Foundation pit construction parameter extraction method and device based on point cloud data and medium

Abstract

The invention provides a foundation pit construction parameter extraction method and device based on point cloud data and a medium, and relates to the technical field of data processing, and the method comprises the steps: collecting three-dimensional point cloud data of a foundation pit excavated by a power transmission line; preprocessing the three-dimensional point cloud data to obtain regional point cloud data of each key feature region, the key feature regions at least comprising a side wall region, a bottom surface region and an excavation boundary region; inputting the region point cloud data of each key feature region into a pre-trained key feature recognition model for feature recognition to obtain key geometric features of the foundation pit, the key geometric features of the foundation pit at least comprising side wall bus features, bottom surface contour line features and excavation boundary features; based on the key geometrical characteristics of the foundation pit, foundation pit construction parameters are calculated through a geometrical calculation model, and the foundation pit construction parameters at least comprise foundation pit overall perpendicularity deviation, foundation pit excavation depth, side wall flatness and excavation size parameters. According to the method, intelligent and accurate extraction of the foundation pit construction parameters can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a pit construction parameter extraction method and device based on point cloud data and a medium. BACKGROUND

[0002] At present, the electric power infrastructure industry is accelerating the transformation from the traditional construction control mode relying on experience to the digital and data-driven quality control. As the core stress structure of the transmission line excavation foundation, the pit directly determines the bearing capacity and long-term operation safety of the power grid tower through key construction parameters such as verticality, excavation depth, and side wall flatness. Therefore, high-precision and high-efficiency detection of pit construction quality has become one of the core needs of the digital transformation of electric power infrastructure.

[0003] However, the current pit quality detection in the industry still mainly relies on manual measurement, and tools such as tape measure, plummet, and total station are used to complete data collection. Multiple people are required to set up equipment, manually record and calculate data, which is significantly disturbed by environmental factors such as rugged terrain, wind, rain, and strong light. In addition, subjective errors exist due to differences in personnel operation experience. Single-pit sampling and inspection is time-consuming and unstable in precision, which has been difficult to adapt to the strict requirements of modern electric power engineering on construction efficiency and quality control. It is urgent to break through this limitation through digital technology. SUMMARY

[0004] Therefore, the present application provides a pit construction parameter extraction method and device based on point cloud data, which can realize intelligent and accurate extraction of pit construction parameters.

[0005] According to a first aspect of the present application, a pit construction parameter extraction method based on point cloud data is provided, which comprises: Collecting three-dimensional point cloud data of a transmission line excavation pit; Preprocessing the three-dimensional point cloud data to obtain regional point cloud data of each key feature region, wherein the key feature region at least includes a side wall region, a bottom surface region, and an excavation boundary region; Inputting the regional point cloud data of each key feature region into a pre-trained key feature recognition model for feature recognition to obtain pit key geometric features, wherein the pit key geometric features at least include side wall bus feature, bottom surface contour line feature, and excavation boundary feature; Based on the pit key geometric features, calculating pit construction parameters through a geometric calculation model, wherein the pit construction parameters at least include pit overall verticality deviation, pit excavation depth, side wall flatness, and excavation size parameters.

[0006] According to a second aspect of the present application, a pit construction parameter extraction device based on point cloud data is provided, which comprises: The collection module is configured to collect three-dimensional point cloud data of the excavated foundation pit of the power transmission line. The processing module is configured to preprocess the three-dimensional point cloud data to obtain regional point cloud data of each key feature region, wherein the key feature region at least includes a side wall region, a bottom surface region, and an excavation boundary region. The input module is configured to input the regional point cloud data of each key feature region into a pre-trained key feature recognition model to perform feature recognition, and obtain key geometric features of the foundation pit, wherein the key geometric features of the foundation pit at least include side wall generatrix features, bottom surface contour line features, and excavation boundary features. The calculation module is configured to calculate foundation pit construction parameters based on the key geometric features of the foundation pit through a geometric calculation model, wherein the foundation pit construction parameters at least include foundation pit overall perpendicularity deviation, foundation pit excavation depth, side wall flatness, and excavation size parameters.

[0007] According to a third aspect of the present application, a storage medium having a computer program stored thereon is provided, wherein the program is executed by a processor to implement the above-mentioned foundation pit construction parameter extraction method based on point cloud data.

[0008] According to a fourth aspect of the present application, an electronic device is provided, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-mentioned foundation pit construction parameter extraction method based on point cloud data.

[0009] Through the above technical solutions, the present application provides a foundation pit construction parameter extraction method, device and medium based on point cloud data. Through three-dimensional point cloud data collection, the method can replace the data collection mode of traditional manual tape measure, plumb ball, total station instrument and other tools, without the need for multiple people to cooperate to set up equipment, and can effectively reduce the interference of environmental factors such as rugged terrain, wind and rain, and strong light on the detection process. After preprocessing, the point cloud data of key feature regions such as side walls, bottom surfaces, and excavation boundaries is accurately segmented, and the key geometric features such as foundation pit side wall generatrix, bottom surface contour line, and excavation boundary are automatically extracted by combining a pre-trained key feature recognition model, which can avoid subjective errors caused by differences in personnel operation experience. Finally, the overall perpendicularity deviation of the foundation pit, the excavation depth, the side wall flatness, and the excavation size parameters are automatically calculated by a geometric calculation model, without the need for manual recording and calculation of data, which can significantly shorten the time consumption of single foundation pit detection, improve the stability of detection precision, and ultimately break through the limitations of traditional manual measurement, accurately adapt to the strict requirements of modern power engineering on construction efficiency and quality control.

[0010] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented in accordance with the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a method for extracting foundation pit construction parameters based on point cloud data, provided in an embodiment of this application, is shown. Figure 2 A flowchart illustrating a method for extracting foundation pit construction parameters based on point cloud data, according to another embodiment of this application, is shown. Figure 3 This illustration shows a structural schematic diagram of a foundation pit construction parameter extraction device based on point cloud data provided in an embodiment of this application; Figure 4 A schematic diagram of a device for extracting foundation pit construction parameters based on point cloud data, provided in another embodiment of this application, is shown. Detailed Implementation

[0012] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0013] Currently, the quality inspection of foundation pits in the industry still mainly relies on manual measurement, using tools such as measuring tapes, plumb bobs, and total stations to complete data collection. This requires multiple people to work together to set up equipment and manually record and calculate data. It is not only significantly affected by environmental factors such as rugged terrain, wind, rain, and strong sunlight, but also by subjective errors due to differences in the operating experience of personnel. The sampling inspection of a single foundation pit is time-consuming and the accuracy is unstable, which is no longer suitable for the stringent requirements of modern power engineering for construction efficiency and quality control. There is an urgent need to overcome this limitation through digital technology.

[0014] To address the aforementioned technical problems, embodiments of the present invention provide a method for extracting foundation pit construction parameters based on point cloud data, such as... Figure 1 As shown, the method includes: Step 110: Collect three-dimensional point cloud data of the excavation pit for the transmission line.

[0015] Among them, the excavation pit of the transmission line refers to the pit structure with a specific depth and outline formed on the ground by manual or mechanical excavation in the construction of transmission line projects to build the foundation of core power facilities such as iron towers; three-dimensional point cloud data refers to a dataset composed of a large number of discrete three-dimensional spatial coordinate points collected by scanning the surface of the excavation pit of the transmission line through technologies such as laser scanning.

[0016] For the embodiments of the present disclosure, in the construction site of the power transmission line excavation foundation pit, the staff can select a three-dimensional data acquisition device (such as a ground three-dimensional laser scanner or a handheld mobile device integrated with a laser scanning module) suitable for complex outdoor environments according to the terrain (such as mountains or plains) and the actual size of the foundation pit. The device is first positioned and calibrated on site, and the coordinate deviation caused by the terrain slope is corrected in combination with the inertial measurement unit (IMU) to ensure the accuracy of the acquisition. Then, the scanning path is planned according to the principle of covering the entire key area. After starting the device, the device transmits a laser beam and receives a reflected signal to calculate the three-dimensional spatial coordinates of each reflection point in real time, and finally generate three-dimensional point cloud data that can completely represent the geometric shape of the key areas such as the side wall, bottom surface and excavation boundary of the foundation pit.

[0017] By replacing the traditional manual data acquisition method using a tape measure and a plummet with a three-dimensional point cloud data acquisition technology, the device can be set up without the need for multiple people to collaborate, effectively solving the problem of difficulty in setting up the device in rugged outdoor terrain, and the problem of interference with manual observation caused by wind, rain and strong light. At the same time, the large number of three-dimensional coordinate points acquired by the device can completely capture the geometric details of each key area of the foundation pit, avoiding subjective errors or omissions of key areas caused by differences in operating experience in manual measurement. The device provides high-precision, full-dimensional basic data for subsequent preprocessing, feature recognition and construction parameter calculation, improving the efficiency and data reliability of foundation pit quality detection from the source of the detection process, and breaking through the limitations of traditional manual measurement.

[0018] Step 120, preprocessing the three-dimensional point cloud data to obtain regional point cloud data of each key feature area.

[0019] The preprocessing of the three-dimensional point cloud data refers to a series of operations for targeted optimization and processing of the original foundation pit three-dimensional point cloud data collected, and the core purpose is to eliminate data noise, simplify data size and extract effective areas to provide high-quality and high-availability data basis for subsequent key feature recognition. The preprocessing usually includes key links such as denoising, downsampling and region segmentation. The key feature area refers to the core area in the three-dimensional shape of the foundation pit that plays a decisive role in extracting geometric features and calculating construction parameters. In the context of the power transmission line excavation foundation pit, the key feature area mainly includes, but is not limited to, the side wall area, the bottom surface area and the excavation boundary area related to the perpendicularity, the excavation depth, the flatness and the excavation size. The side wall area refers to the point cloud area corresponding to the vertical or inclined side of the foundation pit. The bottom surface area refers to the point cloud area corresponding to the plane or approximately plane part of the bottom of the foundation pit. The excavation boundary area refers to the point cloud area corresponding to the contour line at the junction of the top of the foundation pit and the ground.

