for judging whether the camera support system has experienced resonance failure; the values ​​of the steady-state threshold parameter and the covariant threshold parameter are both set to 0.75; by truncating the first 25% of the extreme abnormal distributions in the high-dimensional normalized space, i.e., the [0,1] interval, ordinary wind noise perturbations can be accurately filtered out, and only the destructive benchmark loss-of-lock event is penalized), triggering the visual benchmark wind vibration penalty coefficient (in this embodiment, the visual benchmark wind vibration penalty coefficient represents the attenuation multiplier that forcibly deprives the confidence of the visual observation data when benchmark wind vibration is confirmed to have occurred, and the value of the visual benchmark wind vibration penalty coefficient is set to 2.5, which amplifies the information entropy through a multiplier greater than 1, i.e., the larger the information entropy, the more chaotic the judgment image, thereby giving higher measurement noise and achieving forced isolation of inferior benchmarks); the currently calculated image contrast information entropy is multiplied with the visual benchmark wind vibration penalty coefficient to generate the corrected contrast information entropy; the corrected contrast information entropy is used to dynamically adjust the Kalman measurement noise covariance scalar.

[0041] When dynamically adjusting the Kalman measurement noise covariance scalar using the corrected contrast information entropy, to ensure the dimensional consistency of the dynamic equations, the following strict cross-domain mapping logic is executed: A preset nominal baseline measurement covariance parameter is read from an external structured data carrier (in this embodiment, the preferred range of the nominal baseline measurement covariance parameter is limited to 0.0001 to 0.0025, with a preferred value of 0.0004). The nominal baseline measurement covariance parameter characterizes the inherent physical observation error variance baseline of the visual sensor under ideal wind-free conditions, possessing a precise spatial length square dimension; a pre-calibrated ideal image information entropy constant is obtained (in this embodiment, the ideal image information entropy constant serves as a dimensionless pure numerical base, its…). The priority range is defined between 4.5 and 6.5, with a default priority value of 5.5. The corrected contrast information entropy is divided by the ideal image information entropy constant. The information theory unit is eliminated through division, generating a dimensionless dynamic degradation factor. The exponential decay mapping judgment mechanism is invoked to perform an exponential operation on the dynamic degradation factor with the natural constant as the base, generating the final covariance amplification multiplier. The nominal basic measurement covariance parameter and the covariance amplification multiplier are then subjected to a pure algebraic product operation. The output result retains the covariance attribute that conforms to the real physical dimensions of the physical world, and fully inherits the wind vibration penalty characteristics from the information entropy perspective. Finally, it is used as the input source and substituted into the Kalman gain equation for subsequent asynchronous pose assimilation update operations.

[0042] By constructing an external structured data carrier independent of the core computing program, subsequent parameters are stored. All subsequent derivation calculations directly call the variables already loaded into memory for operation. The external structured data carrier is configured to persistently store various static runtime parameters in key-value pair format. A configuration reading process is initiated to locate and parse the external structured data carrier, deserializing the internally encapsulated data stream into a mapping dictionary structure in memory. The mapping dictionary structure is extracted from the mapping dictionary structure by index key names and assigned to the corresponding memory variables, thereby completing the dynamic loading of pre-loaded constants. The pre-loaded constants include, but are not limited to, a preset hardware consumption constant representing the maximum round-trip delay from end to end (in this embodiment, the effective value of the preset hardware consumption constant is...). The parameters include: an operating range of 10 to 30 milliseconds (with a preferred value of 20 milliseconds); an upper limit parameter for the total global task tolerance delay time characterizing the safety control baseline (in this embodiment, the reasonable setting range for the upper limit parameter for the total global task tolerance delay time is between 40 and 80 milliseconds, with a preferred value of 50 milliseconds); a lower limit parameter for the target isolation weight characterizing the physical limit (in this embodiment, the lower limit parameter for the target isolation weight is allowed to be dynamically optimized between 0.2 and 0.5, with a preferred value of 0.4); and a nominal dynamic wind load overturning safety factor characterizing the structural specification requirements (in this embodiment, the engineering allowable range for the nominal dynamic wind load overturning safety factor is defined as a dimensionless value between 1.4 and 2.0, with a preferred value of 1.67).

