Method and system for monitoring the construction of a net without a crossing frame, computer device and medium

Abstract

The present application relates to the technical field of power construction data processing, and particularly relates to a non-crossing frame construction net closing operation monitoring method and system, computer equipment and medium; the method comprises the following steps: obtaining false echo proportion characteristic value and contrast attenuation amplitude characteristic value by analysis. Perform particle filtering algorithm, output droplet size distribution estimation. According to the deviation of the false echo proportion characteristic value and the contrast attenuation amplitude characteristic value, the fusion weight is calculated. The weighted fusion is carried out on the mapped image definition sequence, and the fusion confidence curve is generated to calculate the distance between the net closing rope and the live wire. In this way, the technical problem that the safety distance judgment accuracy is insufficient due to the lack of particle filtering based real-time estimation of droplet size to dynamically optimize the data processing algorithm of multi-sensor data fusion weight under complex weather conditions is solved, and the accuracy and reliability of high-altitude operation safety monitoring under complex weather conditions are improved.

Description

Technical Field

[0001] This invention relates to the field of power construction data processing technology, and in particular to a method, system, computer equipment, and medium for monitoring power construction network sealing operations without cross-passage. Background Technology

[0002] In the field of data processing technology for power construction safety monitoring, especially in high-risk, high-altitude operations such as netting without scaffolding, accurately and reliably calculating the safe distance between the netting rope and the live conductor from multi-source sensor data is a core data processing challenge for ensuring personal and power grid safety. With the development of sensing technology, data processing solutions that integrate LiDAR point clouds and multispectral visual images have become a mainstream technological trend for dealing with complex environments. These solutions aim to overcome the limitations of single data modalities by integrating data sources based on different physical principles through algorithms. However, in rainy and foggy weather, the interference mechanisms of atmospheric suspended particles on laser and visible light signals differ fundamentally: large-diameter fog droplets have a significant backscattering effect on lasers, easily injecting a large amount of false echo noise into point cloud data; while small-diameter, high-concentration fog droplets scatter visible light more strongly, severely degrading the contrast of image data. This contradiction in interference characteristics causes the confidence levels of both LiDAR and visual data sources to dynamically change with the actual, unknown droplet size distribution, posing a severe challenge to traditional data processing algorithms that use fixed or simple heuristic rules for weighted fusion.

[0003] In the field of dynamic data fusion, existing technologies have attempted to introduce more intelligent weight allocation mechanisms. For example, a prior art document (application publication number CN121502554A) discloses a method for evaluating the dynamic operational status of data fusion, which generates dynamic fusion weights by calculating general indicators such as historical consistency, environmental adaptability, and data stability of each data source. This scheme embodies an idea of ​​fusion based on data source reliability assessment. However, the evaluation dimensions relied upon by this method are relatively macroscopic and do not delve into the data essence of the specific physical scenario of "safe distance monitoring in rain and fog environments." Its parameters such as "environmental adaptability" fail to establish a quantitative and dynamic correlation model with "fog droplet size distribution," the core latent variable that directly determines the quality of multi-source data. Its weight adjustment mechanism is essentially a post-event evaluation, rather than a feedforward optimization based on real-time and accurate estimation of key environmental conditions (i.e., fog droplet size distribution). Therefore, existing technologies lack a data processing algorithm that can estimate the droplet size distribution in real time from time-series observation data (false echo ratio, contrast attenuation amplitude) and dynamically optimize the weighting of multi-sensor data fusion accordingly. As a result, the final safety distance calculation accuracy cannot meet the stringent requirements of high-altitude operations under complex rain and fog weather conditions. Summary of the Invention

[0004] To address the aforementioned shortcomings or deficiencies, this invention provides a method, system, computer equipment, and medium for monitoring construction netting operations without cross-passage scaffolding. This invention solves the technical problem of insufficient accuracy in determining safe distances caused by the lack of a data processing algorithm based on particle filtering to estimate droplet size in real time and dynamically optimize the weighting of multi-sensor data fusion.

[0005] This invention provides a method for monitoring netting operations during construction without scaffolding, comprising: Based on the acquired point cloud data and multispectral image data of the sealing operation area, the feature values ​​of false echo proportion and contrast attenuation amplitude are analyzed. Based on the time series of these feature values, a particle filtering algorithm is executed to output a droplet size distribution estimate. If the droplet size distribution estimate exceeds a preset threshold, the fusion weight of the LiDAR data and multispectral visual data is calculated based on the deviation between the normalized false echo proportion and contrast attenuation amplitude feature values. Based on the fusion weight, the point cloud confidence sequence mapped from the false echo proportion feature values ​​and the image sharpness sequence mapped from the contrast attenuation amplitude feature values ​​are weighted and fused to generate a fusion confidence curve. Based on the first and second peak positions identified from the fusion confidence curve, the distance between the sealing rope and the live wire is calculated.

[0006] According to a second aspect, the present invention provides a monitoring system for construction netting operations without scaffolding crossings, comprising: The monitoring feature extraction module is used to analyze the point cloud data and multispectral image data of the acquired netting operation area to obtain the feature values ​​of false echo proportion and contrast attenuation amplitude.

[0007] The distribution estimation output module is used to execute a particle filtering algorithm based on the time series of spurious echo proportion feature values ​​and contrast attenuation amplitude feature values ​​to output droplet size distribution estimates.

[0008] The fusion weight generation module is used to calculate the fusion weight of LiDAR data and multispectral visual data based on the deviation between the normalized characteristic values ​​of the false echo proportion and the characteristic value of the contrast attenuation amplitude when the estimated droplet size distribution exceeds a preset threshold.

[0009] The confidence curve generation module is used to generate a fused confidence curve by weighted fusion of the point cloud confidence sequence mapped by the false echo proportion feature value and the image sharpness sequence mapped by the contrast attenuation magnitude feature value.

[0010] The monitoring result generation module is used to calculate the distance between the sealing rope and the live conductor based on the first peak position and the second peak position identified from the fusion confidence curve.

[0011] According to a third aspect, the present invention provides a computer device comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform any of the monitoring methods for non-spanning construction netting operations in the embodiments of the present invention.

[0012] According to another aspect of the present invention, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute any of the monitoring methods for non-crossing scaffolding construction netting operations in the embodiments of the present invention.

[0013] The present invention provides a method for processing monitoring data of non-crossing scaffolding construction netting operations. This method is achieved through five core steps: feature analysis, state estimation, weight calculation, data fusion, and spacing output. Specifically, based on the acquired point cloud data and multispectral image data, the characteristic values ​​of false echo proportion and contrast attenuation amplitude are obtained, thus quantifying the degree of rain and fog interference on both lidar and visual raw data. A particle filtering algorithm is executed based on the time series of the two characteristic values ​​to output a fog droplet size distribution estimate, achieving real-time and accurate state estimation of the key environmental latent variable (fog droplet size). In response to the fog droplet size distribution estimate exceeding a preset threshold, a fusion weight is calculated based on the deviation of the two characteristic values ​​after normalization, establishing a dynamic weight allocation mechanism based on real-time environmental state and data quality differences. Based on the fusion weight, the point cloud confidence sequence and image sharpness sequence obtained by mapping the characteristic values ​​are weighted and fused to generate a fusion confidence curve, completing the multi-source data integration after dynamic optimization based on reliability. The spacing is calculated based on the double-peak positions identified from the fusion confidence curve, ultimately outputting accurate safety distance monitoring results.

[0014] In this technical solution, the present invention addresses the problem described in the background art, where the lack of accurate modeling of the dynamic correlation between droplet size distribution and the reliability of multi-source data leads to inaccurate fusion weight allocation. By introducing a particle filtering algorithm to estimate droplet size distribution in real time, the abstract assessment of "environmental adaptability" is transformed into tracking specific, quantifiable environmental state variables (droplet size), thus providing a direct and accurate physical basis for weight calculation. To address the consequence of insufficient accuracy in safety distance judgment caused by this problem, a closed-loop data processing flow of "dynamically calculating weights in response to droplet size estimation exceeding the threshold" is constructed. This allows the allocation of fusion weights to respond in real time to the fundamental impact of droplet size changes on the reliability of laser and visual data. When the environment is favorable to a particular sensing mode, the algorithm automatically assigns it a higher confidence level. Therefore, the data processing method of the present invention solves the technical problem of insufficient accuracy in safety distance judgment caused by the lack of a data processing algorithm based on particle filtering for real-time estimation of droplet size to dynamically optimize the fusion weights of multi-sensor data, thereby improving the accuracy and reliability of high-altitude operation safety monitoring under complex meteorological conditions. Attached Figure Description

[0015] Figure 1 This is a flowchart of a monitoring method for construction netting operation without scaffolding according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating another embodiment of the present invention: an intelligent detection method for safe distance during power transmission line closure operations based on multimodal sensing. Figure 3 This is a flowchart illustrating the logic of another embodiment of the present invention: dynamically adjusting multimodal fusion weights based on dual-channel bias. Figure 4 This is a flowchart illustrating a method for calculating the spatial distance between a sealing rope and a live conductor based on weighted fusion and peak identification, according to another embodiment of the present invention. Figure 5 This is a schematic diagram of the structure of a monitoring system for construction netting operation without scaffolding according to an embodiment of the present invention; Figure 6 This is a block diagram of a computer device for implementing embodiments of the present invention. Detailed Implementation

[0016] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0017] In the data processing method involved in this invention, through modeling and analysis of the interference mechanism of sensor signals, a quantifiable intrinsic correlation is revealed between the droplet size distribution and the reliability of two data sources, lidar and multispectral vision, in rainy and foggy environments: Increased droplet size significantly enhances laser backscattering, leading to an increase in the false echo proportion characteristic value and a decrease in data quality in point cloud data. Conversely, for multispectral vision data, the Mie scattering effect caused by small-diameter, high-concentration droplets results in a more significant image contrast attenuation characteristic value. Based on this physical correlation, the core concept of this invention is to: utilize the two interference indicators, the false echo proportion and contrast attenuation amplitude, directly parsed from the original observation data (point cloud and image), and use a time-series data processing algorithm (particle filtering) to inversely estimate the hidden state (droplet size distribution) that cannot be directly measured; then, based on this estimation result and the real-time quality difference between the two data streams, dynamically generate and allocate fusion weights; finally, fuse and analyze the weighted dual-channel confidence sequence to achieve accurate and robust calculation of the safe distance between the sealing rope and the live conductor. This solution embodies a complete data processing logic of "estimating the environmental state from the data, understanding the data quality based on the state, dynamically fusing the data based on the quality, and enhancing the final decision through fusion".