[0020] For the embodiments of the present disclosure, for the original three-dimensional point cloud data of the excavated foundation pit of the power transmission line collected, a statistical filtering algorithm is first used to denoise the data, isolated interference point clouds generated due to field environment interference (such as weeds, floating soil, and equipment shaking) are removed by analyzing the spatial distribution difference of each point and its neighborhood points; then, the point cloud data after denoising is simplified based on the voxel grid downsampling method, the total amount of data is reduced to reduce the subsequent calculation pressure on the premise of retaining the core geometric morphological characteristics of the foundation pit; finally, combined with the inherent geometric properties of the foundation pit (such as the bottom surface being approximately horizontal and the side wall being approximately vertical) and the ground reference surface elevation data provided by the inertial measurement unit (IMU), a region growing algorithm is used for region segmentation, and finally, the region point cloud data of the three key feature regions is obtained independently and purely.

[0021] By performing the preprocessing procedure on the original three-dimensional point cloud data, on the one hand, the interference point clouds (such as weeds and floating soil) in the field environment can be effectively eliminated to avoid their interference with the subsequent feature recognition and solve the problem of measurement error caused by environmental debris shielding in traditional manual measurement; on the other hand, the downsampling operation can simplify the data amount while retaining the core geometric features, which can significantly improve the efficiency of subsequent key feature recognition and construction parameter calculation and avoid the calculation delay caused by large data amount; and the accurate region segmentation can clearly separate the side wall, the bottom surface, and the excavation boundary of the foundation pit, ensure that the point cloud data of each key feature region is independent and pure, and provide accurate data support for the subsequent extraction of geometric features such as the side wall generatrix, the bottom surface contour line, and the excavation boundary, thereby guaranteeing the accuracy and reliability of the final construction parameter calculation from the data preprocessing link.

[0022] Step 130, input the region point cloud data of each key feature region into the pre-trained key feature recognition model for feature recognition to obtain the key geometric features of the foundation pit.

[0023] The key feature recognition model refers to a model trained by sample data, which has the ability to automatically extract the core geometric information of the foundation pit from the key feature area point cloud data. It can realize the automatic conversion from the area point cloud to the key geometric features. The key geometric features of the foundation pit refer to the core feature set that can directly reflect the geometric shape of the foundation pit and support the calculation of subsequent construction parameters. They are the geometric information extracted from the area point cloud data and have engineering significance. In the context of transmission line excavation foundation pit, they mainly include but are not limited to side wall bus feature, bottom contour feature, and excavation boundary feature. The side wall bus feature refers to the core contour line (which can be a straight line or a regular curve) extracted from the side wall area point cloud data, which runs through the top to the bottom of the side wall. It can reflect the overall inclination trend and the integrity of the side wall. The bottom contour feature refers to the closed contour line around the edge of the bottom surface extracted from the bottom area point cloud data. It can reflect the planar shape and boundary range of the bottom surface. The excavation boundary feature refers to the continuous contour (which can be regular or irregular) extracted from the excavation boundary area point cloud data at the junction of the top of the foundation pit and the ground. It can reflect the actual spatial boundary of the excavation range.

[0024] For the embodiments of the present disclosure, the area point cloud data of the three key feature areas of the side wall, the bottom surface, and the excavation boundary obtained after preprocessing can be input into the input preprocessing module of the key feature recognition model which has been pre-trained. Through coordinate normalization and noise secondary filtering, the data standardization can be realized, and the regular point cloud data with uniform output format and pure quality can be obtained. Then, the multi-scale feature enhancement module of the model allocates the regular point cloud data to the corresponding feature extraction branch according to the feature extraction requirements of different areas. For example, the side wall area point cloud can be input into the branch with small-scale convolution kernel to capture the continuous contour from the top to the bottom of the side wall and generate the initial side wall bus feature. The bottom area point cloud can be input into the branch with medium-scale convolution kernel to focus on the closed shape of the bottom edge and generate the initial bottom contour feature. The excavation boundary area point cloud can be input into the branch with large-scale convolution kernel to strengthen the continuity of the contour at the top junction and generate the initial excavation boundary feature. Subsequently, the feature fusion layer fuses the three initial features and the spatial correlation information between the features (such as the vertical relationship between the side wall bus and the bottom contour, and the top connection relationship between the excavation boundary and the side wall bus) to obtain the fused features. Finally, the feature matching module calls the built-in feature template library covering different foundation pit types and defect types, calculates the similarity between the fused features and each template, selects the target template with the highest matching degree and the similarity that meets the standard, and uses the standard features in the target template and the spatial constraint relationship between the features as the reference to calibrate the initial side wall bus feature, the initial bottom contour feature, and the initial excavation boundary feature, respectively. Finally, the foundation pit key geometric features with standard precision and complete shape are output.

[0025] By using a pre-trained key feature recognition model, the key geometric features of the foundation pit can be automatically extracted. On the one hand, this replaces the traditional subjective judgment of the foundation pit's geometry by humans (such as manually comparing the verticality of the sidewalls with a ruler or visually estimating the bottom outline), avoiding subjective errors caused by differences in human experience. On the other hand, the model adapts to the feature extraction needs of different areas through a multi-scale feature enhancement module (capturing sidewall details at a small scale and maintaining boundary continuity at a large scale), and the template calibration function combined with the feature matching module can effectively correct minor deviations in the initial feature extraction (such as local discontinuities in the sidewall generatrix or slight offsets in the bottom outline), ensuring that the output key geometric features are both complete and accurate. At the same time, the entire recognition process requires no manual intervention, which can significantly shorten the feature extraction time and lay the foundation for rapid calculation of construction parameters.

[0026] Step 140: Based on the key geometric features of the foundation pit, calculate the construction parameters of the foundation pit through a geometric calculation model. The construction parameters of the foundation pit shall include at least the overall verticality deviation of the foundation pit, the excavation depth of the foundation pit, the flatness of the sidewalls, and the excavation size parameters.

[0027] The geometric calculation model refers to a dedicated calculation framework integrating various engineering geometric algorithms (such as line fitting, elevation calculation, principal component analysis, contour fitting, etc.). Its core function is to establish a mapping relationship between key geometric features of the foundation pit and construction quality indicators. It can automatically call appropriate algorithms for different geometric features, realizing the automated transformation from geometric information to quantitative construction parameters. The foundation pit construction parameters refer to the set of core indicators used to quantitatively evaluate the construction quality and compliance of transmission line excavation foundation pits. Among them, the overall verticality deviation of the foundation pit measures the verticality of the sidewalls, the excavation depth measures the burial depth of the pit, the sidewall flatness measures the flatness of the sidewall surface, and the excavation size parameters measure the size of the excavation range. Together, they constitute the key basis for foundation pit quality acceptance. The overall verticality deviation of the foundation pit refers to the maximum deviation between the actual shape of the foundation pit sidewalls and the ideal vertical state, which is determined by the sidewall generatrix characteristics. The angle between the fitted reference geometric element and the vertical reference is calculated and is a core indicator for evaluating the overturning resistance of the foundation pit. The excavation depth of the foundation pit refers to the vertical distance between the lowest point of the foundation pit bottom and the ground reference plane. It is calculated by comparing the lowest elevation of the bottom surface corresponding to the bottom contour line characteristics with the elevation of the ground reference plane collected by the inertial measurement unit (IMU), reflecting whether the foundation pit burial depth meets the design requirements. The sidewall flatness refers to the flatness of the foundation pit sidewall surface. It is obtained by statistically analyzing the distance deviation between the sidewall area point cloud and the sidewall main plane, quantifying the severity of defects such as protrusions and depressions on the sidewall surface, which affects the quality of subsequent foundation pouring. Excavation size parameters are quantitative indicators that characterize the actual size of the foundation pit excavation range. They are obtained by contour fitting of the excavation boundary features or calculation of the distance between feature points (such as diameter, side length, etc.), and determine whether the excavation range is suitable for the installation requirements of power facility foundations.

[0028] By using a geometric calculation model to automatically calculate construction parameters based on key geometric features, it can completely replace traditional manual measurement (such as using a plumb bob to measure verticality and a measuring tape to measure depth), avoid subjective errors caused by differences in personnel operating experience and field environment (strong light, terrain obstruction), significantly improve the accuracy and stability of parameter calculation, and provide reliable data support for subsequent foundation pit quality acceptance and problem rectification, effectively meeting the core needs of modern power engineering for efficient and precise construction quality control.

[0029] In summary, the method for extracting foundation pit construction parameters based on point cloud data provided by this invention can replace traditional manual data acquisition methods using tools such as measuring tapes, plumb bobs, and total stations through three-dimensional point cloud data acquisition. It eliminates the need for multiple people to collaboratively set up equipment, effectively reducing the interference of environmental factors such as rugged terrain, wind, rain, and strong sunlight on the inspection process. Furthermore, after preprocessing, the point cloud data of key feature areas such as sidewalls, bottom surfaces, and excavation boundaries are accurately segmented. Combined with a pre-trained key feature recognition model, key geometric features such as the foundation pit sidewall generatrix, bottom surface outline, and excavation boundaries are automatically extracted, avoiding subjective errors caused by differences in personnel operating experience. Finally, the overall verticality deviation, excavation depth, sidewall flatness, and excavation dimension parameters of the foundation pit are automatically calculated through a geometric calculation model, eliminating the need for manual data recording and calculation. This significantly shortens the inspection time for a single foundation pit and improves the stability of inspection accuracy, ultimately overcoming the limitations of traditional manual measurement and accurately meeting the stringent requirements of modern power engineering for construction efficiency and quality control.

[0030] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, and to fully illustrate the implementation of this embodiment, this embodiment also provides another method for extracting foundation pit construction parameters based on point cloud data, such as... Figure 2 As shown, the method includes: Step 210: Collect three-dimensional point cloud data of the excavation pit for the transmission line.

[0031] Step 220: Preprocess the 3D point cloud data to obtain the regional point cloud data of each key feature region.

[0032] For embodiments of this disclosure, step 220 may include the following steps: Step 220-1: Use a statistical filtering algorithm to denoise the 3D point cloud data and remove interfering point clouds.

[0033] Interference point clouds refer to redundant point clouds in 3D point cloud data that do not belong to the main structure of the excavation pit of the power transmission line. They mainly come from interference factors in the field collection environment (such as weeds on the edge of the pit, falling debris, environmental reflection interference during laser scanning, and abnormal points caused by slight equipment vibration). These point clouds will destroy the spatial continuity and integrity of the point cloud of the pit body. If they are not removed, they will affect the accuracy of subsequent region segmentation and geometric feature extraction.