[0043] The sliding time alignment window calculation steps are as follows: First, obtain the inherent communication and instruction parsing latency between the edge computing node and the on-site distributed network. The underlying communication hardware must possess the key attribute that the maximum end-to-end round-trip latency does not exceed a preset hardware consumption constant. Second, obtain the preset hardware consumption constant (in this embodiment, it is calibrated to 20 milliseconds). Third, obtain the preset global task tolerance total latency upper limit parameter (in this embodiment, the global task tolerance total latency upper limit parameter is set to 50 milliseconds). Fourth, subtract the global task tolerance total latency upper limit parameter from the preset hardware consumption constant to obtain the time difference. Fifth, directly assign the time difference value to the sliding time alignment window, thus arriving at a specific value of 30 milliseconds. This ensures that while fully utilizing the hardware performance margin, the algorithm absolutely does not violate the latency red line for engineering safety.

[0044] The steps for determining the steady-state threshold parameter and the covariant threshold parameter are as follows: Extract historical video footage of the hospital ward environment under multiple normal weather conditions and the corresponding radio frequency multipath signal phase perturbation sequences as samples to establish a sample library; traverse and process all data pairs in the sample library, calculate and extract the corresponding original time-series information entropy dispersion feature set and the original trajectory separation feature index set; perform a global extreme value scan: extract the maximum value record in the original time-series information entropy dispersion feature set as the maximum dispersion extreme value, and extract the maximum value record in the original trajectory separation feature index set as the maximum separation extreme value; use division to divide each element in the original time-series information entropy dispersion feature set by the maximum dispersion extreme value to generate a normalized dispersion feature set; and separate the original trajectory... Each element in the feature index set is divided by the maximum separation extremum to generate a normalized separation feature index set; ensuring that all historical feature data are mapped to a dimensionless range with interval endpoints of 0 and 1; sorting all values ​​in the normalized discrete feature set in ascending order, extracting the values ​​at the 75th percentile of the total data, and directly assigning them to the steady-state threshold parameter; sorting all values ​​in the normalized separation feature index set in the same ascending order, extracting the values ​​at the 75th percentile of the total data, and directly assigning them to the covariant threshold parameter; ensuring that the two output threshold parameters strictly possess dimensionless statistical significance (in this embodiment, the specific values ​​are set to 0.75 based on the actual operating condition distribution), completely eliminating the subjectivity of manual debugging.

[0045] The steps for determining the visual baseline wind vibration penalty coefficient are as follows: First, set a target isolation weight lower limit parameter, which represents the maximum confidence percentage of visual observations allowed to participate in state updates when a baseline wind vibration occurs. In this embodiment, to prevent inferior data from dominating updates, the target isolation weight lower limit parameter is set to 40% (i.e., 0.4). Second, obtain a constant 1 and perform a division operation between the constant 1 and the target isolation weight lower limit parameter. Third, directly map the quotient of the above division operation to the visual baseline wind vibration penalty coefficient, resulting in the value 2.5 used in this embodiment. In the algorithm model, the contrast information entropy is multiplied by a 2.5-fold penalty amplification and then input into the covariance matrix. This allows the Kalman gain of visual observations to be smoothly suppressed to 40% of its original value through matrix inversion, achieving mathematical consistency and closed-loop operation.