[0018] Specifically, traditional data fusion methods based on fixed rules or simple weighted averaging of multiple sensors suffer from technical flaws, leading to inaccurate decisions due to their inability to respond to dynamic changes in key environmental variables (droplet size distribution). These methods rely on statically preset fusion strategies or decision thresholds, failing to incorporate the core latent variable of "droplet size distribution" into the real-time data processing loop. This deficiency manifests in several ways: in large-droplet environments, there is an over-reliance on lidar data subjected to strong backscattering interference; while in small-droplet environments, it is difficult to effectively utilize visual image information that is severely attenuated by scattering. This directly results in a higher misjudgment rate in the final safe distance calculation under specific weather conditions, failing to provide reliable data support for high-altitude operations. In contrast, the dynamic data fusion method based on particle filtering for estimating droplet size proposed in this invention achieves implicit modeling of environmental interference mechanisms through algorithms; it realizes intelligent resource allocation between two sensing data sources by responding to adaptive weight calculation triggered by exceeding the threshold of droplet size estimation; it ensures that system decisions are biased towards the more reliable data source under any particle size distribution by generating fusion weights based on normalized deviation; and it achieves stable and accurate calculation of the distance between key targets in the context of strong noise data by identifying peaks and calculating differences in the confidence curve generated after weighted fusion.

[0019] Therefore, the data processing method for monitoring netting operations without scaffolding provided in the first aspect of this invention can be applied to corresponding safety monitoring systems. This system can be deployed at the netting operation site via local edge computing or cloud-edge collaboration, using monitoring computing units and remote servers, to achieve real-time sensing, calculation, and early warning of the safe distance between the netting ropes and live conductors in rainy or foggy environments. The system has flexible deployment capabilities and can be integrated into embedded industrial control computers, on-site edge servers, or cloud computing platforms on construction operation platforms. This meets the different needs of high-altitude operation sites for low-latency real-time monitoring, and the operation and maintenance center for centralized management and in-depth analysis of multi-point, multi-dimensional data, realizing a closed-loop data processing system from precise on-site sensing to global intelligent operation and maintenance.

[0020] like Figure 1 As shown, the method may include: Step S110: Based on the acquired point cloud data and multispectral image data of the sealing operation area, analyze and obtain the characteristic values ​​of false echo proportion and contrast attenuation amplitude.

[0021] Point cloud data refers to a set of discrete points containing a large number of three-dimensional spatial coordinates and corresponding echo intensity information, acquired by a lidar scanning unit by emitting a laser beam and receiving the reflected echo from the target. Multispectral image data refers to a two-dimensional image array containing spectral information of multiple bands, such as visible light and near-infrared, simultaneously captured by a multispectral vision acquisition unit. The false echo proportion feature value is a scalar parameter used to quantify the degree of interference of rain and fog particle scattering on point cloud data. Its value is equal to the ratio of the number of noise points determined to be caused by rain and fog particle scattering to the total number of points in the point cloud. The contrast attenuation amplitude feature value is a scalar parameter used to quantify the degree of sharpness reduction caused by rain and fog attenuation of multispectral images. Its value reflects the proportion of the decrease in the contour contrast of the target area (such as a rope for sealing a net and a live wire) in the current image relative to the baseline contrast in clear weather.

[0022] Specifically, the system can perform the following operations through the data processing unit: perform spatial clustering and intensity analysis on point cloud data, distinguish target points from scattering noise points based on echo intensity threshold and spatial distribution dispersion, and calculate the ratio to obtain the false echo proportion feature value; perform contour extraction and grayscale analysis on multispectral images, calculate the current grayscale difference of the target area, compare it with the pre-stored baseline grayscale difference, and calculate the attenuation ratio to obtain the contrast attenuation amplitude feature value.

[0023] For example, the system processes a frame of point cloud data containing 100,000 points, and through analysis, identifies those points that are scattered and have an intensity lower than [a certain value]. If 20,000 points with a density of decibels (dB) are identified as scattering noise points, then the characteristic value of the false echo proportion is calculated to be 0.2. Simultaneously, the system measures the current grayscale difference between the area of ​​the sealing rope and the conductor in the multispectral image as 30, while the pre-stored clear baseline grayscale difference is 100. Therefore, the characteristic value of the contrast attenuation amplitude is calculated as follows: .

[0024] Step S120: Based on the time series of spurious echo proportion feature values ​​and contrast attenuation amplitude feature values, execute the particle filtering algorithm to output the droplet size distribution estimate.

[0025] Among them, time series refers to an ordered data sequence composed of the two feature values ​​mentioned above, which are continuously collected and arranged at fixed time intervals (such as 100 milliseconds ms); particle filtering algorithm is a time series data processing algorithm used to estimate the state of nonlinear, non-Gaussian systems. It approximates the posterior probability distribution of the system state through a set of weighted random samples (called particles); fog droplet size distribution estimation refers to the probabilistic description of the statistical characteristics of the size of fog droplets in the current rain and fog environment output by the algorithm. It can usually be represented by a statistical value (such as median size) that characterizes the central tendency of the size.

[0026] Specifically, the system can achieve the following through a state estimation module: It aligns the spurious echo proportion feature value and contrast attenuation amplitude feature value of multiple consecutive time points (e.g., 10 time points within the most recent second) in a temporal sequence to construct a dual-channel observation vector sequence; it initializes a set containing a preset number of particles (e.g., 1000), each particle carrying an assumed state regarding droplet size and concentration; based on a preset state transition model describing the time-varying law of droplet size, and an observation model describing the physical relationship between particle size, concentration, and the two observed feature values, it performs iterative prediction, weighted update, and resampling operations on the particle set; finally, based on the converged state and weight of the particles, it calculates the estimated droplet size at the current time and outputs the estimated values ​​of consecutive time points as droplet size distribution estimates.

[0027] For example, based on the observation sequence of the past second, the system iteratively calculates using a particle filtering algorithm and outputs an estimated median droplet size distribution of 15 micrometers. This estimate indicates that the droplet size in the current environment is within a relatively large range.

[0028] Step S130: In response to the droplet size distribution estimation exceeding the preset threshold, the fusion weight of the lidar data and multispectral visual data is calculated based on the deviation between the false echo proportion feature value and the contrast attenuation amplitude feature value after normalization.

[0029] The preset threshold is a pre-defined critical value used to determine whether the droplet size is large enough to significantly affect the reliability of sensor data, such as 20 micrometers; normalization is a process of scaling feature values ​​to a specific numerical range (e.g., ...). The data standardization operation (within the range); the deviation here specifically refers to the scalar value obtained by subtracting the two normalized feature values, which is used to quantify the degree of divergence between the two values; the fusion weight is two weight coefficients respectively assigned to the lidar data and the multispectral visual data, which are used to adjust the contribution of each data in the subsequent fusion, and the sum of the two weight coefficients is a fixed value of 1.

[0030] Specifically, the system can achieve this through a weight calculation module: when the estimated droplet size distribution (e.g., median size) exceeds a preset threshold (e.g., 20 micrometers), it reads the spurious echo proportion feature value and the contrast attenuation amplitude feature value at the current moment; it normalizes the two feature values ​​respectively; it calculates the deviation between the two normalized values; and based on preset rules, combined with the deviation value and the relationship between the two normalized values, it dynamically calculates a set of fusion weights. Typically, the rules are designed to give greater weight to the data source with higher reliability.

[0031] For example, assume the current droplet size is estimated to be 25 micrometers, exceeding the 20-micrometer threshold. The current false echo proportion feature value is normalized to 0.8, and the contrast attenuation amplitude feature value is normalized to 0.4, with a deviation of 0.4. According to preset rules, since a high false echo proportion indicates significant interference with the LiDAR, while a small contrast attenuation amplitude indicates a relatively clear visual image, the system calculates and assigns fusion weights: LiDAR data weight is 0.3, and multispectral visual data weight is 0.7.