[0034] In this embodiment of the disclosure, the core parameters of the statistical filtering algorithm can be set based on the density characteristics of the pit point cloud (e.g., the spacing between adjacent points in the pit point cloud collected in the field is usually 2-5mm). These parameters include the number of neighboring points to be analyzed for each point (e.g., selecting 30-50 neighboring points to ensure the representativeness of the statistical results) and the standard deviation threshold for judging outliers (e.g., setting it to 1.5-2 times the standard deviation to balance the noise removal rate and the retention rate of the main body point cloud). The algorithm then traverses each point in the point cloud one by one, searches for and determines a preset number of neighboring points for each point, calculates the spatial distance from the point to all neighboring points, and further calculates the average and standard deviation of the distance between the neighboring points. Then, the actual distance of each point is compared with the normal range of the average ± standard deviation threshold. If the distance of the point exceeds this range (i.e., it belongs to an outlier that deviates from the normal distribution), it is judged as an interfering point cloud and marked for removal. Finally, the output is point cloud data that retains only the main structure of the pit (side walls, bottom surface, excavation boundary), ensuring that subsequent downsampling, region segmentation and other operations can be carried out based on clean point cloud data.

[0035] Step 220-2: Downsample the denoised 3D point cloud data by voxel grid downsampling to obtain compressed point cloud data that retains the geometric features of the foundation pit.

[0036] Downsampling refers to the process of reducing the total number of point cloud data (such as millions or even tens of millions of coordinate points) that may still exist after denoising. The core purpose is to reduce the computational load of subsequent data processing (such as region segmentation and feature recognition) and improve the overall process efficiency. At the same time, it is necessary to ensure that the key geometric features of the foundation pit (such as sidewall busbars, bottom contour lines, and excavation boundaries) are not lost. Compressed point cloud data refers to a point cloud dataset that has been significantly reduced in total number of points after downsampling but retains the core geometric information completely. Its compression is reflected in the reduction of data volume (number of points) rather than the loss of geometric features. It can reduce storage and computing costs while meeting the requirements of subsequent region segmentation and key feature recognition for data accuracy and integrity.

[0037] In this embodiment of the present disclosure, when downsampling the denoised 3D point cloud data of the excavated foundation pit of the transmission line, the cube side length of the voxel grid can be set first according to the actual size of the foundation pit and the accuracy requirements of key geometric features. Then, the 3D space where the denoised point cloud data is located is uniformly divided into several non-overlapping voxel grids according to the set side length, so that each voxel contains a different number of point cloud data. Next, for all points in each voxel, the average value of its 3D coordinates (or geometric centroid coordinates) is calculated, and the point corresponding to the average value is retained as a representative point, while other redundant points in the voxel are removed. Finally, the representative points of all voxels are summarized to generate compressed point cloud data with a reduced total number of point clouds but complete retention of key geometric features such as the inclination trend of the foundation pit sidewall, the bottom edge contour, and the excavation boundary shape.

[0038] Step 220-3: Use the region growing algorithm to segment the compressed point cloud data into regions to obtain the region point cloud data of each key feature region.

[0039] For the embodiments of this disclosure, step 220-3 may include the following steps: screening candidate points on the bottom surface of the foundation pit from the compressed point cloud data and selecting seed points from them; starting from the seed points, performing region growth according to preset conditions to sequentially form preliminary regional point cloud data of the bottom surface region, side wall region and excavation boundary region; removing non-target point clouds from the preliminary regional point cloud data by using grayscale threshold and distance threshold to obtain regional point cloud data of each key feature region.

[0040] Among them, bottom candidate points refer to the set of point clouds selected from compressed point cloud data that are highly likely to belong to the bottom of the foundation pit. The selection criteria usually combine the inherent attributes of the foundation pit bottom (such as the normal vector direction of approximately horizontal points and the elevation difference with the ground reference plane conforming to the design excavation depth range), which is the basis for the subsequent selection of seed points. Seed points are reference points selected from bottom candidate points that can represent the core features of the foundation pit bottom. They need to have typical location (such as the central area of ​​the bottom, points without local protrusions / indentations). Their attributes (normal vector, elevation, etc.) are the reference standards for judging whether neighboring points belong to the bottom during region growth. Preset conditions refer to the core criteria for judging whether the point cloud belongs to the current region during region growth. For foundation pit scenarios, they usually include the similarity of the normal vector direction of the point cloud (such as the normal vector of bottom points being close to perpendicular to the ground, and the normal vector of side wall points being close to parallel to the ground). The system employs several criteria to ensure the consistency of the growing area. These criteria include: 1) a threshold for elevation difference between point clouds (e.g., the elevation difference between adjacent bottom points does not exceed a preset value); 2) a similarity threshold for grayscale values ​​(e.g., the grayscale difference of point clouds in the same area is within an allowable range); 3) a threshold for grayscale values ​​between point clouds; and 4) a similarity threshold for grayscale values ​​between point clouds. The system ensures the uniformity of the growing area by considering the differences in grayscale values ​​between the point cloud and the core area. The initial area point cloud data refers to the set of point clouds corresponding to the bottom surface, sidewalls, and excavation boundaries that are initially separated after the area grows. At this stage, there may still be a small number of point clouds with similar attributes to the target area but not belonging to the foundation pit itself (e.g., slightly raised soil points at the edge of the pit). Further purification is required. The grayscale threshold and distance threshold are screening criteria used for secondary purification of the initial area point cloud. The grayscale threshold eliminates interference points that do not belong to the foundation pit itself (e.g., the grayscale of weeds and gravel differs from that of the foundation pit wall / soil) based on the difference in grayscale values ​​reflected by the point cloud. The distance threshold eliminates discrete, isolated points based on the difference in distance between the point cloud and the core area, thus ensuring the purity of the area point cloud.

[0041] Step 230: Based on the preset geometric attributes and point cloud density distribution of each key feature region, detect whether there is missing point cloud data in each key feature region.

[0042] Among them, the preset geometric attributes refer to the predefined geometric morphology standards based on the inherent structural characteristics and engineering design requirements of each key feature area (sidewall, bottom, and excavation boundary) of the transmission line excavation pit. For example, the sidewall must have a continuous vertical (or preset tilt angle) outline, the bottom must be a closed planar shape, and the excavation boundary must be a continuous closed outline. These serve as the geometric reference benchmark for judging whether the point cloud is complete. Point cloud density distribution refers to the spatial density distribution of point cloud data within each key feature area. It is usually quantified by the number of point clouds per unit three-dimensional spatial volume or the average distance between adjacent point clouds. Under normal circumstances, the point cloud density of the same key area should remain relatively uniform. If the density of a certain area drops sharply or there are blanks, there may be missing point clouds. Missing point cloud data refers to the situation where, due to interference from the acquisition environment (such as occlusion, scanning blind spots) or equipment errors, the area that should have been covered did not collect point cloud data, resulting in blank points or a density far below the normal level. Such missing data will destroy the integrity of the geometric morphology of the area and affect the accuracy of subsequent key feature identification and construction parameter calculation.

[0043] In this embodiment of the present disclosure, when detecting whether there are missing point clouds in each key feature region, the predefined preset geometric attributes of each region can be called first, and the point cloud density distribution of each key feature region can be calculated by the algorithm. Then, the actual point cloud shape of each region is compared with the preset geometric attributes one by one. For example, it is checked whether there are discontinuities from the bottom surface to the excavation boundary in the side wall point cloud, whether there are unclosed gaps in the bottom surface point cloud, and whether there are obvious breaks in the excavation boundary point cloud. At the same time, it is combined with the density map to determine whether there are blank units with a density much lower than the average density of the region or no point cloud at all. If the actual point cloud shape of a certain region does not conform to the preset geometric attributes (such as discontinuities in the side wall) and the density at the corresponding position is abnormal, it is determined that there is missing point cloud data in the region. Otherwise, it is determined that the point cloud is complete, providing a basis for judgment on whether to perform point cloud completion in the future.

[0044] By combining preset geometric attributes with point cloud density distribution to detect missing points, the system achieves several advantages. First, the preset geometric attributes serve as a clear reference, avoiding misjudgments caused by relying solely on density (such as misjudging normal low-density point clouds at a bottom design recess as missing), thus ensuring the accuracy of missing point detection. Second, the point cloud density distribution serves as a quantitative basis, avoiding biases caused by subjective judgments based solely on geometric shapes, enabling precise location of missing points (such as localized missing points caused by weeds obscuring the sidewall). Furthermore, this detection method can preemptively identify point cloud issues before key feature recognition, preventing subsequent extraction of broken sidewall generatrices or incomplete bottom contours due to missing point clouds, which could lead to errors in construction parameter calculations. This provides a clear direction for subsequent point cloud completion, ensuring the reliability of the entire foundation pit construction parameter extraction process and reducing the risk of quality misjudgments due to point cloud data defects.

[0045] Step 240: If yes, then select the corresponding point cloud completion strategy based on the missing area of ​​the missing point cloud data, and complete the missing point cloud data according to the point cloud completion strategy.

[0046] The missing area of ​​a point cloud refers to the area of ​​missing point cloud data within each key feature region, occupying the area on the two-dimensional projection plane. This area is used to determine the appropriate completion strategy (lightweight algorithms for small areas, robust algorithms for large areas). A point cloud completion strategy is a pre-defined set of highly adaptable completion schemes for point cloud data with different degrees of missing data. The core logic is to select the algorithm as needed, choosing an algorithm that balances accuracy and efficiency based on the size of the missing area, avoiding the problem of a single algorithm being inefficient in small-area completion or inaccurate in large-area completion. Point cloud completion strategies include at least using neighborhood coordinate interpolation to complete the missing point cloud data corresponding to the first missing area, and using region generation... The long algorithm completes the missing point cloud data corresponding to the missing area of ​​the second point cloud. The missing area of ​​the first point cloud is less than or equal to the preset area threshold, and the missing area of ​​the second point cloud is greater than the preset area threshold. The neighborhood coordinate interpolation method is a completion algorithm suitable for small-area point cloud missing data. The core principle is to use the three-dimensional coordinates (neighborhood point coordinates) of the complete point cloud around the missing area to calculate the point cloud coordinates of the missing position through mathematical fitting (such as linear interpolation, polynomial interpolation). It is characterized by low computational load, high completion accuracy, and the ability to quickly restore the shape of local small missing data. The preset area threshold is a critical area value (such as 0.01㎡-0.05㎡) set in advance according to the accuracy requirements of the key feature area of ​​the foundation pit and the adaptability of the algorithm.