[0046] To meet the hard boundary constraint of a total latency of less than 50 milliseconds for complex hoisting tasks at the hospital site, the fusion computation must complete the closed loop at the edge computing node. The state data used for auxiliary adjustment is explicitly defined as the image contrast information entropy of the current on-site environmental video. Based on this auxiliary data, the algorithm calculates a specific correction value, namely the Kalman measurement noise covariance scalar (a smaller entropy value indicates a clearer image, and a lower corresponding measurement noise covariance value indicates high confidence in visual observation). The correction parameter is used to adjust the main process, that is, the corrected covariance scalar is substituted into the Kalman gain equation to adjust the correction intensity for the high-frequency radio frequency extrapolation data, thereby outputting a feature matrix that includes both high-frequency vibration characteristics and eliminates drift errors. The absolute state reconstruction delay time is set as the collaborative gain KPI. The low-frequency absolute macroscopic displacement (visual source) and the high-frequency micro-perturbation spectrum (radio frequency source) are calibrated in a unified dimension. The dimensional consistency of the underlying physical absolute defense ensures that the output feature matrix possesses both the absolute spatial accuracy of vision and the millisecond-level sensitivity of radio frequency.

[0047] The steps for obtaining the first frequency domain feature parameter are as follows: Read the pre-stored complex Morlet wavelet basis function as the mother wavelet basis; generate a series of sub-wavelets with different center frequencies by performing continuous time axis translation and scaling operations on the mother wavelet basis; obtain the high-frequency data sequence of the resonance fatigue risk feature matrix, and perform convolution integration operations on the high-frequency data sequence and each sub-wavelet to generate a three-dimensional time-frequency spectrum composed of time coordinates, scale coordinates, and energy coefficients; extract the target scale parameter corresponding to the energy coefficient reaching its global maximum value in the three-dimensional time-frequency spectrum; obtain the preset wavelet center frequency constant, divide the wavelet center frequency constant by the target scale parameter, and then divide it by the basic sampling period parameter (in this embodiment, the priority range of the basic sampling period parameter is defined between 0.002 and 0.01, corresponding to a sampling rate of 100 to 500 Hz, where the exact priority value is set to 0.005), thereby generating the first frequency domain feature parameter characterizing the current transient vibration main frequency of the target medical container ward through pure algebraic transformation.

[0048] Furthermore, to ensure that frequency domain feature extraction can accurately capture the unique low-frequency resonant fatigue characteristics of medical containers, the complex Morlet wavelet basis is explicitly given a strict static data structure calibration during the offline model pre-configuration stage; specifically, the core characteristic parameter system of the mother wavelet basis is uniquely defined by two system-level preset scalars: namely, the system bandwidth parameter. and center frequency constant These two key parameters were determined before system initialization, based strictly on the historical vibration experimental database of standard elastic modes of steel used in medical container structures, and through a balance between the mutual constraints of resolution accuracy in the frequency domain and time domain.

[0049] Among them, system bandwidth parameters In the feature space mapping, a physical constraint index characterizing the attenuation rate of the sub-wavelet group energy envelope is used. Larger values ​​correspond to higher frequency domain resolution but lead to decreased time domain accuracy. Considering the transient vibration characteristics of the engineering site, in this embodiment, the preferred range is 1.0 to 3.0, with a preferred value of 1.5. The center frequency is constant. The core frequency reference parameter characterizing the inherent oscillation period of the mother wavelet in the model is lower in value, which better matches the low-frequency slow-varying resonance response of the container steel structure. In this embodiment, its priority range is 0.5 to 2.0, and the priority value is 1.0. The aforementioned value range is derived from the statistical extraction of standard container empty to full load conditions. In actual deployment, it can be finely adjusted based on the stiffness of different container models.