[0032] Step S140: Based on the fusion weight, the point cloud confidence sequence mapped by the false echo proportion feature value and the image sharpness sequence mapped by the contrast attenuation amplitude feature value are weighted and fused to generate a fusion confidence curve.

[0033] Among them, the point cloud confidence sequence refers to representing the reliability of point cloud data at each distance unit along the preset detection direction (from the sealing rope to the live wire) as a numerical sequence. This sequence is generated by substituting the false echo proportion feature value into a negatively correlated mapping function (such as 1 minus the feature value); the image sharpness sequence refers to representing the image sharpness corresponding to each position in the same detection direction as a numerical sequence. This sequence is generated by substituting the contrast attenuation amplitude feature value into a negatively correlated mapping function; weighted fusion refers to the operation of multiplying each position element of the above two sequences by the corresponding fusion weight and then adding them together; the fusion confidence curve is a one-dimensional curve distributed along the detection direction generated after weighted fusion, and its peak position corresponds to the probability of the existence of a target (rope or wire).

[0034] Specifically, the system can achieve this through a data fusion module: First, based on a preset mapping relationship, the false echo proportion feature value and the contrast attenuation amplitude feature value are converted into a point cloud confidence sequence and an image sharpness sequence distributed along the detection direction, respectively; then, the LiDAR data fusion weights calculated in step S130 are multiplied by the point cloud confidence sequence, and the multispectral visual data fusion weights are multiplied by the image sharpness sequence; finally, the two weighted sequences are added element by element to generate the final fusion confidence curve.

[0035] For example, the system generates a detection direction sequence of length 256. The false echo proportion feature value of 0.2 is then mapped using a mapping function. Convert to a point cloud confidence sequence, where each element is approximately 0.37; the contrast attenuation magnitude feature value of 0.7 is mapped using a function. Convert to an image sharpness sequence, where each element is approximately 0.25. Use fusion weights. When weighted fusion is performed, the fusion confidence curve will form two significant local peaks near the actual locations of the rope and wire.

[0036] Step S150: Based on the first peak position and the second peak position identified from the fusion confidence curve, calculate the distance between the sealing rope and the live conductor.

[0037] Among them, the first peak position and the second peak position refer to the coordinate index or physical distance along the detection direction corresponding to the two local maxima points identified from the fusion confidence curve; the spacing refers to the difference between the coordinates of the two peak positions, which, after unit conversion, is the actual distance between the sealing rope and the live wire.

[0038] Specifically, the system can achieve the following through the result calculation module: Execute a peak detection algorithm (e.g., find local maxima) on the fusion confidence curve generated in step S140; Based on the prior spatial distribution range of the sealing rope and the live conductor, filter and confirm the coordinates of the detected peaks, and mark them as the first peak position (corresponding to the sealing rope) and the second peak position (corresponding to the live conductor), respectively; Calculate the difference between the indices of the two peak positions, and convert the difference into the actual physical distance and output it according to the distance-index conversion ratio calibrated by the system (e.g., each index represents 0.01 meters (m)).

[0039] For example, the system identifies two significant peaks from the fusion confidence curve, with index positions of 120 and 185, respectively. Given that the system's calibrated distance resolution is 0.01 meters per index, the distance between the sealing rope and the live conductor is calculated as follows: Meters. This result will be output as a safety monitoring result.

[0040] Therefore, according to the above implementation method, the system achieves its function through five core steps: feature analysis, state estimation, weight calculation, data fusion, and distance output. Specifically, based on the acquired point cloud data and multispectral image data, the system analyzes the characteristic values ​​of the false echo ratio and contrast attenuation amplitude, thus quantitatively characterizing the degree of rain and fog interference on both the lidar and visual raw data. Based on the time series of the two characteristic values, a particle filtering algorithm is executed to output a fog droplet size distribution estimate, achieving real-time and accurate state estimation of the key environmental latent variable (fog droplet size). In response to the fog droplet size distribution estimate exceeding a preset threshold, a fusion weight is calculated based on the deviation of the two characteristic values ​​after normalization, establishing a dynamic weight allocation mechanism based on real-time environmental state and data quality differences. Based on the fusion weight, the point cloud confidence sequence and image sharpness sequence obtained from the feature value mapping are weighted and fused to generate a fusion confidence curve, completing the multi-source data integration after dynamic optimization based on reliability. The distance is calculated based on the double-peak positions identified from the fusion confidence curve, ultimately outputting accurate safety distance monitoring results.

[0041] Specifically, in this embodiment, the technical solution addresses the problem described in the background art where the lack of accurate modeling of the dynamic correlation between droplet size distribution and multi-source data reliability leads to inaccurate fusion weight allocation. This is achieved by introducing a particle filtering algorithm to estimate droplet size distribution in real time, transforming the abstract "environmental adaptability" assessment into tracking specific, quantifiable environmental state variables (droplet size), thus providing a direct and accurate physical basis for weight calculation. To address the resulting insufficient accuracy in safety distance judgment, a closed-loop data processing flow is constructed that "dynamically calculates weights in response to droplet size estimation exceeding the threshold." This allows the allocation of fusion weights to respond in real time to the fundamental impact of droplet size changes on the reliability of laser and visual data. When the environment favors a particular sensing mode, the algorithm automatically assigns it a higher confidence level. Therefore, the data processing method of this invention solves the technical problem of insufficient accuracy in safety distance judgment caused by the lack of a data processing algorithm based on real-time droplet size estimation using particle filtering to dynamically optimize multi-sensor data fusion weights, thereby improving the accuracy and reliability of high-altitude operation safety monitoring under complex meteorological conditions.

[0042] In another embodiment, Table 1 shows a comparison table of interference characteristics and particle filter particle size estimation of a multimodal sensing system under different environmental conditions. For example, during power transmission line grid closure operations, the system continuously monitors environmental changes and records data. At time T0, the weather is clear, serving as the baseline environment. At this time, the proportion of false echoes collected by the lidar is 10%, and the contrast attenuation of the multispectral image is 5%. Since neither indicator exceeds the preset threshold, the particle filter algorithm does not need to perform particle size estimation and is marked as "not exceeding the threshold". As the environment changes to time T1, light fog appears on site, and the environment description is marked as "small particle size dominant". The proportion of false echoes from the lidar rises to 15%, with slight interference; while the image contrast attenuation significantly increases to 45%, with significant interference. The system determines that particle size estimation is necessary, and after particle filtering, the output fog droplet particle size estimate is 8. The confidence level is high. As time progresses to T2, the weather changes to dense fog, and the environment description is marked as "large droplet-dominated." The proportion of false echoes from the lidar increases sharply to 65%, indicating severe interference; conversely, the image contrast attenuation decreases to 25%, indicating minor interference. This reflects that large droplets cause more severe visual obstruction, while their impact on laser scattering exhibits different characteristics. At this point, the system estimates the droplet size to be 55 mm. The confidence level is high. At time T3, the weather evolved into a complex mixture of rain and fog, and the environment was characterized as "extremely uneven particle size distribution." At this time, the false echo rate of the lidar reached as high as 80%, and the image contrast attenuation also reached 70%, with both channels suffering extremely severe interference. The system-estimated fog droplet size soared to 82. However, due to the drastic fluctuations in the data, which exceeded the optimal applicability of the physical model, it was marked as having low confidence. Based on the data in this table, the system will trigger a higher-level dynamic weight adjustment mechanism at time T3 to cope with extreme interference.

[0043] In another embodiment, Table 2 shows a comparison table of the dynamic weight adjustment logic and calculation process based on dual-channel normalization bias. For example, in real-time monitoring of power transmission line closure operations, the system sets the initial fusion weights of LiDAR and multispectral vision to 0.5 each. At time T0, the environment is clear with no significant interference, and the system maintains the initial weights, with the LiDAR weight WI remaining at 0.5. As the environment changes and we enter time T1, light fog appears on site. The system extracts the false echo ratio of LiDAR as 15%, which is mapped to parameter A=0.30 after normalization; it also extracts the image contrast attenuation amplitude as 45%, mapped to parameter B=1.125. The dual-channel bias value is then calculated. Since parameter B is significantly greater than parameter A, it indicates that the visual channel is more severely affected by fog. The system determines that the visual weight needs to be reduced and the laser weight increased. According to the weight adjustment formula, the laser weight WI is adjusted to 0.665, and the visual weight Wv is adjusted to 0.335. When time progresses to T2, the weather changes to dense fog. At this time, the proportion of false laser echoes surges to 65% (A=1.30), while the image attenuation decreases to 25% (B=0.625). The calculated deviation value D=0.675. Since parameter A is greater than parameter B, it indicates that the laser channel is severely interfered with by dense fog scattering. The system executes the opposite weight adjustment strategy, reducing the laser weight WI to 0.365 and increasing the visual weight Wv to 0.635. At time T3, the environment evolves into a mixed rain and fog state, and the interference in both channels is extremely severe. The proportion of false laser echoes reaches 80% (A=1.60), and the image attenuation reaches 70% (B=1.75). The calculated deviation value D=0.15 is small. The system determined that both sensors were severely malfunctioning at this point, so it made only a very small adjustment to the weights, and finally output the laser weight WI as 0.53 and the visual weight Wv as 0.47, and triggered a system alarm to prompt manual intervention for verification.