[0047] In this embodiment of the present disclosure, when missing point cloud data is detected in each key feature region, the missing area of ​​the point cloud on the two-dimensional projection plane can be calculated first using an algorithm; then, the calculated missing area is compared with a preset area threshold. If the missing area is ≤ the preset area threshold (i.e., the first missing point cloud area), then the neighborhood coordinate interpolation method is used for completion: a complete point cloud within a range of 5-10 times the surrounding area of ​​the missing region is selected as a neighborhood point, and the coordinate distribution law of the neighborhood point is fitted by linear interpolation or polynomial interpolation to calculate the three-dimensional coordinates of the missing position and generate a completed point cloud to fill the missing gap; if the missing area is > the preset area threshold (i.e., the second missing area), then the missing area is filled by a neighborhood coordinate interpolation method. If the point cloud is missing, a region growing algorithm is used to complete it: points with attributes (normal vector, gray level, elevation) that are consistent with the expected attributes of the missing region are selected from the complete point cloud at the edge of the missing region as seed points. Under the conditions that the normal vector similarity is greater than or equal to the preset value, the elevation difference is less than or equal to the preset value, and the gray level difference is within the allowable range, the point cloud is gradually grown from the seed points to the missing region, and the complete point cloud with the same attributes as the surrounding point cloud is generated, ensuring that the geometric shape of the complete region is consistent with the original region. Finally, after completion by both strategies, the connection between the complete point cloud and the original point cloud is smoothed to eliminate splicing traces and obtain complete and consistent point cloud data for each key feature region.

[0048] The point cloud completion scheme, which selects strategies based on missing area, achieves two key advantages. First, it ensures precise matching between the completion algorithm and the degree of missing data. For small missing areas, neighborhood coordinate interpolation is used to quickly restore local details with minimal computation, avoiding redundant calculations in region growing algorithms for small-area completion. For large missing areas, region growing algorithms are used to ensure geometric continuity between the completed and original regions through attribute similarity growth, avoiding morphological distortion caused by insufficient data support in large-area completion using neighborhood interpolation, thus balancing completion efficiency and accuracy. Second, the completed point cloud data eliminates gaps and breaks in key feature regions, ensuring no data defects in subsequent key feature identification and avoiding feature extraction errors caused by point cloud missing data, thereby guaranteeing the accuracy of construction parameter calculations. Furthermore, the completion process requires no manual intervention, adapting to the automated processing needs of field engineering scenarios, further improving the reliability and practicality of the entire foundation pit construction parameter extraction process, and providing a complete data foundation for high-quality engineering quality control.

[0049] Step 250: Input the regional point cloud data of each key feature region into the pre-trained key feature recognition model for feature recognition to obtain the key geometric features of the foundation pit. The key geometric features of the foundation pit include at least the side wall generatrix features, the bottom contour line features, and the excavation boundary features.

[0050] The key feature recognition model includes an input preprocessing module, a multi-scale feature enhancement module, and a feature matching module. The multi-scale feature enhancement module sets up a first feature extraction branch, a second feature extraction branch, a third feature extraction branch, and a feature fusion layer at different scales. The feature matching module has a built-in feature template library covering different types of foundation pits and defect types. Each feature template in the feature template library contains standard features and spatial constraints between standard features. The standard features include standard sidewall busbar features, standard bottom surface outline features, and standard excavation boundary features.

[0051] In specific application scenarios, when pre-training a key feature recognition model, the implementation steps may include: determining sample point cloud data and the key geometric features of the sample foundation pit corresponding to the sample point cloud data; using the sample point cloud data as input features and the key geometric features of the sample foundation pit as training labels, iteratively training the key feature recognition model until the loss function of the key feature recognition model is less than a preset threshold, and determining that the key feature recognition model training is complete; wherein, during the iterative training of the key feature recognition model, initial features of the samples are extracted through a multi-scale feature enhancement module, and after feature fusion and correlation analysis, sample fusion features and sample feature space correlation information are obtained; the sample fusion features and sample feature space correlation information are matched with a feature template library through a feature matching module, and the key geometric features of the foundation pit predicted by the model are output based on the matching results; the loss function is calculated based on the predicted key geometric features of the foundation pit and the key geometric features of the sample foundation pit, and the model parameters are adjusted in reverse when the loss function is greater than or equal to a preset threshold, and the prediction of the key geometric features of the foundation pit is repeated.

[0052] In this embodiment of the disclosure, the trained key feature recognition model can be reused for feature recognition to obtain the key geometric features of the foundation pit. Accordingly, step 250 of the embodiment may include the following steps: Step 250-1: Standardize the regional point cloud data of each key feature region through the input preprocessing module, and output regular point cloud data.

[0053] The input preprocessing module is a functional unit in the key feature recognition model specifically designed to optimize the input regional point cloud data. Its core function is to eliminate data format differences, residual noise, and redundant information, transforming the original regional point cloud into standard input data that meets the requirements of the model's subsequent feature extraction branches. It serves as a bridge connecting the regional point cloud data and the model's core computation module. Standardization processing refers to a series of unification and purification operations performed on the regional point cloud data of each key feature region. The core operations include coordinate normalization (unifying the spatial coordinate reference), secondary noise filtering (removing minor interference points left over from region segmentation), and data format regularization (unifying the storage and transmission format of point cloud data). The aim is to eliminate the data heterogeneity of different regional point clouds and ensure that the model can process them stably. Regularized point cloud data refers to point cloud data that has been standardized by the input preprocessing module, possessing characteristics such as unified coordinate reference, no residual noise, standardized format, and complete geometric features. It can directly adapt to the input requirements of the model's multi-scale feature enhancement module and is the basic data form for the model to accurately extract initial features.

[0054] In this embodiment of the disclosure, the preprocessed point cloud data of the three key feature regions—sidewall, bottom, and excavation boundary—can be simultaneously input into the input preprocessing module of the key feature recognition model. The module first performs coordinate normalization on the point clouds of each region to eliminate the interference of absolute coordinate differences under different acquisition scenarios on model recognition. Next, it performs secondary noise filtering by analyzing local point cloud density (removing isolated points with densities far below the average density of the region) and verifying normal vector consistency (removing abnormal points where the normal vector deviates significantly from the normal vector of the main body of the region). This further removes minor interference points that may remain from the region segmentation process (such as loose soil points attached to the pit wall or false points generated by scanning reflections). Finally, it performs data format unification, converting the coordinate data, normal vector data, and other information of the point clouds in different regions into a tensor format preset by the model (such as fixed-dimensional matrix data), ensuring that the point cloud data of the sidewall, bottom, and excavation boundary can be efficiently read by the corresponding branches of the subsequent multi-scale feature enhancement module. After all operations are completed, the module outputs well-organized point cloud data with unified coordinates, no residual noise, and a standardized format, preparing for subsequent feature extraction.

[0055] By standardizing the input preprocessing module, the heterogeneity of point cloud data in different key feature regions can be eliminated, avoiding feature extraction deviations caused by unstable data input. On the other hand, the normalized point cloud data can directly adapt to the input requirements of the model's multi-scale feature enhancement module, without the need for subsequent modules to handle data format or noise issues. This can significantly improve the overall processing efficiency of the model and provide a prerequisite for the model to accurately extract initial features and finally output high-quality key geometric features of the foundation pit.

[0056] Step 250-2: Input the side wall region point cloud data in the regular point cloud data into the first feature extraction branch. Use the small-scale convolution kernel configured in the first feature extraction branch to identify the symmetrical contour line covering the top to bottom of the side wall as the initial side wall generatrix feature.

[0057] The first feature extraction branch refers to a functional branch under the multi-scale feature enhancement module in the key feature recognition model, specifically adapted to the point cloud data of the sidewall region, used to extract the geometric features (sidewall generatrices) specific to the sidewall. The small-scale convolution kernel refers to the operation unit configured in the first feature extraction branch with a small three-dimensional convolution window size (such as 3×3×3). Its core feature is a small receptive field (only covering a local point cloud region), which can accurately capture subtle contour changes and continuous edge information in the point cloud, avoiding the loss of local details caused by large-scale convolution kernels, and adapting to the extraction requirements of the sidewall generatrices to retain continuous morphological details. The initial sidewall generatrice feature refers to the preliminary contour line feature that runs through the top to the bottom of the sidewall, identified and output by the first feature extraction branch from the point cloud of the sidewall region.

[0058] In this embodiment of the present disclosure, the point cloud data corresponding to the side wall region can be extracted separately from the regular point cloud data output by the input preprocessing module and input into the first feature extraction branch. This branch first performs three-dimensional spatial convolution initialization on the point cloud of the side wall region. The point cloud is scanned point by point by point through the built-in small-scale convolution kernel. The small-scale convolution kernel has the advantage of small receptive field, which accurately captures the spatial position relationship and contour continuity of adjacent point clouds of the side wall, while filtering out local interference caused by small protrusions and depressions on the side wall surface. During the convolution operation, the branch simultaneously performs coherence analysis and splicing on the scanned local contour segments, and prioritizes retaining the contour line that extends from the top to the bottom of the side wall and conforms to the inherent shape of the side wall (such as the left and right symmetry of the vertical side wall and the linear symmetry of the inclined side wall). Finally, it integrates and outputs a contour line that runs through the entire height of the side wall and can reflect the overall tilt trend and basic shape of the side wall, which is the initial side wall generatrix feature.

[0059] The first feature extraction branch is specifically used to process the point cloud of the sidewall region and is configured with small-scale convolution kernels. On the one hand, the small receptive field of the small-scale convolution kernels can accurately capture the local contour details from the top to the bottom of the sidewall, avoiding the blurring of contour edges or the breakage of continuous shape caused by the excessive coverage of the large-scale convolution kernels. This ensures that the initial sidewall generatrix features can completely preserve the continuous trend and subtle shape changes of the sidewall. On the other hand, the specialized processing of this branch decouples the sidewall feature extraction from the feature extraction of the bottom surface and the excavation boundary. This can not only improve the efficiency of single-region feature extraction, but also ensure the adaptability of the initial sidewall generatrix features to the sidewall region. This provides a high-quality preliminary feature foundation for subsequent feature fusion and template calibration, directly ensuring the accuracy of the final key geometric features of the foundation pit (final sidewall generatrix), and thus supporting the accurate calculation of subsequent construction parameters such as verticality deviation and sidewall flatness.