[0050] It also includes an adaptive degradation control step for extreme working conditions: when a link interruption is detected in the input first state data, or the confidence level of the absolute deviation vector of spatial pose is continuously lower than the preset blocking limit (in this embodiment, the priority range of the preset blocking limit is set between 0.5 and 0.7, that is, a confidence probability of 50% to 70%, and the factory-preconfigured preferred value is 0.6; the setting of this limit is not a subjective assumption, but to ensure that before cutting off visual input and degrading to a pure radio frequency blind test mode, false coordinate data with an error rate higher than 40% will never be adopted, thereby building the most robust automated construction safety defense line), an adaptive perception degradation instruction is generated; based on the adaptive perception degradation instruction, the absolute coordinate update operation in the asynchronous pose assimilation model based on Kalman filtering is frozen, so that the asynchronous pose assimilation model is degraded to a single-source operation mode; according to the pre-read configuration information, the structural allowable tolerance bandwidth parameter of the target medical container ward in the global monitoring layer is shrunk by a preset ratio, and the current lifting speed limit intervention instruction is updated.

[0051] For frequency domain feature extraction in S4, the edge program calls the continuous wavelet transform algorithm to analyze the high-frequency part of the matrix; S5 and the degradation strategy are implemented through explicit logic gating. When a visual blind spot occurs (such as a sharp drop in contrast caused by strong gusts of wind and dust), the program intercepts the status flag, blocks the Kalman update input of the visual observation, prevents inferior data from polluting the status matrix, reads the preset normal tolerance bandwidth and degradation shrinkage ratio coefficient in the external configuration table (in this embodiment, the degradation shrinkage ratio coefficient is 0.6, which reduces the lifting redundancy speed by 40% as a physical cost when a heterogeneous sensor fails on one side, in exchange for the absolute safety bottom line of the structure in the blind test state), performs algebraic multiplication operation and then issues it, forcing the lifting control center to execute subsequent lifting with more stringent speed and deflection angle restrictions; demonstrating robustness and degradation mechanism; when one of the heterogeneous sensor networks fails, it will not cause a crash or output outrageous control commands, but will automatically switch to the backup mode by relying on the degradation algorithm, and fulfill the safety commitment through physical deceleration and cost reduction, ensuring the safety bottom line of complex medical engineering.

[0052] The steps for determining the degradation shrinkage ratio coefficient are as follows: Call the interface data of the steel structure construction safety specification to obtain the nominal dynamic wind load overturning safety factor of the target medical container ward at the current height level (in this embodiment, the nominal dynamic wind load overturning safety factor is 1.67); extract the value 1, and perform a division operation between the value 1 and the nominal dynamic wind load overturning safety factor to obtain the reciprocal; round the result of the above reciprocal operation down to one decimal place to obtain the final attenuation parameter; assign the attenuation parameter to the degradation shrinkage ratio coefficient (i.e., 1 divided by 1.67 is approximately equal to 0.598, rounded down to 0.6).

[0053] The steps for obtaining the adjustment parameters are as follows: Obtain the first frequency domain characteristic parameter currently being judged, and extract the center reference value of the target container's natural frequency range from the configuration file; perform a subtraction operation on the first frequency domain characteristic parameter and the center reference value and take the absolute value to obtain the absolute value of the resonance frequency difference; obtain the preset Hooke's law stress conversion coefficient (in this embodiment, the Hooke's law stress conversion coefficient characterizes the linear mapping relationship between the frequency shift and the residual stress inside the steel structure node); multiply the absolute value of the resonance frequency difference with the Hooke's law stress conversion coefficient to generate a correction base with mechanical dimensions; add the correction base to the current foundation lifting force command of the corresponding lifting node, and output the local lifting point stress release adjustment parameter for directly driving the winch to release millimeter-level to centimeter-level micro-motion compensation.

[0054] In a specific on-site engineering logic deduction, due to the cross-domain mapping from intangible frequency drift characteristics to the actual mechanical driving quantity of the winch, the process of multiplying the absolute value of the resonant frequency difference with the stress conversion coefficient follows the continuous algebraic conversion model to eliminate the difference in physical dimensions: The extracted current correction base with mechanical dimensions is set as... The absolute value of the resonant frequency difference is set to The computing bus follows the algebraic model as follows Perform a dimensionless to dimensional characteristic transformation; in this conversion relation, the multiplier parameters This is the aforementioned pre-set Hooke's Law stress transformation coefficient.