[0044] In another embodiment, Table 3 shows a comparison table of real-time calculation and safety judgment of the netting spacing based on dynamic weight adjustment and multimodal fusion results. For example, during live-line crossing netting operations, the system adjusts the fusion weights of lidar and multispectral vision in real time according to changes in environmental interference intensity, and performs target detection and spacing calculation accordingly. At time T0, the weather is clear, and the system uses a preset baseline weight (visual weight Wv is 0.5) for routine detection. At this time, pure visual detection shows that the netting rope position is 1.00m and the conductor position is 1.78m, and the system calculates the spatial distance between the two to be 0.78m. Since this distance is greater than the minimum limit of 0.70m set by the safety regulations, the system determines that the operation is safe and does not trigger an alarm. As the environment changes and we enter time T1, light fog appears on site, and vision is affected by fog interference, resulting in a large error. According to the logic above, the system lowers the visual weight Wv to 0.335, focusing on using reliable laser point cloud data. At this point, the rope position detected by pure vision shifted to 0.335m (affected by blurred images), while the lidar, with its strong penetrating power, accurately locked the conductor position at 1.02m. The system calculated a spacing of 0.75m based on the data corrected by the fusion strategy, still within the safe range, and did not trigger an alarm. As time progressed to T2, the fog intensified, and false signals increased in the lidar echo. The system then increased the visual weight Wv to 0.635, relying on high-resolution image data for primary judgment. At this time, pure vision accurately detected the conductor position at 0.99m, while the false peaks in the lidar point cloud due to interference were suppressed by the system (the detected value was displayed as 0.635m). The system calculated a spacing of 0.74m, which, although narrower than before, still met the safety requirement of being greater than 0.70m, and the system maintained its decision not to trigger an alarm. At T3, the environment evolved into an extremely harsh mixed rain and fog condition, and both channels of data were severely distorted. Although the system underwent a very minor weight adjustment (visual weight Wv was 0.53), due to extreme noise, the visual module was unable to extract valid rope peak values ​​(displaying "cannot extract"). Furthermore, the noise peak values ​​from the lidar misled the system into calculating the conductor position as 1.62m, resulting in a final calculated spacing of only 1.62m (an invalid calculation result influenced by erroneous peak values). Given that all sensors had failed and the calculation results were completely unreliable, the system determined that an abnormal process had been triggered, forcibly activating the highest-level alarm signal, prompting ground command personnel to immediately cease automated monitoring and switch to manual intervention to ensure operational safety.

[0045] In another embodiment, such as Figure 2The flowchart illustrates an intelligent detection method for safe distance during power transmission line closure operations based on multimodal perception. For example, during a live-line crossing closure operation on a high-voltage transmission line, the monitoring platform deployed a LiDAR scanning unit and a multispectral vision acquisition unit. During the operation, the system executes step S101 in real time, acquiring raw point cloud and image data of the closure rope area and the live conductor area. Suddenly, dense fog appears, causing numerous false echoes caused by fog droplets in the LiDAR echo, and the contrast between the foreground rope and the background conductor is severely reduced in the multispectral camera image due to fog obstruction. The system then proceeds to step S102, extracting the current proportion of false echoes and the contrast attenuation, using a particle filtering algorithm to process rain and fog interference, and estimating the median particle size of the fog droplets in the current environment to be approximately 18. Because the particle size exceeds the preset 10... To ensure safety, the system triggers the dynamic weight adjustment mechanism in step S103. Assuming the observed false echo percentage is 0.6 and the contrast attenuation is 0.4, the system calculates the normalization deviation after normalization processing. It determines that the visual data is more severely affected by fog, therefore automatically increasing the fusion weight of the LiDAR data to 0.7 and decreasing the visual data weight to 0.3. In step S104, the system performs weighted fusion of point cloud and image data according to the new weight scheme, generating a fusion confidence curve along the detection direction. The curve clearly shows two peaks, one at 11.5m and the other at 21.3m. The system identifies the 11.5m location as the sealing rope and the 21.3m location as the live wire. Finally, in step S105, the system calculates the spatial distance between the sealing rope and the live wire to be 9.8m. Because the distance was below the minimum 10m limit required by safety regulations, the system immediately triggered an alarm signal, reminding ground command personnel that there was a safety hazard at the current work location, and that the personnel should be instructed to immediately stop deploying and adjust the position of the netting.

[0046] In another embodiment, such as Figure 3 The diagram illustrates the logic flowchart for dynamically adjusting multimodal fusion weights based on dual-channel bias. For example, during real-time monitoring of a power transmission line closure operation, the system observes a sudden dense fog in the environment through a multispectral visual acquisition unit. At this time, the system executes the dynamic weight adjustment process: first, it extracts the current environmental features and obtains an estimated fog droplet size of 25. The system compares this estimated value with the preset safe particle size upper limit threshold of 10. The system compares the observed values ​​and determines that the current environment has significantly exceeded safety limits, triggering a dynamic weight adjustment mechanism. Subsequently, the system obtains the observed values ​​for the false echo percentage (0.6) and contrast attenuation (0.4). To fairly compare these two metrics with different dimensions, the system normalizes them, calculating a normalized false echo percentage of 0.75 and a normalized contrast attenuation of 0.5. Next, the system calculates the absolute difference between the two, yielding a dual-channel deviation of 0.25. Based on the relative magnitudes of the normalized echo percentage and attenuation, since the normalized false echo percentage (0.75) is greater than the normalized contrast attenuation (0.5), the system determines that the noise generated by rain and fog interference is more significant in the current LiDAR data, while the multispectral visual data, although showing a decrease in contrast, still retains more effective contour information. Therefore, the system implements a strategy of reducing the weight of laser and increasing the weight of vision, increasing the fusion weight of multispectral vision data to 0.65 and correspondingly reducing the fusion weight of lidar data to 0.35, ultimately obtaining an optimized weight allocation scheme for the current poor line-of-sight conditions, and using this to guide subsequent target detection and distance calculation.

[0047] In another embodiment, such as Figure 4 The flowchart illustrates a method for calculating the spatial distance between a power transmission line sluice gate rope and a live conductor based on weighted fusion and peak identification. For example, in a real-world test scenario of a power transmission line crossing a power line sluice gate, the system has already obtained the fusion weights of LiDAR and multispectral vision through preliminary data processing. The LiDAR weight is set to 0.6, and the multispectral vision weight is set to 0.4. The system then executes steps S202 to S204 to generate and fuse the confidence sequence: First, the system extracts the proportion of false echoes caused by fog droplets in the current LiDAR point cloud data as 0.2, maps it to a point cloud reliability score of 0.8, and after multiplying by a weight of 0.6, calculates the distribution of the weighted point cloud confidence sequence along the detection direction. Simultaneously, the system extracts the image contrast attenuation magnitude as 0.4, mapping it to an image sharpness score of 0.6. After multiplying by a weight of 0.4, the weighted image confidence sequence distribution is obtained as follows: The system sums the confidence scores at the same locations to form the final fused confidence curve data. Next, the system executes step S205 to identify local maxima in the fusion confidence curve. Analysis reveals two distinct peaks: the first peak occurs at a distance of 11.5m in the detection direction, with a fusion confidence level of 1.4; the second peak occurs at a distance of 21.8m in the detection direction, with a fusion confidence level of 0.95. The system marks these two locations as the safety net rope and the live conductor, respectively. Finally, step S206 is executed, where the system extracts the spatial coordinates corresponding to these two peaks, calculates the difference between them, determines the spatial distance between the safety net rope and the live conductor to be 10.3m, and transmits this coordinate data to the work monitoring platform for subsequent safety assessment.

[0048] In some embodiments, based on the time series of spurious echo proportion characteristic values ​​and contrast attenuation amplitude characteristic values, a particle filtering algorithm is executed to output a droplet size distribution estimate, including: The time-aligned spurious echo proportion feature value and the contrast attenuation amplitude feature value are used to construct a dual-channel observation vector sequence.

[0049] Temporal alignment refers to pairing two feature values ​​collected within the same time period and at the same timestamp; the dual-channel observation vector sequence is a two-dimensional data structure, where each row represents a time point and contains two elements. The first column is the feature value of the proportion of false echoes, and the second column is the feature value of the contrast attenuation amplitude. The entire sequence describes the changes of these two observations over time.

[0050] Specifically, the system can read feature value data from the most recent N consecutive sampling times (e.g., N=10) in the cache, ensuring that the timestamps of the two feature values ​​at each time point match perfectly. Then, these N pairs of data are arranged in chronological order to form an N×2 matrix, which is the dual-channel observation vector sequence. For example, if the system reads data sampled every 100ms within the most recent second, it will obtain 10 feature value pairs: This was constructed as a 10-row, 2-column observation sequence.

[0051] The particle set is initialized based on a preset number of state vectors, and iterative prediction, weighting and resampling are performed according to the state transition model and observation model to update the particle weights and states.

[0052] Here, the state vector is an array describing the hidden states of the system (droplet size and concentration in this case); the particle set is a collection of multiple state vectors, with each particle representing a possible assumption of the system state; the state transition model is a mathematical model describing how the system state (droplet size and concentration) evolves from the previous moment to the current moment; the observation model is a mathematical model describing the theoretical values ​​of the observed feature values ​​(false echo percentage, contrast attenuation amplitude) given a certain system state (hypothetical droplet size and concentration); prediction refers to advancing the state of each particle from the previous moment to the current moment according to the state transition model; weighting refers to calculating and updating the weight of each particle based on the difference between the actual observation value at the current moment (from the dual-channel observation vector sequence) and the predicted observation value calculated by the observation model, with a smaller difference resulting in a higher weight; resampling involves resampling the particle set according to the particle weights, eliminating low-weight particles and replicating high-weight particles to prevent particle degradation.

