A foundation pit working condition real-time analysis method based on holographic perception and semantic recognition

Abstract

The application discloses a kind of based on holographic perception and semantic recognition's foundation pit working condition instant analysis method, comprising the following steps: carry out the selection and parameter configuration of foundation pit attention feature;Machine vision data acquisition work is carried out by the form of camera and point cloud radar combination;The target point cloud data set T is obtained by carrying out semantic recognition analysis to the machine vision data collected;The position of target point cloud data set T in the point cloud data set P of total amount is calculated, and the element identification of construction object and construction area is carried out according to the set semantic label parameter, and the feature index of construction area is extracted;Working condition sensitive perception and continuous application.The application has the advantages that: the parameter planning, instant solution and efficient analysis of underground engineering construction working condition attention feature are realized, object elements in project can be automatically "understood", working condition changes are discovered early in "will move but not move" moment to achieve early warning, and methods and tools are provided for efficient and accurate foundation pit engineering control.

Description

Technical Field

[0001] This invention belongs to the field of foundation pit construction technology, specifically involving a method for real-time analysis of foundation pit working conditions based on holographic perception and semantic recognition. Background Technology

[0002] In the construction of underground spaces, the construction conditions are crucial, directly affecting the stability, safety, and economy of the project. During underground excavation, the soil is disturbed, and reasonable excavation conditions can effectively control the project's environmental impact, ensuring stability and construction safety. Excavation conditions mainly present risks such as over-excavation and untimely support construction; many engineering accidents often stem from oversights in condition management. Traditional methods rely primarily on manual inspections or retrospective review and judgment of monitoring images, and the analysis results are not easily applicable to various analyses. Overall, risk information collection suffers from issues of timeliness, data accuracy, and analytical usability. Furthermore, existing models cannot cover construction excavation around the clock, creating regulatory blind spots and are unsuitable for such large-scale foundation pit projects with numerous overlapping construction sequences. The acquisition and analysis of dynamic conditions has become an important direction for technological development in the industry.

[0003] Machine vision is a technology that uses visual elements such as images and point clouds to acquire information about objects. It can be used to acquire and analyze shape and appearance data of construction elements in underground engineering projects. However, underground engineering projects are complex due to visual occlusion, continuous dynamic changes, and uncertain identification features, posing significant challenges to the automated identification and calculation efficiency of underground engineering condition information analysis. While some machine vision-based monitoring and data processing solutions exist, existing methods and technologies often employ data collection and calculation methods for fixed areas / designated points. On the one hand, the extracted data represents macroscopic indicators of a specified area / object, calculated through averaging, interpolation, and point cloud patching. These macroscopic indicators may "smooth out" the risks of local anomalies. On the other hand, due to the coarse granularity of monitoring, it is impossible to sensitively locate the area under construction through changes in indicator data, nor can it quickly pinpoint the area where a risk occurs in the first instance. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of the prior art by providing a real-time analysis method for foundation pit working conditions based on holographic perception and semantic recognition. This real-time analysis method uses collected machine vision elements to design effective strategies and methods for analysis, calculation, and data processing, completing the conversion from real-world features of underground engineering to working condition index data, thus providing support for underground engineering construction monitoring and automated working condition analysis.

[0005] The objective of this invention is achieved through the following technical solutions: A real-time analysis method for foundation pit working conditions based on holographic perception and semantic recognition, the real-time analysis method comprising the following steps: S1: Feature and Parametric Programming Select and configure the features of interest for the foundation pit, including the acquisition parameters, detection scene parameters, target area and semantic label parameters. S2: Data Input: Based on the preset acquisition parameters, machine vision data acquisition is carried out through a combination of camera and point cloud radar; the acquired machine vision data is converted into single-frame image data I, point cloud dataset P, and image-point cloud coupling relationship matrix. S3: Semantic Recognition A large semantic segmentation model is used to perform image segmentation and basic semantic extraction on each single frame image data I. A two-dimensional thematic semantic recognition model is used to analyze the image segmentation results and take the union of the basic semantic recognition and thematic semantic recognition. The union is processed by a verification function to obtain the target object dataset, which is the image semantic segmentation output result. Using the image-point cloud coupling relationship matrix, the geometric features of the image semantic segmentation output in the two-dimensional range are transformed by spatial mapping, and the image range is compared with the point cloud dataset P to obtain the target object and store it as the target point cloud dataset T. S4: Data Settlement The point cloud dataset P is processed according to the preset detection scene parameters to make the spatial vector of the working condition change direction in the scene consistent with the preset detection scene parameters. Calculate the position of the target point cloud dataset T in the full point cloud dataset P, and identify the construction object and construction area elements according to the set semantic label parameters; Based on the identified construction objects and construction areas, feature indicators of the construction areas are extracted. S5: Operating Condition Sensing and Continuous Application: Based on the dynamic monitoring of construction areas and construction objects, a preset rule model is adopted to input the characteristic indicators of the construction area into the rule model for calculation, and to determine whether different sequences of working conditions are reached, so as to realize the automatic evolution and identification of working conditions. Repeat the above steps to implement full-process monitoring of foundation pit construction.