[0055] Among them, the correction base The physical meaning refers to the specific physical tensile force value required for additional compensation loading or unloading at the target lifting node to release residual stress; the absolute value of the resonant frequency difference. In the characteristic space, the physical offset of the monitored resonant frequency from the nominal steady-state center is represented; multiplier parameter The comprehensive elastic-plastic coefficient of the wire rope of the bridge crane and the special steel lifting lug of the medical container is based on the stiffness test limit calibration of the national standard structural steel. In this embodiment, the priority range of its operating domain is defined between 5000 and 15000, and the exact system priority point value is set to 8500.

[0056] The above description of the embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A method for real-time monitoring of safety risks at the construction site of a steel structure container hospital, characterized in that, The specific steps include: S1. Obtain the first state data representing the three-dimensional displacement characteristics of the target medical container ward. The first state data includes the spatial pose absolute deviation vector obtained by comparing the background occlusion mask matrix of the multi-view two-dimensional projection contour library based on the BIM model with the background occlusion mask matrix of the hospital ward environment video. S2. Obtain second-state data characterizing the high-frequency vibration features of the target medical container ward. The second-state data includes the radio frequency multipath signal phase perturbation sequence extracted based on the channel state information of wireless communication at the construction site. S3. Using a pre-built asynchronous pose assimilation model based on Kalman filtering, feature alignment and fusion are performed: the phase perturbation sequence of the radio frequency multipath signal is converted into a high-frequency relative amplitude extrapolation sequence with length dimensions; the absolute spatial pose deviation vector under the current timestamp is extracted as the absolute coordinate observation benchmark, and the long-term cumulative position drift error of the high-frequency relative amplitude extrapolation sequence is dynamically corrected using the benchmark; a resonant fatigue risk feature matrix that can simultaneously characterize the absolute spatial position, high-frequency vibration intensity, and local node spatial displacement divergence rate is generated. S4. Extract the first frequency domain feature parameter representing the main peak value of vibration frequency and the local node spatial displacement divergence rate from the resonance fatigue risk feature matrix. Extract the center reference value of the preset container natural frequency range. When the absolute value of the difference between the first frequency domain feature parameter and the center reference value is less than the safety margin threshold and the local node spatial displacement divergence rate exceeds the normal operation baseline, generate a node connection strength degradation diagnosis command pointing to the associated spatial node. S5. Based on the node connection strength degradation diagnosis command, generate local suspension point stress release adjustment parameters for associated spatial nodes, and simultaneously output the lifting speed limit intervention command in the global monitoring layer to characterize the decrease in the overall wind load safety margin.

2. The method for real-time monitoring of safety risks at the construction site of a steel structure container hospital according to claim 1, characterized in that: The steps for obtaining the first state data in S1 are as follows: obtain the two-dimensional background occlusion mask matrix after edge desensitization processing, retaining only the rigid body shadow contour; calculate the cross-union ratio error between the two-dimensional background occlusion mask matrix and the multi-view two-dimensional projection contour library through the inverse ray tracing inference algorithm, and inversely solve the absolute deviation vector of spatial pose.