[0053] Specifically, the system can first be within a reasonable range of values ​​(e.g., particle size 1~100). M state vectors (e.g., M=1000) are randomly generated at a concentration of 0.1~10 g / m³, and each vector is assigned the same initial weight (1 / M) to initialize the particle ensemble. Then, for each time step, the following iterations are performed: (1) Prediction: Update the state vector carried by each particle according to the state transition model (e.g., a random walk model with random noise); (2) Weighting: For each particle, its current state (particle size) is weighted. ,concentration Substitute the values ​​into the observation model to calculate the corresponding predicted value of the spurious echo proportion. and the predicted value of contrast attenuation By comparison Compared with the actual observed value at the current time The differences (e.g., calculating the inverse of the Euclidean distance as the likelihood) are used to update the particle's weights. .

[0054] (3) Resampling: All weights are normalized, and then the entire particle set is resampled according to the normalized weights to obtain a new, equally weighted particle set. This new set represents the posterior probability distribution of the system state at the current moment. For example, the system initializes with 1000 particles. At a certain iteration moment, a particle's state is (particle size = 20...). (Concentration = 1.5 g / m³). The predicted value was obtained through observation model calculation. The actual observed value is... The system calculates the difference between the predicted and observed values ​​and assigns a higher weight to the particle accordingly. After weighting all particles, the system resamples, removing particles with lower weights (e.g., those with a particle size of 5). Particles with high weights (whose predicted values ​​differ greatly from the observed values) are eliminated, while high-weighted particles are replicated, and the state distribution of the particle set converges in a direction that is closer to the true situation.

[0055] Based on the iteratively updated particle states and weights, the posterior estimate of the droplet size at each time step is calculated.

[0056] The posterior estimate refers to the optimal estimate of the hidden state of the system (specifically, the droplet size) after incorporating the observation data at the current moment.

[0057] Specifically, after each iteration's weighting (before resampling), the system can use the droplet size state values ​​carried by all particles in the current particle set and their normalized weights to calculate the weighted average of these states as the posterior estimate of the droplet size at the current moment. The calculation formula is: Estimated value = Σ(Particle size value of the i-th particle * Normalized weight of the i-th particle). For example, suppose that at a certain moment, the particle size values ​​in the particle set are mainly concentrated at 15. and 25 Nearby. After weighted calculation, the particle size is 20. Nearby particles generally have higher weights. The system calculates a weighted average of all particle sizes, yielding a posterior estimate of the droplet size at the current moment of 18.7. .

[0058] The output consists of a time-series distribution of posterior estimates, which is used as an estimate of droplet size distribution.

[0059] Among them, time series distribution refers to a sequence or trend chart formed by arranging the posterior estimates calculated at multiple consecutive time points in chronological order.

[0060] Specifically, the system can maintain a fixed-length first-in-first-out queue or array to store the posterior estimates of droplet size calculated for the most recent K consecutive time intervals (e.g., K=50, corresponding to the most recent 5 seconds of data). Whenever a new estimate is calculated, it is added to this sequence, and the updated sequence is output as the droplet size distribution estimated by the system. For example, if the system runs continuously and outputs a posterior estimate every 100 milliseconds, the sequence of estimates output by the system in the most recent 5 seconds would be as follows: This sequence shows the trend of droplet size gradually increasing over time, which is the droplet size distribution estimate output by the system.

[0061] Therefore, according to the above implementation method, the system can use continuous observation data and a probabilistic algorithm such as particle filtering to dynamically and adaptively estimate the real-time state and changing trend of key environmental variables (droplet size) that cannot be directly measured, providing accurate data input for subsequent intelligent decision-making based on environmental state.

[0062] In some embodiments, the fusion weights of LiDAR data and multispectral visual data are calculated based on the deviation between the false echo proportion feature value and the contrast attenuation amplitude feature value after normalization, including: Read the characteristic values ​​of the proportion of false echoes and the characteristic value of the contrast attenuation amplitude at the current moment when the estimated droplet size distribution exceeds the preset threshold.

[0063] The preset threshold is a pre-set value stored in the system, used as a decision threshold for whether to initiate the dynamic weight calculation process. When the estimated droplet size distribution (usually expressed as median size) is greater than this threshold, it indicates that the rain and fog environment has had a significant and differentiated impact on the reliability of sensor data, and dynamic weight adjustment needs to be initiated.

[0064] Specifically, the system can continuously receive and monitor the droplet size distribution estimate from the state estimation module (which executes the particle filtering algorithm) in real time. Once the estimate is detected to exceed a preset threshold, the system immediately extracts the spurious echo proportion feature value and contrast attenuation amplitude feature value, calculated by the feature extraction module, from the data cache, which have the same timestamp as the estimate. For example, the preset droplet size threshold is 20. When the system detects that the posterior estimate of the droplet size at the current moment is 25... If the threshold is exceeded, the system will then read the characteristic value of the proportion of false echoes (e.g., 0.22) and the characteristic value of contrast attenuation amplitude (e.g., 0.65) that correspond to the timestamp.

[0065] Normalize the read spurious echo proportion feature value and contrast attenuation amplitude feature value to generate normalized echo proportion and normalized attenuation amplitude.

[0066] In this step, normalization specifically refers to a linear scaling operation, the purpose of which is to map two feature values ​​to the same numerical range (e.g., ...). This eliminates dimensional differences, facilitating fair comparisons and calculations in the future.

[0067] Specifically, the system can perform minimum-maximum normalization on the currently read feature value based on the possible range (maximum and minimum values) of each feature value obtained through pre-statistical analysis or experimental calibration under typical rainy and foggy weather conditions. The formula is: Normalized value = (Current feature value - Lower limit of feature value) / (Upper limit of feature value - Lower limit of feature value). For example, assuming the typical range of the feature value of false echo proportion is... The typical range of the contrast attenuation amplitude characteristic value is: For the currently read feature value The normalized echo ratio was calculated to be: Normalized attenuation magnitude .

[0068] Calculate the difference between the normalized echo percentage and the normalized attenuation amplitude.

[0069] Here, the difference value is defined as the absolute value of the difference between two normalized values, which is used to quantify the relative magnitude of the interference between the LiDAR and vision data.

[0070] Specifically, the system can perform simple scalar subtraction and take the absolute value of the result. The calculation formula is: Difference value = |Normalized echo percentage - Normalized attenuation amplitude|. For example, based on the normalized value obtained in the previous step... Calculate the difference value .

[0071] Based on preset rules, the fusion weights of LiDAR data and multispectral visual data are dynamically calculated according to the normalized echo ratio, normalized attenuation amplitude, and difference value.

[0072] The preset rules are a set of predefined logical judgments and numerical calculation criteria used to calculate the final weights based on the input parameters (two normalized feature values ​​and their difference). The core design principle of these rules is to give greater fusion weights to data sources that are less affected by interference in the current environment and have relatively higher data quality.

[0073] Specifically, the system can implement the following rules through the weight calculation module: (1) Compare the normalized echo ratio (denoted as ) ) and normalized decay magnitude (denoted as The size of ).

[0074] (2) Set the initial weight base for lidar data and multispectral visual data, for example, both are 0.5.

[0075] (3) If This indicates that the LiDAR is experiencing more severe interference. Therefore, based on the difference value (denoted as D), the LiDAR weight base value is lowered, and the multispectral vision weight base value is increased. The adjustment amount can be proportional to the difference value D, for example: the LiDAR weight base value. Multispectral visual weighting base .

[0076] (4) If If so, then the opposite operation will be performed.

[0077] (5) To ensure that the sum of the two weights is 1, the adjusted weight base is finally normalized: LiDAR fusion weight = LiDAR weight base / (LiDAR weight base + Multispectral vision weight base); Multispectral vision fusion weight = 1 - LiDAR fusion weight. For example, continuing from the previous example, .because According to the rules, the weight base of multispectral vision is lowered, while the weight base of LiDAR is increased. The calculated weight base for multispectral vision is: LiDAR weighting base After normalization, the lidar fusion weights Multispectral visual fusion weight = 0.395.

[0078] Therefore, according to the above implementation method, the system can automatically and dynamically calculate the optimal sensor data fusion weights based on the judgment of the current environmental state (particle size exceeding the threshold) and the real-time data quality assessment (normalized eigenvalues ​​and difference values) through a set of explicit and programmable rules, thereby ensuring that the fusion process always tilts towards the more reliable current data source.

[0079] In some embodiments, based on fusion weights, a weighted fusion is performed on the point cloud confidence sequence mapped by the false echo proportion feature value and the image sharpness sequence mapped by the contrast attenuation magnitude feature value to generate a fusion confidence curve, including: The spurious echo proportion feature value and the contrast attenuation amplitude feature value are respectively converted into the corresponding point cloud confidence sequence and image sharpness sequence.