[0060] Step 250-3: Input the bottom region point cloud data from the regular point cloud data into the second feature extraction branch. Use the mesoscale convolution kernel configured in the second feature extraction branch to identify the closed contour line of the bottom edge as the initial bottom contour line feature.

[0061] The second feature extraction branch refers to a functional branch under the multi-scale feature enhancement module in the key feature recognition model, specifically adapted to the point cloud data of the bottom region, used to extract the unique geometric features (bottom contour lines) of the bottom surface. The mesoscale convolution kernel refers to the operational unit configured in the second feature extraction branch with a three-dimensional convolution window size between small and large scales (such as 5×5×5). Its core feature is a moderate receptive field, which can cover a certain range of point cloud at the bottom edge to capture the overall contour trend, while also preserving local edge details. This avoids contour breaks caused by the limited field of view of small-scale kernels or edge offsets caused by blurred details of large-scale kernels, thus meeting the extraction requirements of the closed contour lines of the bottom surface. The initial bottom contour line feature refers to the preliminary closed contour line feature surrounding the bottom edge that is identified and output by the second feature extraction branch from the point cloud of the bottom region.

[0062] In this embodiment of the present disclosure, the point cloud data corresponding to the bottom region can be extracted separately from the regular point cloud data output by the input preprocessing module and then input into the second feature extraction branch. This branch first performs a three-dimensional convolutional scan on the point cloud of the bottom region, and analyzes the spatial distribution pattern of the bottom point cloud through the built-in mesoscale convolution kernel. The mesoscale convolution kernel, with its moderate receptive field, can both cover the adjacent point cloud of the bottom edge to capture the edge continuity and accurately distinguish the bottom body point cloud from the transition point cloud of the side wall bottom. During the convolution operation, the branch simultaneously performs closure verification and splicing on the identified bottom edge segments, prioritizing the retention of edge lines that surround the bottom without obvious breaks, and supplementing the segments with slight breaks by fitting the trends of adjacent edges. Finally, it integrates and outputs a closed contour line that surrounds the bottom edge and reflects the overall boundary range of the bottom, i.e., the initial bottom contour line feature.

[0063] The second feature extraction branch is specifically used to process the point cloud of the bottom area and is configured with a mesoscale convolution kernel. On the one hand, the moderate receptive field of the mesoscale convolution kernel can balance the capture of the overall trend of the bottom edge and the preservation of local details, avoiding edge fragmentation caused by the limited field of view of the small-scale convolution kernel (which requires frequent splicing and is prone to errors), or edge offset caused by the blurring of details of the large-scale convolution kernel (which cannot accurately locate the bottom boundary). This ensures that the initial bottom contour features are both complete and closed and have accurate edges. On the other hand, the dedicated processing of this branch decouples the bottom feature extraction from other areas, which can improve the feature extraction efficiency. Moreover, the closure and accuracy of the initial bottom contour features can provide a high-quality foundation for subsequent feature fusion (related to the vertical relationship with the side wall generatrix) and template calibration (correcting local offsets), which directly supports the accuracy of the subsequent foundation pit excavation depth calculation (the excavation depth depends on the bottom elevation associated with the bottom contour line), ensuring the reliability of the construction parameter calculation.

[0064] Step 250-4: Input the point cloud data of the excavation boundary region in the regular point cloud data into the third feature extraction branch. Use the large-scale convolution kernel configured in the third feature extraction branch to identify the set of continuous contour points at the top intersection as the initial excavation boundary features.

[0065] The third feature extraction branch refers to the functional branch in the multi-scale feature enhancement module of the key feature recognition model that is specifically adapted to the point cloud data of the excavation boundary area. Its core responsibility is to extract the specific geometric features of the excavation boundary. The large-scale convolution kernel refers to the large-size (e.g., 7×7×7) three-dimensional convolution window configured in the third feature extraction branch. Its core feature is a large receptive field (which can cover a large area of ​​the point cloud), which can effectively capture the continuity of the contours in the point cloud over a large area and avoid contour breaks caused by the limited field of view of small-scale kernels. It is adapted to the large-scale contour extraction needs of the excavation boundary at the junction of the ground and the sidewall. The initial excavation boundary features refer to the set of preliminary continuous contour points located at the junction of the top of the foundation pit and the ground, which are identified and output by the third feature extraction branch from the point cloud of the excavation boundary area.

[0066] In this embodiment of the present disclosure, the point cloud data of the excavation boundary region in the regular point cloud data output by the input preprocessing module can be extracted separately and input into the third feature extraction branch. This branch first performs three-dimensional spatial convolution initialization on the point cloud of the excavation boundary region, and scans the point cloud region by region through the built-in large-scale convolution kernel. With the advantage of a large receptive field, the large-scale convolution kernel simultaneously covers the excavation boundary points at the junction of the top of the foundation pit, the surrounding ground points, and the top points of the sidewall, accurately distinguishing the spatial distribution differences of the three (such as the excavation boundary points having both ground elevation and sidewall edge coordinate characteristics). During the convolution operation, the branch simultaneously performs continuous filtering and aggregation on the scanned point cloud that conforms to the excavation boundary attributes, removes interference points such as ground debris and redundant points on the top of the sidewall, and splices the remaining points in spatial order to form a continuous contour point set covering the top of the foundation pit. Finally, the output is the initial excavation boundary feature that can reflect the basic range and continuous shape of the excavation boundary.

[0067] The third feature extraction branch is used to specifically process the point cloud of the excavation boundary area and is configured with a large-scale convolution kernel. On the one hand, the large receptive field of the large-scale convolution kernel can effectively capture the continuity of the large-scale contour at the junction of the top of the foundation pit, avoiding the excavation boundary contour breakage or local loss caused by the limited field of view of the small-scale kernel, and ensuring the integrity of the initial excavation boundary features. On the other hand, the decoupling processing of the dedicated branch can improve the extraction efficiency of excavation boundary features, and the continuity and accuracy of the initial features provide a high-quality foundation for subsequent feature fusion (related to the connection relationship with the top of the side wall busbar) and template calibration (correcting local offsets), directly supporting the accuracy of subsequent excavation size parameter calculation and ensuring the reliability of construction parameter extraction.

[0068] Step 250-5: Use the feature fusion layer to perform feature fusion and correlation analysis on the initial sidewall busbar features, initial bottom surface contour features and initial excavation boundary features to obtain fused features and feature space correlation information.

[0069] Among them, the feature fusion layer refers to the core functional unit in the multi-scale feature enhancement module of the key feature recognition model, which is responsible for integrating initial features from multiple regions and analyzing the logical relationships between features. Its core function is to break the information isolation of each feature extraction branch. Through unified integration and correlation analysis, it generates comprehensive feature data that combines geometric information and spatial logic of the entire region, providing complete information support for subsequent feature matching. The fused feature refers to the comprehensive feature generated by the feature fusion layer after aggregating information from the three initial features. It retains the detailed information of each initial feature and eliminates the limitations of individual initial features through information complementarity, forming a unified feature that can characterize the overall geometric shape of the foundation pit. The feature spatial correlation information refers to the geometric logical relationship between each initial feature calculated simultaneously by the feature fusion layer when integrating the initial features. For example, the vertical angle between the side wall generatrix and the bottom contour line, the top connection position between the excavation boundary and the side wall generatrix, and the projection overlap degree between the bottom contour line and the excavation boundary are core data that reflect the spatial correlation of each key area of ​​the foundation pit.

[0070] In this embodiment of the present disclosure, the initial sidewall generatrix features, initial bottom contour features, and initial excavation boundary features output by the first, second, and third feature extraction branches can be simultaneously input into the feature fusion layer. The feature fusion layer can perform feature information aggregation and extract common information (such as the overall geometric center of the foundation pit and the size ratio of each area) and complementary information (such as the height data of the sidewall generatrix to supplement the vertical dimension information of the bottom contour line and the range data of the excavation boundary to supplement the horizontal dimension information of the sidewall generatrix) of each initial feature through a weighted fusion algorithm, forming a fusion feature covering the entire area of ​​the foundation pit. At the same time, the fusion layer analyzes and records the spatial relationship logic between each initial feature through a geometric relationship calculation module, such as calculating the angle between the plane where the sidewall generatrix and the bottom contour line are located (to determine whether they are perpendicular), locating the connection position between the top endpoint of the excavation boundary and the sidewall generatrix (to determine whether they are continuous), and verifying the overlap between the bottom contour line and the excavation boundary on the horizontal projection plane (to determine whether they match), generating feature spatial relationship information. Finally, the feature fusion layer synchronously outputs the fusion feature that integrates the details and spatial logic of the entire area, as well as the feature spatial relationship information that records the geometric relationship between the features.

[0071] By using feature fusion and correlation analysis in the feature fusion layer, the information isolation of each feature extraction branch can be broken, avoiding the judgment bias caused by subsequent feature matching relying solely on a single initial feature. At the same time, the feature space correlation information can provide implicit logical support for subsequent construction parameter calculation, avoiding the distortion of results caused by missing logical relationships between regions during parameter calculation. It can improve the completeness and accuracy of key geometric features of the foundation pit from both information integration and logical correlation dimensions, laying the foundation for high-quality construction parameter extraction.

[0072] Step 250-6: In the feature matching module, based on the fused features and feature space association information, calculate the similarity between the feature template and each feature template in the feature template library, filter the target template with a similarity greater than or equal to the preset similarity threshold, and calibrate the initial sidewall busbar features, initial bottom contour features and initial excavation boundary features respectively based on the standard features and spatial constraint relationships of the target template to obtain the final excavation boundary features.