3. The method for real-time monitoring of safety risks at the construction site of a steel structure container hospital according to claim 2, characterized in that: The pixel-level filtering steps of the two-dimensional background occlusion mask matrix are as follows: acquire the video stream of the hospital ward environment, and extract the pixel matrix of the single-frame color image under the current timestamp; for any coordinate pixel point in the single-frame color image pixel matrix, extract the pixel values ​​of the red channel, green channel, and blue channel respectively. Obtain three preset human eye physiological perception weight constants; perform a product operation on the extracted red channel pixel value and the red weight constant to obtain the first product; similarly, calculate the second product of the green channel pixel value and the green weight constant, and the third product of the blue channel pixel value and the blue weight constant; perform a chain addition operation on the first product, the second product and the third product, and extract the integer part of the chain addition result as the brightness channel value that uniquely determines the coordinate pixel; after traversing the entire matrix and performing weighted synthesis, compare the extracted full brightness channel values ​​with the preset semantic segmentation binarization threshold one by one; When the value of the brightness channel is determined to be greater than the semantic segmentation binarization threshold, it represents dynamic background, strong reflection or human features, and the pixel status bit at the corresponding coordinate position is forcibly set to 0. When the value of the brightness channel is less than or equal to the semantic segmentation binarization threshold, that is, the rigid shadow contour of the backlit surface, the pixel state of the corresponding coordinate position is set to 1. After traversing the entire matrix, the output is a two-dimensional background occlusion mask matrix consisting of state bits 0 and 1, which retains only the rigid body shadow features.

4. The method for real-time monitoring of safety risks at the construction site of a steel structure container hospital according to claim 3, characterized in that: The steps for obtaining the second state data in S2 are as follows: extract the complex matrix of the underlying orthogonal frequency division multiplexing subcarriers of the distributed nodes at the hospital construction site within the idle heartbeat period of the wireless network; perform signal phase continuity processing on the complex matrix of the underlying orthogonal frequency division multiplexing subcarriers, extract the instantaneous Doppler frequency shift features that characterize the change of electromagnetic wave transmission path, and generate a radio frequency multipath signal phase perturbation sequence.

5. The method for real-time monitoring of safety risks at the construction site of a steel structure container hospital according to claim 4, characterized in that: The steps for generating a continuous radio frequency sequence from instantaneous Doppler frequency shift features are as follows: Extract the real and imaginary parts of each discrete carrier signal from the complex matrix of the underlying orthogonal frequency division multiplexing subcarriers; perform an arctangent function operation on the ratio of the imaginary to the real part to extract the original folded phase sequence truncated between negative and positive pi; calculate the absolute value of the phase difference between two adjacent time sampling points in the original folded phase sequence; when the absolute value of the phase difference is greater than the constant pi, phase winding is determined to have occurred, and algebraic compensation operations of adding or subtracting twice the constant pi are performed on the phase values ​​of the current and subsequent sampling points; continuously iterate the detection and compensation operations until the absolute value of all adjacent phase differences is less than the constant pi, outputting a continuous radio frequency multipath signal phase disturbance sequence that eliminates physical jumps and fully characterizes the minute changes in multipath transmission distance.

6. The method for real-time monitoring of safety risks at the construction site of a steel structure container hospital according to claim 5, characterized in that: The steps for dynamically correcting the long-term cumulative position drift error of the high-frequency relative amplitude extrapolation sequence in S3 are as follows: A sliding time alignment window is set based on the difference between the first sampling frequency for obtaining the absolute spatial pose deviation vector and the second sampling frequency for obtaining the RF multipath signal phase perturbation sequence; within the sliding time alignment window, spline interpolation is performed on the high-frequency relative amplitude extrapolation sequence with length dimensions to extract the RF compensation parameters after timestamp alignment synchronized with the timestamp of the absolute spatial pose deviation vector; the image contrast information entropy is calculated based on the current on-site environmental video, and the Kalman measurement noise covariance scalar is dynamically adjusted according to the image contrast information entropy. The measurement noise covariance scalar, the spatial pose absolute deviation vector, and the RF compensation parameters are all substituted into the Kalman gain equation to output the resonant fatigue risk characteristic matrix that completes spatial reference alignment.