[0080] In this context, transformation refers to the operation of expanding and mapping a scalar feature value into a one-dimensional sequence with a spatial distribution (along the detection direction). Each element of the point cloud confidence sequence represents the corresponding position along the detection direction; the LiDAR point cloud data reflects the confidence score of the probability of the target's existence, with higher scores indicating greater probability. Each element of the image sharpness sequence represents the corresponding position along the detection direction; the multispectral image data reflects the sharpness score of the target contour, with higher scores indicating better sharpness. These two sequences are typically generated based on a negative correlation mapping function, meaning that the higher the input feature value (interference level), the lower the output sequence value (confidence or sharpness).

[0081] Specifically, the system can predefine the detection direction, for example, taking the construction platform as the origin and pointing away from the live wire, and discretize it into L equally spaced detection units (e.g., L=256). For each detection unit index j (j from 1 to L), the system uses a mapping function F to convert the scalar characteristic value a (false echo ratio) into a confidence value for that location. Similarly, the scalar characteristic value b (contrast attenuation magnitude) is converted into a sharpness value. Functions F and G are typically monotonically decreasing, such as exponentially decreasing functions. Functions F and G must be designed to be monotonically decreasing; for example, they can be in the form of exponentially decreasing functions. Among them, the attenuation coefficient and This is a constant greater than 0, and its specific value needs to be determined through experimental calibration to reflect the actual impact of the proportion of false echoes or the magnitude of contrast attenuation on the data confidence level. In an exemplary calibration result, it can be taken as... Assume the current spurious echo proportion characteristic value is a = 0.2, and the contrast attenuation amplitude characteristic value is b = 0.7. Define the mapping function. , ,in Therefore, for all detection units j, each element of the generated point cloud confidence sequence is approximately Each element of the image sharpness sequence is approximately Therefore, two one-dimensional arrays with length L=256 and element values ​​of approximately 0.368 and 0.247 are generated.

[0082] The fusion weights of LiDAR data and multispectral visual data are applied to the corresponding point cloud confidence sequence and image sharpness sequence, respectively, to obtain the weighted point cloud sequence and image sequence.

[0083] Here, "applying" refers to multiplying the scalar weights with each element in the sequence. The weighted point cloud sequence is a new sequence obtained by multiplying each element of the point cloud confidence sequence by the LiDAR data fusion weights. The weighted image sequence is a new sequence obtained by multiplying each element of the image sharpness sequence by the multispectral visual data fusion weights.

[0084] Specifically, the system obtains two fusion weights from the weight calculation module, denoted as... (LiDAR weights) and (Multispectral visual weights), and satisfy Then, the system performs two element-wise scalar multiplication operations: for j from 1 to L, calculate... ,as well as This results in two new sequences, still of length L. For example, the sequence from the previous step and the assumed fusion weights are... Then each element of the weighted point cloud sequence becomes... The weighted image sequence becomes each element as follows: .

[0085] An element-wise fusion operation is performed on the weighted point cloud sequence and the image sequence to generate a fusion confidence curve along the detection direction from the sealing rope to the live conductor.

[0086] In this context, element-wise fusion specifically refers to the operation of adding the element values ​​at the same index position between two sequences. The fusion confidence curve is a new one-dimensional sequence obtained after the fusion operation. The value at each position combines the weighted LiDAR and visual information, and the peak on the curve indicates a high probability that a target (netting rope or live wire) exists at that position.

[0087] Specifically, the system processes the weighted point cloud sequence and weighted image sequences Perform element-wise addition. That is, for each detection unit index j (j ranges from 1 to L), calculate... The final result This is the desired fusion confidence curve, a one-dimensional array of length L describing the distribution of the overall confidence along the detection direction. For example, continuing the previous example, for each position j, the fusion confidence value is calculated as 0.110 + 0.173 = 0.283. Therefore, the generated fusion confidence curve is a sequence where each element has a value of approximately 0.283. In real-world scenarios, near the actual locations of the safety net ropes and live wires, the original point cloud and image will respond at these locations. Therefore, after the above mapping, weighting, and fusion, the curve will form obvious local peaks at these locations, rather than a flat straight line.

[0088] Therefore, according to the above implementation method, the system can combine the feature values ​​reflecting the global data quality with the sequence representation of spatial local information, and perform adaptive weighting through dynamic weights, and finally fuse them to generate a confidence curve that can highlight the real target location and suppress noise interference, laying the foundation for subsequent accurate identification of target location and calculation of distance.

[0089] In some embodiments, the distance between the sealing rope and the live conductor is calculated based on the first peak position and the second peak position identified from the fusion confidence curve, including: A peak detection algorithm is applied to the fusion confidence curve to analyze the first peak position located in the preset spatial interval of the sealing rope and the second peak position located in the preset spatial interval of the live conductor.

[0090] Peak detection algorithms are digital signal processing methods used to automatically identify local maxima (peak points) in a one-dimensional data sequence, such as the sliding window comparison method. The preset spatial interval for the safety net rope is a range of distances (represented by a sequence index) along the detection direction, pre-defined according to the construction plan and sensor deployment locations. The peak value of the safety net rope is expected to appear within this interval, for example, the index range. The preset spatial range for the charged conductor is another pre-defined distance range along the detection direction, within which the peak value of the charged conductor is expected to appear, such as the index range. The first peak position and the second peak position refer to the sequence index numbers corresponding to the peak points with the largest amplitude values ​​found by the algorithm within the two preset spatial intervals mentioned above.

[0091] Specifically, the system can employ a sliding window peak detection method: a sliding window of fixed width (e.g., 5 index points) is set, moving from the starting index of the fusion confidence curve to the ending index; when the value of the center point of the window is greater than the values ​​of all other points within the window, that center point is marked as a candidate peak point. Then, based on preset spatial interval constraints, the system filters out peak points from all candidate peak points that fall within preset spatial intervals for the sealing rope and the live conductor, respectively. Finally, within each interval, the peak point with the largest amplitude is selected, and its index is recorded as the first peak position and the second peak position, respectively. For example, the system performs peak detection on a fusion confidence curve of length 256, finding 4 candidate peaks with indices of 80, 120, 180, and 220, corresponding to amplitudes of 0.40, 0.55, 0.60, and 0.30, respectively. The preset sealing rope interval is... The section with live conductors is After screening, it fell into The peak values ​​in the interval are 80 and 120, with the amplitude (0.55) being larger at 120. Therefore, the first peak value is determined to be 120. The peak values ​​in the interval are 180 and 220, with the amplitude (0.60) at 180 being larger. Therefore, the second peak value is determined to be 180.

[0092] Based on the preset spatial interval constraints, the coordinates of the first peak position and the second peak position are confirmed.

[0093] Among them, coordinate confirmation refers to the process of verifying and finally assigning values ​​to the peak positions initially analyzed, so as to ensure that the selected peak positions are indeed located near the expected physical target and are the most significant responses within the interval.

[0094] Specifically, after obtaining the initial first and second peak positions, the system will check again whether these two positions strictly fall within their respective preset spatial intervals. If interference causes the initially analyzed positions to slightly exceed the interval boundaries, the system can perform a secondary search or reselect the point with the largest amplitude within the interval boundaries. After confirmation, these two index values ​​are used as the final coordinates for calculating the spacing. For example, continuing the previous example, the initially analyzed first peak position of 120 and second peak position of 180 both conform to the preset intervals (…). The system confirmed the coordinates and finally output the coordinates of the first peak position as 120 and the coordinates of the second peak position as 180.

[0095] Calculate the difference between the coordinates of the first peak position and the coordinates of the second peak position, and output the difference as the distance between the sealing rope and the live conductor.

[0096] The difference refers to the arithmetic difference between the coordinate indices of the two peak positions. The spacing is the actual distance after physical calibration and conversion, and its value is equal to the difference multiplied by the distance resolution pre-calibrated by the system (i.e., the actual physical length represented by each sequence index).

[0097] Specifically, the system first calculates the index difference between the two confirmed coordinates: Spacing Index Difference = |Second Peak Position Coordinate - First Peak Position Coordinate|. Then, the system reads the pre-calibrated and stored distance resolution parameter R (unit: meters / index). Finally, it calculates the actual physical spacing = Spacing Index Difference * R, and outputs the result as the final spacing for safety monitoring. For example, given a system distance resolution R = 0.01 meters / index, a confirmed first peak position coordinate of 120, and a second peak position coordinate of 180, the index difference is calculated as |180 - 120| = 60. Therefore, the actual spacing = 60 * 0.01 meters = 0.6 meters. The system outputs "The spacing between the sealing net rope and the live conductor is 0.6 meters."

[0098] Therefore, according to the above implementation method, the system can stably and accurately locate the positions of two key targets from the fused confidence curve through peak detection constrained by spatial prior knowledge, and convert them into precise physical distances according to the system calibration parameters, thereby reliably outputting the core results of safety monitoring of the net enclosure operation.

[0099] In some embodiments, a particle set is initialized based on a preset number of state vectors, and iterative prediction, weighting, and resampling are performed according to a state transition model and an observation model to update particle weights and states, including: A set of particles is generated and initialized within a preset parameter space, where each particle is associated with a state vector and assigned an initial weight.

[0100] Here, the parameter space refers to the two-dimensional search space consisting of the numerical range of all possible values ​​for the system state (droplet size and concentration). The initial weights are the same importance score assigned to each particle at the start of the algorithm, with the sum of the initial weights for all particles being 1.