[0073] The feature matching module is the core functional unit in the key feature recognition model responsible for comparing the comprehensive features extracted by the model with standard templates and calibrating the initial features based on the templates. Its core function is to correct the errors of the initial features by referring to the standard templates, ensuring that the final output features conform to engineering geometric standards. The feature template library is a set of standard feature templates pre-stored in the feature matching module, covering different types of foundation pits (circular, rectangular, irregular) and design specifications. Each template contains standard features (such as standard sidewall generatrices, standard bottom contour lines, and standard excavation boundaries) and spatial constraints (such as the vertical relationship between the sidewall generatrices and the bottom contour lines in the standard template, and the top connection relationship between the excavation boundary and the sidewall generatrices), providing a reference benchmark for feature comparison and calibration. Similarity refers to the degree of matching between the fused features, feature spatial correlation information and a certain feature template, quantified by a specific algorithm (such as feature morphology matching degree and spatial relationship consistency calculation). The higher the similarity value, the better the fit between the features extracted by the model and the standard template. The higher the similarity, the more accurate the selection of the template. The preset similarity threshold is a critical value (e.g., 0.8-0.9) set in advance to determine whether a feature is compatible with a template. Only when the similarity is greater than or equal to this threshold is the corresponding feature template considered a compatible template, thus avoiding calibration deviations caused by using templates that differ too much from the current foundation pit shape. The target template is a standard template selected from the feature template library whose similarity to the fused features and feature space association information is greater than or equal to the preset threshold. The standard features and spatial constraints contained in the target template best match the actual design shape of the current foundation pit. The calibration process refers to the operation of correcting errors in the initial sidewall busbar features, initial bottom contour features, and initial excavation boundary features using the standard features of the target template as the morphological benchmark and the spatial constraints as the logical benchmark. This includes correcting local offsets of the initial excavation boundary, filling in minor breaks in the initial sidewall busbar, and adjusting irregular edges of the initial bottom contour. The core purpose is to eliminate the extraction error of the initial features and make them conform to engineering standards.

[0074] Calibration is performed based on the standard features and spatial constraints of the target template. This not only corrects the extraction error of the initial features but also ensures the spatial logic consistency between the features, avoiding deviations in subsequent construction parameter calculations due to initial feature errors or spatial logic confusion. At the same time, the entire matching and calibration process is fully automated and requires no manual intervention. This avoids subjective errors in manually judging template suitability and correcting features, while also improving feature calibration efficiency. The final output of accurate excavation boundary features provides reliable data support for the calculation of subsequent foundation pit excavation size parameters, further ensuring the accuracy and automation level of the entire foundation pit construction parameter extraction process, and meeting the high-quality and high-efficiency quality control requirements of modern power engineering.

[0075] Step 260: Based on the key geometric features of the foundation pit, calculate the construction parameters of the foundation pit through a geometric calculation model. The construction parameters of the foundation pit shall include at least the overall verticality deviation of the foundation pit, the excavation depth of the foundation pit, the flatness of the sidewalls, and the excavation size parameters.

[0076] For embodiments of this disclosure, step 260 may include the following steps: Step 260-1: Based on the geometric properties of the sidewall busbar characteristics, fit the reference geometric elements corresponding to the excavation pit of the transmission line; calculate the angle between the reference geometric elements and the vertical reference in each region, and take the maximum value of the angle in each region as the overall verticality deviation of the pit.

[0077] Among them, the geometric attributes of the sidewall generatrix feature refer to the inherent geometric morphological characteristics presented by the sidewall generatrix feature, such as morphological type (straight line type, regular curved surface type), spatial extension trend (vertical extension, inclined extension at a preset angle), and overall continuity (whether it presents a continuous and unbroken morphology). The reference geometric element refers to the geometric carrier that can characterize the overall morphology and inclination trend of the sidewall, obtained by adaptation algorithms (such as straight line fitting, curved surface axis extraction) based on the geometric attributes of the sidewall generatrix feature. For example, a straight generatrix corresponds to a straight axis, and a curved sidewall corresponds to the curved surface center axis or local plane normal vector. It is the core reference object for subsequent calculation of verticality deviation. The vertical reference refers to the reference based on the inertial measurement unit. The ideal vertical direction obtained by gravity direction calibration provided by the IMU (Integrated Measurement Unit) is the standard vertical state direction that the foundation pit sidewall should achieve. It serves as a reference benchmark for measuring whether the reference geometric elements (sidewall morphology) are tilted. The included angle of each region refers to the angle between the reference geometric element of each region and the vertical benchmark after the foundation pit sidewall is divided according to preset rules (such as dividing it into several regions along the circumference). It can reflect the local tilt degree of different sidewall regions. The overall verticality deviation of the foundation pit refers to the maximum value selected from the included angles of each region. This value can reflect the most severe tilt degree of the foundation pit sidewall. It conforms to the principle of assessing the overall stability in the engineering field with the most unfavorable situation and is the core indicator for quantifying whether the verticality of the foundation pit meets the standard.

[0078] Step 260-2: Based on the point cloud data corresponding to the bottom contour line features, determine the lowest bottom elevation value by fitting the lowest bottom elevation surface; call the ground reference elevation data collected by the inertial measurement unit (IMU) to calculate the difference between the ground reference elevation and the lowest bottom elevation value, which is used as the excavation depth of the foundation pit.

[0079] Among them, the fitted lowest elevation surface of the bottom surface refers to the selection of a subset of point clouds at the lowest point of the bottom surface from the point cloud data associated with the bottom surface contour line through a specific geometric algorithm (such as RANSAC random sampling consensus algorithm, least squares method), and then fitting a plane that can represent the overall elevation trend of the lowest area of ​​the bottom surface. The core purpose is to accurately locate the lowest height benchmark of the bottom surface and avoid interference from local bulges / depressions of a single lowest point. The lowest elevation value of the bottom surface refers to the elevation coordinate value corresponding to the fitted lowest elevation surface of the bottom surface, which can reflect the actual lowest height of the bottom surface of the foundation pit. The ground reference elevation data refers to the standard elevation data of the ground where the foundation pit is located after being collected and calibrated by an inertial measurement unit (IMU), representing the original ground height before the foundation pit was excavated.

[0080] In this embodiment of the present disclosure, the point cloud data corresponding to the bottom contour features output by the key feature recognition model can be extracted first, and a subset of point clouds covering the bottom edge and bottom body with complete elevation information can be selected. Using a fitting algorithm with strong anti-interference ability such as RANSAC, abnormal high points caused by local protrusions and loose soil residue on the bottom surface are removed from the subset of point clouds, and the point cloud that can reflect the lowest area of ​​the bottom surface is retained. The retained lowest area point cloud is fitted with a plane using the least squares method to obtain the lowest elevation surface of the bottom surface that can characterize the trend of the lowest height of the bottom surface. The elevation coordinates of this plane are read as the lowest elevation value of the bottom surface. Then, the ground reference elevation data synchronously recorded and calibrated by the inertial measurement unit (IMU) during the foundation pit acquisition stage is called, and the difference between the ground reference elevation and the lowest elevation value of the bottom surface is the construction parameter that can truly reflect the excavation depth of the foundation pit.

[0081] Step 260-3: Use principal component analysis to extract the principal plane of the sidewall region from the point cloud data associated with the sidewall busbar features. Determine the normal vector of the principal plane by calculating the point cloud covariance matrix and construct the principal plane equation. Based on the principal plane equation, traverse all point clouds in the sidewall region, calculate the distance deviation of each point cloud to the principal plane, and count the proportion of point clouds with distance deviation greater than the preset distance deviation threshold to the total number of point clouds in the sidewall as the sidewall flatness.

[0082] Principal component analysis (PCA) is a statistical analysis algorithm used to extract the main distribution direction of data from point cloud data in the sidewall region. Its core function is to find the main direction that can characterize the overall flatness trend of the sidewall through dimensionality reduction. The sidewall principal plane is a reference plane that reflects the overall flatness trend of the sidewall, obtained by fitting the PCA results. The point cloud covariance matrix is ​​a matrix constructed by calculating the variance and covariance of the point cloud data in the sidewall region in three-dimensional space, which quantifies the dispersion and directional characteristics of the point cloud data. The principal plane normal vector is a vector perpendicular to the sidewall principal plane, obtained by solving the eigenvectors of the point cloud covariance matrix, and its direction is unique. The spatial orientation of the principal plane is determined; the principal plane equation refers to the three-dimensional plane mathematical expression established based on the coordinates of points on the principal plane and the normal vector of the principal plane (e.g., Ax+By+Cz+D=0), which can be directly used to calculate the vertical distance from any point cloud to the principal plane; the distance deviation refers to the vertical distance from a single point cloud in the sidewall region to the sidewall principal plane. The larger the value, the further the point deviates from the overall flatness trend of the sidewall; the preset distance deviation threshold refers to the critical distance (e.g., 2mm-5mm) set in advance according to engineering quality standards to judge whether the point cloud is uneven. Point clouds with a distance deviation greater than this threshold are judged as uneven points, and vice versa.

[0083] In this embodiment of the present disclosure, point cloud data covering the entire sidewall region, associated with the sidewall busbar features, can be extracted first. Principal component analysis (PCA) is then used to reduce the dimensionality of this point cloud data, focusing on the principal direction that reflects the overall distribution trend of the sidewall. Based on the PCA results, the three-dimensional covariance matrix of the point cloud is calculated, and the eigenvectors of this matrix are solved. The eigenvector that characterizes the attitude of the sidewall principal plane is selected as the principal plane normal vector. Combining the coordinates of any point on the sidewall principal plane (such as the center point of the sidewall) with the principal plane normal vector, a principal plane equation for calculating distance is constructed. Subsequently, all point clouds in the sidewall region are traversed, and the three-dimensional coordinates of each point cloud are substituted into the principal plane equation to calculate the vertical distance from each point cloud to the principal plane, i.e., the distance deviation. Finally, the number of uneven points with a distance deviation greater than a preset distance deviation threshold is counted, and this number is divided by the total number of point clouds in the sidewall region. The resulting ratio is the sidewall flatness, which quantifies the sidewall flatness quality.

[0084] Step 260-4: Extract the key geometric information corresponding to the excavation boundary features. The key geometric information includes closed contour point cloud data or feature point coordinates. Perform geometric fitting on the closed contour point cloud data using the least squares method, or calculate the straight-line distance between adjacent feature points based on the feature point coordinates to obtain the actual feature size representing the excavation range of the foundation pit, which is used as the excavation size parameter.