7. The method for real-time monitoring of safety risks at the construction site of a steel structure container hospital according to claim 6, characterized in that: For spline interpolation within the sliding time alignment window, the following cubic polynomial construction logic is performed: obtain multiple discrete input timestamps of the high-frequency relative amplitude extrapolation sequence within the sliding time alignment window, and a corresponding set of relative amplitude observations; construct a system of cubic polynomial equations between adjacent discrete input timestamps; The following boundary constraints are set for the cubic polynomial equation system: the function output value of the cubic polynomial equation system at each discrete input timestamp must be equal to the corresponding relative amplitude observation value, and the first and second derivatives of the entire cubic polynomial equation system at all intersection points must remain absolutely continuous; the equation system is solved to obtain the interpolation coefficient set; the target absolute timestamp corresponding to the spatial pose absolute deviation vector is obtained, and the target absolute timestamp is substituted into the cubic polynomial equation system with the interpolation coefficient set for algebraic operations to output the radio frequency compensation parameters that are absolutely synchronized with the visual reference timestamp.

8. The method for real-time monitoring of safety risks at the construction site of a steel structure container hospital according to claim 7, characterized in that: The dynamic adjustment of the Kalman measurement noise covariance scalar based on the image contrast information entropy also includes the following sub-steps: extracting the historical image contrast information entropy sequence within the sliding time alignment window, and calculating the temporal information entropy dispersion characteristics of the historical image contrast information entropy sequence. The radio frequency compensation parameters and spatial pose absolute deviation vector synchronized with the preceding timestamp are obtained, and the data pairs corresponding to the discrete points on the time axis are extracted. The rate of change difference between each data pair is calculated. The absolute values ​​of all rate of change differences are obtained to generate a sequence of absolute difference values. The logarithmic operation is performed on each element in the sequence of absolute difference values, and then the summation operation is performed to calculate the trajectory separation feature index, which characterizes the degree of deviation between the radio frequency compensation parameters and the spatial pose absolute deviation vector in the same manifold space. The temporal information entropy dispersion feature and trajectory separation feature index are mapped to preset normalized dimensionless intervals respectively; when the normalized temporal information entropy dispersion feature is greater than the steady-state threshold parameter and the normalized trajectory separation feature index is greater than the covariance threshold parameter, the visual reference wind vibration penalty coefficient is triggered; the currently calculated image contrast information entropy and the visual reference wind vibration penalty coefficient are multiplied to generate the corrected contrast information entropy; the corrected contrast information entropy is used to dynamically adjust the Kalman measurement noise covariance scalar.

9. The method for real-time monitoring of safety risks at the construction site of a steel structure container hospital according to claim 8, characterized in that: The steps for obtaining the first frequency domain feature parameters are as follows: read the pre-stored complex Morlet wavelet basis function as the mother wavelet basis; generate a series of sub-wavelet sets with different center frequencies by performing continuous time axis translation and scaling operations on the mother wavelet basis. The high-frequency data sequence of the resonant fatigue risk feature matrix is ​​obtained. The high-frequency data sequence is then convolved with each sub-wavelet to generate a three-dimensional time-frequency spectrum consisting of time coordinates, scale coordinates, and energy coefficients. The target scale parameter corresponding to the energy coefficient reaching the global maximum value in the three-dimensional time-frequency spectrum is extracted. The preset wavelet center frequency constant is obtained. The wavelet center frequency constant is divided by the target scale parameter and then by the basic sampling period parameter. Through pure algebraic transformation, the first frequency domain feature parameter characterizing the current transient vibration main frequency of the target medical container ward is generated.

10. The method for real-time monitoring of safety risks at the construction site of a steel structure container hospital according to claim 9, characterized in that: It also includes adaptive degradation control steps for extreme operating conditions: when a link interruption is detected in the input first state data, or when the confidence level of the absolute deviation vector of spatial pose is continuously lower than the preset blocking limit, an adaptive perception degradation instruction is generated; based on the adaptive perception degradation instruction, the absolute coordinate update operation in the asynchronous pose assimilation model based on Kalman filtering is frozen, so that the asynchronous pose assimilation model is degraded to a single-source operation mode; according to the pre-read configuration information, the structural allowable tolerance bandwidth parameter of the target medical container ward in the global monitoring layer is shrunk by a preset ratio, and the current lifting speed limit intervention instruction is updated.

Images