[0101] Specifically, the system can pre-set reasonable value ranges for state variables (particle size d, concentration c), such as particle size. ,concentration Within this parameter space, the system independently generates M state vectors (M being a preset number, such as 1000) through methods such as uniform random sampling or Gaussian distribution sampling. Each state vector... Associated with a particle, and assigned the same initial weight to each particle. For example, the system generates M=1000 particles. The first particle is randomly assigned a state vector (particle size = 15.3 μm, concentration = 2.1 g / m³) with an initial weight of 0.001. The second particle's state could be (particle size = 45.7 μm). (Concentration = 0.5 g / m³), with an initial weight of 0.001. This process continues until 1000 particles covering the parameter space are generated.

[0102] State prediction is performed based on the state transition model, and the dual-channel observation prediction value corresponding to each particle is calculated based on the observation model.

[0103] State prediction refers to using a state transition model to advance the state of each particle at time k-1 to the current time k, thus obtaining the predicted state of that particle. Dual-channel observation prediction refers to a two-dimensional vector composed of the theoretically predicted false echo proportion and the predicted contrast attenuation amplitude, calculated by substituting the predicted state of each particle into the observation model.

[0104] Specifically, at the beginning of each iteration (processing the data at time k), the system performs the following operation for the i-th particle in the particle set: (1) Prediction: Based on the state transition model Calculate the predicted state of the particle at time k. Here, x represents the state vector (particle size, concentration), and f is the state transition function (which can be simplified to a random walk). It is process noise.

[0105] (2) Calculate the predicted observations: The predicted state Substitute into the observation model Calculate the corresponding dual-channel observation prediction values. ,in It is the predicted proportion of false echoes. This is the predicted contrast decay rate. For example, for a certain particle, its state at time k-1 is (particle size = 18 micrometers, concentration = 1.8 g / m³). This is processed using a state transition model (assuming a mean of 0 and a standard deviation of 1 is added). and After prediction using Gaussian noise, the predicted state at time k is obtained as (particle size = 18.5). The concentration was 1.75 g / m³. Substituting this predicted state into the observation model (which establishes the physical relationship between particle size, concentration and the two observed values), the predicted values ​​for the dual-channel observation were calculated as follows: (predicted false echo ratio = 0.25, predicted contrast attenuation amplitude = 0.52).

[0106] The weights of each particle are calculated and updated based on the difference between the predicted values ​​from dual-channel observations and the dual-channel observation vector sequence.

[0107] Here, the difference refers to the degree of inconsistency between the observed predicted value of each particle and the actual observed value at the current moment (from the dual-channel observation vector sequence), usually measured by Euclidean distance or Mahalanobis distance. Weight update refers to calculating a likelihood based on this difference and multiplying this likelihood by the particle's prior weight to obtain the particle's posterior weight.

[0108] Specifically, the system reads the actual observed values ​​from the dual-channel observation vector sequence at time k. For the i-th particle, the system calculates its observed prediction value. Compared with actual observed values Differences between The smaller the difference, the closer the assumed state of the particle matches the actual situation. The system determines the appropriate state based on the difference. Calculate the likelihood of the particle. (For example, (where σ is the standard deviation of the observation noise). Then, the weights of the particle are updated: After traversing all particles, all weights are normalized so that their sum equals 1. For example, the actual observed value at the current moment is... For the particle in the example above, its observed predicted value is (0.25, 0.52). Calculate the Euclidean distance difference. Assuming the standard deviation of the observed noise σ = 0.1, calculate the likelihood. Assuming the prior weight of this particle is 0.001, the updated non-normalized weight is 0.001 * 0.914 = 0.000914. After all particles have completed this operation, the system performs weight normalization.

[0109] The particle set is resampled based on the updated weights, the particle state and distribution are updated, and the next iteration is triggered.

[0110] Resampling is an operation that resamples a particle set based on particle weights, aiming to replicate high-weight particles and eliminate low-weight particles, thereby avoiding particle degradation (i.e., the weights of most particles tending to 0). Updating particle states and distribution means that after resampling, the state vector distribution of the particle set will be more concentrated in the high-likelihood region, thus better approximating the true posterior state distribution of the system.

[0111] Specifically, the system resamples the current particle set based on the normalized particle weights. A common method is system resampling: generating M particles in... Random numbers uniformly distributed across intervals are used. Based on the cumulative distribution function of particle weights, the index of each random number corresponds to the sampled particle. Particles with higher weights have a higher probability of being selected multiple times, while particles with lower weights may not be selected. The selected particles (which may be repeated) form a new particle set, in which all particles are assigned the same weight 1 / M. This new set represents the posterior distribution of the current state and serves as the starting point for processing data at the next time step (k+1), thus triggering the next iteration. For example, suppose there are 5 particles with normalized weights... During resampling, the second particle with a higher weight (weight 0.5) has a higher probability of being sampled multiple times, while the fifth particle with a lower weight (weight 0.05) is likely not sampled at all. The new particle set obtained after resampling may contain: particle 2, particle 2, particle 3, particle 2, and particle 4. The weights of all particles in the new set are reset to 0.2, and the particle state distribution is more concentrated in the region represented by the previously high-weighted particles 2 and 3.

[0112] Therefore, according to the above implementation method, the system can dynamically converge the particle set and its weights to the optimal estimate of the true hidden state of the system (droplet size distribution) through a series of iterative data processing steps such as initialization, prediction, weighting, and resampling, thereby providing the key environmental state input for the entire monitoring algorithm.

[0113] In some embodiments, based on preset rules, the fusion weights of LiDAR data and multispectral visual data are dynamically calculated according to the normalized echo ratio, normalized attenuation amplitude, and difference value, including: Perform a comparison calculation on the normalized echo percentage and the normalized attenuation magnitude.

[0114] The comparison operation is a logical judgment operation that aims to determine which of the two input values ​​is larger. Its output is a binary logical value (true or false), which is used to indicate the direction of subsequent weight adjustments.

[0115] Specifically, the system can implement this step using a simple numerical comparator. This comparator takes two inputs: the normalized echo ratio (denoted as ) ) and normalized decay magnitude (denoted as Comparator judgment Is it greater than And output the result of the judgment (True or False). For example, suppose... Perform comparison operations. Since 0.44 > 0.65 is not true, the output result is False, indicating that the current normalized attenuation amplitude is greater than the normalized echo ratio.

[0116] Based on the comparison results and the difference values, adjustment instructions are generated for the initial weight base of LiDAR data and multispectral visual data.

[0117] The initial weight base is the initial reference weight assigned to the LiDAR data and multispectral vision data. It is typically set to two equal values ​​before dynamic adjustment, for example, 0.5 each. The adjustment command is a data structure or logic signal containing two key pieces of information: (1) The direction of adjustment, i.e., which sensor’s weight base should be increased and which should be decreased, is determined by the comparison results; (2) The adjustment range, which is usually proportional to the size of the difference value. The larger the difference value, the larger the adjustment range.

[0118] Specifically, the system can be designed with a rule processor. This processor receives the comparison result (C) and the difference value (D). If the comparison result is true ( If the comparison result is false, the following instruction is generated: "Lower the base weight of the LiDAR, increase the base weight of the multispectral vision, and adjust the magnitude in relation to D." If the comparison result is false, the opposite instruction is generated. The adjustment magnitude can be quantified by a scaling factor proportional to D (e.g., D / 2). For example, continuing from the previous example, if the comparison result C=False, the difference value D=0.21. According to the preset rule, when C is False, it means that the visual interference is more severe ( (The difference is greater), therefore, we should "lower the multispectral visual weight base and increase the lidar weight base". The adjustment magnitude is quantized as D / 2=0.21 / 2=0.105. The generated adjustment command can be expressed as: {Target: visual weight base, Operation: Decrease, Quantity: 0.105; Target: lidar weight base, Operation: Increase, Quantity: 0.105}.

[0119] According to the adjustment instruction, the initial weight base is adjusted in the opposite direction to obtain the adjusted weight base.

[0120] In this context, reverse adjustment refers to making opposite numerical changes to the two weight bases, i.e., increasing one while decreasing the other by the same amount, to ensure that the sum of the two bases remains unchanged (e.g., still 1) before normalization. The adjusted weight bases are the intermediate results obtained after performing the addition and subtraction operations.

[0121] Specifically, the system sets the initial weight base for both LiDAR data and multispectral visual data to 0.5. The system parses the adjustment commands, performing addition (initial value + adjustment amount) on weight bases specified as "increased" and subtraction (initial value - adjustment amount) on weight bases specified as "decreased". For example, the initial weight base is: Based on the adjustment instructions generated in the previous step, the visual weight base is reduced: Increment the laser weight base: The adjusted weight base is obtained as follows: .

[0122] Normalization is performed on the adjusted weight base to obtain the final fusion weight with a fixed sum.

[0123] In this context, normalization specifically refers to the process of converting a set of weight bases into final weights that sum to 1. The final fusion weights are weight coefficients output after normalization calculations, which can be directly used for data fusion, and the sum of the two weights is strictly equal to 1.