[0085] Among them, the key geometric information corresponding to the excavation boundary features refers to the core data extracted from the final excavation boundary features that can directly characterize the shape and size of the excavation area; geometric fitting refers to matching and calculating the closed contour point cloud data with the preset geometric model (such as circle or rectangle) based on the least squares method to obtain the ideal geometric shape that can completely characterize the overall shape of the excavation boundary (such as the fitted circular boundary or rectangular boundary), with the aim of transforming the discrete point cloud into a regular geometric shape whose size can be directly calculated; feature point coordinates refer to the three-dimensional coordinates of points marked manually or by algorithms on the excavation boundary that can reflect the key shape of the boundary (such as the four corner points of a rectangular foundation pit or the turning point of an irregular foundation pit). The positional relationship of these points can directly reflect the local size and overall range of the excavation boundary, and the size can be calculated without relying on the complete contour fitting; actual feature size refers to the size of the ideal geometric shape obtained through geometric fitting (such as the diameter of the fitted circle or the length and width of the fitted rectangle), or the distance between adjacent feature points calculated based on the feature point coordinates (such as the side length between the corner points of a rectangle), which can truly reflect the actual size of the excavation area of ​​the foundation pit.

[0086] In this embodiment of the disclosure, key geometric information that can characterize the excavation range can first be extracted from the final excavation boundary features output by the key feature recognition model. If the excavation boundary shape is regular (such as a circle or rectangle) and the point cloud is complete, then the closed contour point cloud data covering the boundary is obtained. If the boundary shape is irregular or the point cloud is locally dense (such as the point cloud being concentrated at the corner), then the coordinates of feature points on the boundary (such as turning points or vertices) are extracted. Then, the calculation method is selected according to the type of key geometric information: if it is closed contour point cloud data, the least squares method is used to fit the point cloud with a preset regular geometric model (such as a circle or rectangle). For example, the diameter of a circle is fitted, and the length and width of a rectangle are fitted, to obtain the actual feature size characterizing the excavation range. If it is feature point coordinates, the straight-line distance between adjacent feature points is calculated using the spatial distance formula (such as the distance between adjacent corner points of a rectangular foundation pit is the side length), and the actual feature size is obtained in the same way. Finally, the obtained actual feature size is determined as the excavation size parameter for quantifying the excavation range of the foundation pit.

[0087] In summary, the technical solution of this application, through 3D point cloud data acquisition, can replace traditional manual data acquisition methods using tools such as measuring tapes, plumb bobs, and total stations. It eliminates the need for multiple people to collaboratively set up equipment, effectively reducing the interference of environmental factors such as rugged terrain, wind, rain, and strong sunlight on the inspection process. Furthermore, after preprocessing, point cloud data of key feature areas such as sidewalls, bottom surfaces, and excavation boundaries are accurately segmented. Combined with a pre-trained key feature recognition model, key geometric features such as the sidewall generatrix, bottom contour line, and excavation boundary are automatically extracted, avoiding subjective errors caused by differences in personnel operating experience. Finally, a geometric calculation model automatically calculates the overall verticality deviation, excavation depth, sidewall flatness, and excavation dimension parameters of the foundation pit, eliminating the need for manual data recording and calculation. This significantly shortens the inspection time for a single foundation pit while improving the stability of inspection accuracy, ultimately overcoming the limitations of traditional manual measurement and accurately meeting the stringent requirements of modern power engineering for construction efficiency and quality control.

[0088] Furthermore, as Figure 1 and Figure 2 The specific implementation of the method shown in this embodiment provides a device for extracting foundation pit construction parameters based on point cloud data, such as... Figure 3 As shown, the device may include: acquisition module 31, processing module 32, input module 33, and calculation module 34.

[0089] Acquisition module 31 can be used to acquire three-dimensional point cloud data of the excavation pit of the power transmission line; Processing module 32 can be used to preprocess the three-dimensional point cloud data to obtain the regional point cloud data of each key feature region. The key feature regions include at least the side wall region, the bottom surface region and the excavation boundary region. Input module 33 can be used to input the regional point cloud data of each key feature region into the pre-trained key feature recognition model for feature recognition, and obtain the key geometric features of the foundation pit. The key geometric features of the foundation pit include at least the side wall generatrix features, the bottom surface outline features and the excavation boundary features. The calculation module 34 can be used to calculate the construction parameters of the foundation pit based on the key geometric features of the foundation pit through a geometric calculation model. The construction parameters of the foundation pit include at least the overall verticality deviation of the foundation pit, the excavation depth of the foundation pit, the flatness of the sidewalls, and the excavation size parameters.

[0090] In some embodiments of this application, the processing module 32 can be specifically used to perform denoising processing on the three-dimensional point cloud data using a statistical filtering algorithm to remove interfering point clouds; to perform downsampling processing on the denoised three-dimensional point cloud data through voxel grid downsampling to obtain compressed point cloud data after retaining the geometric features of the foundation pit; and to perform region segmentation on the compressed point cloud data using a region growing algorithm to obtain regional point cloud data of each key feature region.

[0091] In some embodiments of this application, when using a region growing algorithm to segment compressed point cloud data into regions and obtain regional point cloud data for each key feature region, the processing module 32 can specifically be used to screen candidate points on the bottom surface of the foundation pit from the compressed point cloud data and select seed points from them; starting from the seed points, perform region growing according to preset conditions to sequentially form preliminary regional point cloud data for the bottom surface region, side wall region, and excavation boundary region; and remove non-target point clouds from the preliminary regional point cloud data by using grayscale thresholds and distance thresholds to obtain regional point cloud data for each key feature region.

[0092] In some embodiments of this application, such as Figure 4 As shown, the device also includes: a detection module 35; The detection module 35 can be used to detect whether there is missing point cloud data in each key feature region based on the preset geometric attributes and point cloud density distribution of each key feature region. The processing module 32 can also be used to select the corresponding point cloud completion strategy based on the missing point cloud area of ​​the missing point cloud data if there are missing point cloud data in each key feature region, and to complete the missing point cloud data according to the point cloud completion strategy. The point cloud completion strategy includes at least using the neighborhood coordinate interpolation method to complete the missing point cloud data corresponding to the first missing point cloud area, and using the region growing algorithm to complete the missing point cloud data corresponding to the second missing point cloud area. The first missing point cloud area is less than or equal to a preset area threshold, and the second missing point cloud area is greater than the preset area threshold.

[0093] In some embodiments of this application, the key feature recognition model includes an input preprocessing module, a multi-scale feature enhancement module, and a feature matching module. The multi-scale feature enhancement module sets up a first feature extraction branch, a second feature extraction branch, a third feature extraction branch, and a feature fusion layer at different scales. The feature matching module has a built-in feature template library covering different types of foundation pits and defect types. Each feature template in the feature template library contains standard features and spatial constraint relationships between standard features. The standard features include standard sidewall busbar features, standard bottom surface contour features, and standard excavation boundary features. Correspondingly, input module 33 can be used to standardize the regional point cloud data of each key feature region through the input preprocessing module, and output regular point cloud data; input the side wall region point cloud data in the regular point cloud data into the first feature extraction branch, and use the small-scale convolution kernel configured in the first feature extraction branch to identify the symmetrical contour line covering the top to the bottom of the side wall as the initial side wall generatrix feature; input the bottom surface region point cloud data in the regular point cloud data into the second feature extraction branch, and use the medium-scale convolution kernel configured in the second feature extraction branch to identify the closed contour line of the bottom edge as the initial bottom surface contour line feature; input the excavation boundary region point cloud data in the regular point cloud data into the third feature extraction branch, and use the third feature extraction... The large-scale convolutional kernel configured in the branch identifies the continuous set of contour points at the top intersection as the initial excavation boundary features. The feature fusion layer is used to perform feature fusion and correlation analysis on the initial sidewall busbar features, initial bottom contour line features, and initial excavation boundary features to obtain fused features and feature space correlation information. In the feature matching module, based on the fused features and feature space correlation information, the similarity between the feature and each feature template in the feature template library is calculated. Target templates with similarity greater than or equal to the preset similarity threshold are selected. Based on the standard features and spatial constraints of the target templates, the initial sidewall busbar features, initial bottom contour line features, and initial excavation boundary features are calibrated to obtain the final excavation boundary features.

[0094] In some embodiments of this application, such as Figure 4 As shown, the device also includes: a training module 36; Training module 36 can be used to determine sample point cloud data and the key geometric features of the sample foundation pit corresponding to the sample point cloud data. Using the sample point cloud data as input features and the key geometric features of the sample foundation pit as training labels, iterative training of the key feature recognition model is performed until the loss function of the key feature recognition model is less than a preset threshold, at which point the training of the key feature recognition model is considered complete. Specifically, during the iterative training of the key feature recognition model, the initial features of the samples are extracted through a multi-scale feature enhancement module, and after feature fusion and correlation analysis, sample fusion features and sample feature space correlation information are obtained. The sample fusion features and sample feature space correlation information are matched with a feature template library through a feature matching module. Based on the matching results, the model outputs the predicted key geometric features of the foundation pit. The loss function is calculated based on the predicted key geometric features of the foundation pit and the sample key geometric features. When the loss function is greater than or equal to a preset threshold, the model parameters are adjusted in reverse, and the prediction of the key geometric features of the foundation pit is repeated.

[0095] In some embodiments of this application, the calculation module 34 can also be used to fit the reference geometric elements corresponding to the excavation pit of the transmission line based on the geometric properties of the sidewall busbar features; calculate the angles between the reference geometric elements and the vertical reference in each region, and take the maximum value of the angles in each region as the overall verticality deviation of the pit; determine the minimum elevation value of the bottom surface by fitting the minimum elevation surface of the bottom surface based on the point cloud data corresponding to the bottom surface contour features; call the ground reference surface elevation data collected by the inertial measurement unit (IMU) to calculate the difference between the ground reference surface elevation and the minimum elevation value of the bottom surface as the pit excavation depth; and apply principal component analysis to the point cloud data of the sidewall region associated with the sidewall busbar features. Extract the principal plane of the sidewall, determine the principal plane normal vector by calculating the point cloud covariance matrix, and construct the principal plane equation; based on the principal plane equation, traverse all point clouds in the sidewall region, calculate the distance deviation from each point cloud to the principal plane, and count the proportion of point clouds with distance deviations greater than a preset distance deviation threshold to the total number of point clouds in the sidewall, as the sidewall flatness; extract the key geometric information corresponding to the excavation boundary features, which includes closed contour point cloud data or feature point coordinates; perform geometric fitting on the closed contour point cloud data using the least squares method, or calculate the straight-line distance between adjacent feature points based on the feature point coordinates, to obtain the actual feature size characterizing the excavation range of the foundation pit, as the excavation size parameter.

[0096] It should be noted that other corresponding descriptions of the functional units involved in the excavation pit construction parameter extraction device based on point cloud data provided in this embodiment can be found in [reference needed]. Figure 1 and Figure 2 The corresponding descriptions in [the document] will not be repeated here.

[0097] Based on the above, Figure 1 and Figure 2 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method. Figure 1 and Figure 2 The method for extracting foundation pit construction parameters based on point cloud data is shown.

[0098] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause an electronic device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0099] Based on the above, Figure 1 and Figure 2 The method shown, and Figure 3 , 4To achieve the above objectives, the present application also provides an electronic device, specifically a personal computer, tablet computer, server, or other network device, as shown in the virtual device embodiment. This device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described objectives. Figure 1 and Figure 2 The method for extracting foundation pit construction parameters based on point cloud data is shown.

[0100] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0101] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0102] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0103] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.

[0104] This invention utilizes 3D point cloud data acquisition, replacing traditional manual data collection methods using tools such as measuring tapes, plumb bobs, and total stations. It eliminates the need for multiple people to collaboratively set up equipment, effectively reducing interference from environmental factors like rugged terrain, wind, rain, and strong sunlight during the inspection process. Pre-processing precisely segments point cloud data from key feature areas such as sidewalls, bottom surfaces, and excavation boundaries. Combined with a pre-trained key feature recognition model, it automatically extracts key geometric features such as the sidewall generatrix, bottom contour, and excavation boundaries, avoiding subjective errors caused by differences in operator experience. Finally, a geometric calculation model automatically calculates the overall verticality deviation, excavation depth, sidewall flatness, and excavation dimensions of the foundation pit, eliminating the need for manual data recording and calculation. This significantly shortens the inspection time for a single foundation pit while improving the stability of inspection accuracy, ultimately overcoming the limitations of traditional manual measurement and accurately meeting the stringent requirements of modern power engineering for construction efficiency and quality control.

[0105] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0106] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for extracting a foundation pit construction parameter based on point cloud data, characterized in that, The method comprises the following steps: collecting three-dimensional point cloud data of a power transmission line excavation pit; preprocessing the three-dimensional point cloud data to obtain regional point cloud data of each key feature region, the key feature region at least including a side wall region, a bottom surface region, and an excavation boundary region; inputting the regional point cloud data of each key feature region into a pre-trained key feature recognition model to perform feature recognition, to obtain pit key geometric features, the pit key geometric features at least including side wall generatrix features, bottom surface contour line features, and excavation boundary features; based on the pit key geometric features, calculating pit construction parameters through a geometric calculation model, the pit construction parameters at least including pit overall perpendicularity deviation, pit excavation depth, side wall flatness, and excavation size parameters.

2. The method of claim 1, wherein, The preprocessing of the three-dimensional point cloud data to obtain regional point cloud data of each key feature region comprises the following steps: performing denoising processing on the three-dimensional point cloud data by using a statistical filtering algorithm to remove interference point clouds; performing downsampling processing on the denoised three-dimensional point cloud data by using voxel grid downsampling, to obtain compressed point cloud data after retaining pit geometric features; performing regional segmentation on the compressed point cloud data by using a region growing algorithm, to obtain regional point cloud data of each key feature region.

3. The method of claim 2, wherein, The preprocessing of the three-dimensional point cloud data to obtain regional point cloud data of each key feature region comprises the following steps: selecting pit bottom surface candidate points from the compressed point cloud data and selecting seed points therefrom; taking the seed points as a starting point, performing region growing according to a preset condition, to sequentially form preliminary regional point cloud data of the bottom surface region, the side wall region, and the excavation boundary region; removing non-target point clouds in the preliminary regional point cloud data by using a gray threshold and a distance threshold, to obtain regional point cloud data of each key feature region.

4. The method of claim 1, wherein, Before inputting the regional point cloud data of each key feature region into a pre-trained key feature recognition model to perform feature recognition, the method further comprises the following steps: based on preset geometric attributes and point cloud density distribution of each key feature region, detecting whether there is missing point cloud data in each key feature region; if yes, based on a point cloud missing area of the missing point cloud data, selecting a corresponding point cloud completion strategy, and performing completion processing on the missing point cloud data according to the point cloud completion strategy, wherein the point cloud completion strategy at least includes using a neighborhood coordinate interpolation method to complete missing point cloud data corresponding to a first point cloud missing area smaller than or equal to a preset area threshold, and using a region growing algorithm to complete missing point cloud data corresponding to a second point cloud missing area larger than the preset area threshold.

5. The method of claim 1, wherein, The key feature recognition model comprises an input preprocessing module, a multi-scale feature enhancement module, and a feature matching module. The multi-scale feature enhancement module is provided with first, second, and third feature extraction branches of different scales and a feature fusion layer. The feature matching module internally comprises a feature template library covering different foundation pit types and defect types. Each feature template of the feature template library comprises standard features and spatial constraint relationships between the standard features. The standard features comprise standard sidewall generatrix features, standard bottom surface contour line features, and standard excavation boundary features. The region point cloud data of each key feature region is input into the pre-trained key feature recognition model for feature recognition to obtain key geometric features of the foundation pit, including: The region point cloud data of each key feature region is input into the pre-trained key feature recognition model for feature recognition to obtain key geometric features of the foundation pit, including: The sidewall region point cloud data in the regularized point cloud data is input into the first feature extraction branch, and small-scale convolution kernels configured in the first feature extraction branch are used to identify a symmetric contour line covering the top to the bottom of the sidewall as an initial sidewall generatrix feature. The bottom surface region point cloud data in the regularized point cloud data is input into the second feature extraction branch, and medium-scale convolution kernels configured in the second feature extraction branch are used to identify a closed contour line of the edge of the bottom surface as an initial bottom surface contour line feature. The excavation boundary region point cloud data in the regularized point cloud data is input into the third feature extraction branch, and large-scale convolution kernels configured in the third feature extraction branch are used to identify a continuous contour point set at the top junction as an initial excavation boundary feature. The feature fusion layer is used to perform feature fusion and correlation analysis on the initial sidewall generatrix feature, the initial bottom surface contour line feature, and the initial excavation boundary feature to obtain fused features and feature spatial correlation information. In the feature matching module, the similarity between each feature template in the feature template library and the fused features and the feature spatial correlation information is calculated, and a target template corresponding to a similarity greater than or equal to a preset similarity threshold is screened. The standard features and the spatial constraint relationships of the target template are used as a reference to calibrate the initial sidewall generatrix feature, the initial bottom surface contour line feature, and the initial excavation boundary feature, respectively, to obtain final excavation boundary features.

6. The method of claim 5, wherein, The method further comprises a training method of the key feature recognition model; Sample point cloud data and sample foundation pit key geometric features corresponding to the sample point cloud data are determined; The sample point cloud data is used as input features, and the sample foundation pit key geometric features are used as training labels. Iterative training of the key feature recognition model is performed until the loss function of the key feature recognition model is less than a preset threshold, and it is determined that the training of the key feature recognition model is completed. In the iterative training of the key feature recognition model, the sample initial features are extracted through the multi-scale feature enhancement module, and the sample fusion features and sample feature space correlation information are obtained through feature fusion and correlation analysis. The sample fusion features and sample feature space correlation information are matched with the feature template library through the feature matching module, and the predicted key geometric features of the foundation pit are output based on the matching results. The loss function is calculated based on the predicted key geometric features of the foundation pit and the sample key geometric features of the foundation pit. When the loss function is greater than or equal to the preset threshold, the model parameters are adjusted in reverse, and the prediction of the key geometric features of the foundation pit is repeated.

7. The method of claim 1, wherein, The key geometric features of the foundation pit are used to calculate the construction parameters of the foundation pit through a geometric calculation model, including: According to the geometric properties of the side wall generatrix feature, the reference geometric element corresponding to the power transmission line excavation foundation pit is fitted; the angles between the reference geometric element and the vertical reference of each region are calculated, and the maximum value of the angles is taken as the overall verticality deviation of the foundation pit; Based on the point cloud data corresponding to the bottom surface contour line feature, the bottom surface lowest elevation value is determined by fitting the bottom surface lowest elevation surface; the difference between the ground reference surface elevation data collected by the inertial measurement unit (IMU) and the bottom surface lowest elevation value is calculated as the excavation depth of the foundation pit; The principal component analysis is used to extract the side wall main plane from the point cloud data of the side wall region associated with the side wall generatrix feature, and the principal plane equation is constructed by calculating the principal plane normal vector; based on the principal plane equation, all point clouds in the side wall region are traversed, the distance deviation of each point cloud to the principal plane is calculated, and the proportion of point clouds with distance deviation greater than a preset distance deviation threshold to the total number of side wall point clouds is calculated as the side wall flatness. The key geometric information corresponding to the excavation boundary feature is extracted, including closed contour point cloud data or feature point coordinates; the actual feature size representing the excavation range of the foundation pit is obtained by geometric fitting of the closed contour point cloud data through the least squares method or calculation of the straight line distance between adjacent feature points based on the feature point coordinates, and is taken as the excavation size parameter. 8.A device for extracting a foundation pit construction parameter based on point cloud data, characterized in that, It includes: The acquisition module is used to acquire three-dimensional point cloud data of the power transmission line excavation foundation pit; The processing module is used to preprocess the three-dimensional point cloud data to obtain regional point cloud data of each key feature region, and the key feature region at least includes a side wall region, a bottom surface region and an excavation boundary region; The input module is used to input the regional point cloud data of each key feature region into a pre-trained key feature recognition model for feature recognition to obtain key geometric features of the foundation pit, and the key geometric features of the foundation pit at least include side wall generatrix features, bottom surface contour line features and excavation boundary features; The calculation module is used to calculate the construction parameters of the foundation pit based on the key geometric features of the foundation pit through a geometric calculation model, and the construction parameters of the foundation pit at least include overall verticality deviation of the foundation pit, excavation depth of the foundation pit, side wall flatness and excavation size parameter.

9. A storage medium having stored thereon a computer program, characterized in that The computer program, which is executed by a processor, implements the method of any one of claims 1 to 7.

10. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, The processor, when executing the computer program, implements the method of any one of claims 1 to 7.

Images