[0124] Specifically, the system adjusts the two weight bases (denoted as...). and Normalization is performed. The calculation formula is: final fusion weight of LiDAR. Multispectral visual final fusion weights .because The sum remains 1 (0.605 + 0.395 = 1). This step is an identity operation in this example, but it ensures that the sum of the output weights is a fixed value of 1 in all cases (such as when there are small rounding errors in the adjustment calculation). For example, based on the adjusted weight base. The total calculated value is 1.0. Normalized calculation: LiDAR fusion weight = 0.605 / 1.0 = 0.605; Multispectral visual fusion weight = 0.395 / 1.0 = 0.395. The final output fusion weight is... .

[0125] Therefore, according to the above implementation method, the system can combine the abstract environmental interference comparison result (comparison operation) with the quantitative difference of the interference degree (difference value), automatically generate accurate weight adjustment amount through a set of clear and programmable rules, and output a set of final fusion weights that reflect the real-time reliability difference of sensor data and meet the mathematical requirements of the fusion algorithm (sum value is 1) after robust normalization processing.

[0126] Figure 5 This is a structural block diagram of a monitoring system for construction netting operation without scaffolding according to an embodiment of the present invention.

[0127] like Figure 5 As shown, the monitoring system for the construction netting operation without a scaffolding includes: The monitoring feature extraction module 210 is used to analyze the false echo proportion feature value and contrast attenuation amplitude feature value based on the acquired point cloud data and multispectral image data of the netting operation area.

[0128] The distribution estimation output module 220 is used to execute a particle filtering algorithm based on the time series of spurious echo proportion feature values ​​and contrast attenuation amplitude feature values ​​to output droplet size distribution estimates.

[0129] The fusion weight generation module 230 is used to calculate the fusion weight of lidar data and multispectral visual data based on the deviation between the false echo proportion feature value and the contrast attenuation amplitude feature value after normalization when the estimated droplet size distribution exceeds a preset threshold.

[0130] The confidence curve generation module 240 is used to generate a fused confidence curve by weighted fusion of the point cloud confidence sequence mapped by the false echo proportion feature value and the image sharpness sequence mapped by the contrast attenuation amplitude feature value.

[0131] The monitoring result generation module 250 is used to calculate the distance between the sealing rope and the live conductor based on the first peak position and the second peak position identified from the fusion confidence curve.

[0132] The specific functions and examples of each module and submodule of the device in this embodiment of the invention can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.

[0133] According to embodiments of the present invention, the above-described method of the present invention can be applied to a computer device and a readable storage medium.

[0134] Figure 6A schematic block diagram of an example computer device 600 that can be used to implement embodiments of the present invention is shown. The computer device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The computer device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0135] like Figure 6 As shown, the computer device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. The RAM 603 may also store various programs and data required for the operation of the computer device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0136] Multiple components in computer device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of monitors, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows computer device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0137] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as a method for monitoring construction netting operations without scaffolding. For example, in some embodiments, a method for monitoring construction netting operations without scaffolding can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on the computer device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of a method for monitoring construction netting operations without scaffolding described above can be performed. Alternatively, in other embodiments, the computing unit 601 may be configured by any other suitable means (e.g., by means of firmware) to perform a method for monitoring a non-scaffolding construction netting operation.

[0138] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0139] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0140] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0141] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0142] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0143] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers or servers in a distributed system.

[0144] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0145] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for monitoring netting operations during construction without scaffolding, characterized in that, include: Based on the point cloud data and multispectral image data of the acquired netting operation area, the characteristic values ​​of false echo proportion and contrast attenuation amplitude are obtained by analysis. Based on the time series of spurious echo proportion feature values ​​and contrast attenuation amplitude feature values, a particle filtering algorithm is executed to output a droplet size distribution estimate. In response to the droplet size distribution estimation exceeding a preset threshold, the fusion weight of lidar data and multispectral visual data is calculated based on the deviation between the false echo proportion feature value and the contrast attenuation amplitude feature value after normalization. Based on the fusion weights, the point cloud confidence sequence mapped by the false echo proportion feature value and the image sharpness sequence mapped by the contrast attenuation amplitude feature value are weighted and fused to generate a fusion confidence curve. Based on the first peak position and the second peak position identified from the fusion confidence curve, the distance between the sealing rope and the live wire is calculated.

2. The method according to claim 1, characterized in that, The time series based on the spurious echo proportion feature value and the contrast attenuation amplitude feature value is used to execute a particle filtering algorithm and output a droplet size distribution estimate, including: The time-aligned spurious echo proportion feature value and the contrast attenuation amplitude feature value are used to construct a dual-channel observation vector sequence; The particle set is initialized based on a preset number of state vectors, and iterative prediction, weighting and resampling are performed according to the state transition model and observation model to update the particle weights and states. Based on the iteratively updated particle states and weights, the posterior estimate of the droplet size at each time step is calculated. The time-series distribution composed of the posterior estimates is output as the droplet size distribution estimate.

3. The method according to claim 2, characterized in that, The step of calculating the fusion weight of LiDAR data and multispectral visual data based on the deviation between the false echo proportion feature value and the contrast attenuation amplitude feature value after normalization includes: Read the spurious echo proportion feature value and contrast attenuation amplitude feature value at the current moment when the estimated droplet size distribution exceeds a preset threshold; Normalization is performed on the read false echo proportion feature value and contrast attenuation amplitude feature value to generate normalized echo proportion and normalized attenuation amplitude. Calculate the difference between the normalized echo ratio and the normalized attenuation amplitude; Based on preset rules, the fusion weights of LiDAR data and multispectral visual data are dynamically calculated according to the normalized echo ratio, normalized attenuation amplitude, and difference value.

4. The method according to claim 3, characterized in that, The step of weighted fusion of the point cloud confidence sequence mapped by the false echo proportion feature value and the image sharpness sequence mapped by the contrast attenuation magnitude feature value, based on the fusion weight, to generate a fusion confidence curve includes: The false echo proportion feature value and the contrast attenuation amplitude feature value are respectively converted into the corresponding point cloud confidence sequence and image sharpness sequence; The fusion weights of the lidar data and multispectral visual data are applied to the corresponding point cloud confidence sequence and image sharpness sequence, respectively, to obtain the weighted point cloud sequence and image sequence. An element-wise fusion operation is performed on the weighted point cloud sequence and the image sequence to generate the fusion confidence curve along the detection direction from the sealing rope to the live conductor.

5. The method according to claim 4, characterized in that, The calculation of the distance between the sealing rope and the live conductor based on the first peak position and the second peak position identified from the fusion confidence curve includes: A peak detection algorithm is applied to the fusion confidence curve to analyze the first peak position located in the preset spatial interval of the sealing rope and the second peak position located in the preset spatial interval of the live conductor. Based on the preset spatial interval constraints, the coordinates of the first peak position and the second peak position are confirmed. Calculate the difference between the coordinates of the first peak position and the coordinates of the second peak position, and output the difference as the distance between the sealing rope and the live wire.

6. The method according to claim 2, characterized in that, The process of initializing the particle set based on a preset number of state vectors and performing iterative prediction, weighting, and resampling according to the state transition model and observation model to update particle weights and states includes: A set of particles is generated and initialized within a preset parameter space, where each particle is associated with a state vector and assigned an initial weight. State prediction is performed based on the state transition model, and dual-channel observation prediction values ​​for each particle are calculated based on the observation model. The weights of each particle are calculated and updated based on the difference between the dual-channel observation predictions and the dual-channel observation vector sequence. The particle set is resampled based on the updated weights, the particle state and distribution are updated, and the next iteration is triggered.

7. The method according to claim 3, characterized in that, The method, based on preset rules, dynamically calculates the fusion weights of LiDAR data and multispectral visual data according to the normalized echo ratio, normalized attenuation amplitude, and difference value, including: Perform a comparison operation on the normalized echo ratio and the normalized attenuation amplitude; Based on the comparison results and the difference values, an adjustment instruction is generated for the initial weight base of the lidar data and the multispectral visual data. According to the adjustment instruction, a reverse adjustment is applied to the initial weight base to obtain the adjusted weight base; The adjusted weight base is normalized to obtain the final fusion weight with a fixed sum.

8. A monitoring system for construction netting operations without scaffolding crossings, characterized in that, include: The monitoring feature extraction module is used to analyze the point cloud data and multispectral image data of the acquired netting operation area to obtain the feature values ​​of false echo proportion and contrast attenuation amplitude. The distribution estimation output module is used to execute the particle filtering algorithm based on the time series of spurious echo proportion feature values ​​and contrast attenuation amplitude feature values ​​to output droplet size distribution estimates. The fusion weight generation module is used to calculate the fusion weight of lidar data and multispectral visual data based on the deviation between the false echo proportion feature value and the contrast attenuation amplitude feature value after normalization when the estimated droplet size distribution exceeds a preset threshold. The confidence curve generation module is used to perform weighted fusion of the point cloud confidence sequence mapped by the false echo proportion feature value and the image sharpness sequence mapped by the contrast attenuation amplitude feature value based on the fusion weight, and generate a fusion confidence curve. The monitoring result generation module is used to calculate the distance between the sealing rope and the live conductor based on the first peak position and the second peak position identified from the fusion confidence curve.

9. A computer device, characterized in that, include: At least one processor; and a memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, in, Computer instructions are used to cause a computer to perform the method according to any one of claims 1-7.

Images