[0006] In step S1, the configuration acquisition parameters are the data acquisition frequency, the detection scene parameters are the scene registration parameters with the geoid screen normal direction as the scene registration parameter, the target area is the excavation surface of the foundation pit, and the semantic label parameters are the excavation machinery labels used.

[0007] In step S3, the semantic segmentation big model is the SAM semantic segmentation big model; the image segmentation refers to segmenting personnel elements, mechanical parts elements, and scene elements; the two-dimensional thematic semantic recognition model is the U-Net two-dimensional thematic semantic recognition model.

[0008] In step S4, the semantic label parameter is the excavation machinery label. The location of the excavation operation is captured by the positioning of the excavation machinery point cloud, and the construction target is located before the change in the operation state. Based on the identified excavation machinery elements, the actual excavation depth of the current foundation pit is estimated by the empirical formula of "target point cloud dataset T bounding box center elevation - standard data of center height of this type of excavator equipment". The actual excavation depth of the foundation pit is used as the feature index of the construction area for extraction.

[0009] In step S5, the rule model is a calculation model that sets an excavation depth threshold within a specified time period. The actual excavation depth of the foundation pit is used as a characteristic indicator of the construction area and substituted into the rule model for calculation. The excavation depth and the threshold are compared to determine whether over-excavation or completion of the excavation standard for this stage has been achieved.

[0010] The advantages of this invention are: it enables parameter planning, real-time calculation and efficient analysis of the characteristics of underground engineering construction conditions, and further, it can automatically "understand" the object elements in the project, and keenly detect changes in the working conditions at the moment of "before the movement", thus providing a method and tool for efficient and accurate foundation pit engineering management. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the real-time analysis method for foundation pit working conditions based on holographic perception and semantic recognition in this invention. Detailed Implementation

[0012] The features and other related features of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments, so as to facilitate understanding by those skilled in the art: Example: Figure 1 As shown, this embodiment specifically relates to a method for real-time analysis of foundation pit conditions based on holographic perception and semantic recognition, including the following steps: S1: Feature and Parametric Programming This embodiment focuses on the typical support type of foundation pit engineering, which combines concrete supports and diaphragm walls. It selects the features of interest in the foundation pit and configures the parameters, including the acquisition parameters, detection scene parameters, target area, and semantic label parameters.

[0013] The data acquisition parameters are configured as follows: data acquisition frequency, such as data acquisition frequency every 2 hours; detection scene parameters are based on the geoid screen normal direction as the scene registration parameter; target area is the excavation face of the foundation pit; semantic label parameters are the labels of the excavation machinery used.

[0014] S2: Data Input: Based on the preset acquisition parameters of the parameter configuration module, machine vision data acquisition is carried out through a combination of camera and point cloud radar. The acquired machine vision data is converted into single-frame image data I, point cloud dataset P, and image-point cloud coupling relationship matrix for subsequent process processing. These three types of data are conducive to using visual images for high-accuracy and rapid semantic extraction, and combining two-dimensional and three-dimensional linkage to focus on the target construction object. Combined with mathematical calculation methods, the solution efficiency is improved and the indicators of changes in the working conditions of the construction area are sensitively identified.

[0015] S3: Semantic Recognition The SAM semantic segmentation model is used to perform image segmentation and basic semantic extraction on each single frame image data I. The image segmentation refers to segmenting personnel elements, mechanical parts elements, scene elements, etc. The image segmentation results are then substituted into a pre-trained U-Net two-dimensional thematic semantic recognition model for analysis (this model can be mainly trained on the excavation machinery sample set). The image segmentation results are analyzed and the union is taken through basic semantic recognition and thematic semantic recognition. After the union is processed by the verification function, the target object dataset is obtained. The target object dataset is the image semantic segmentation output result.

[0016] Since this study uses excavation machinery to estimate the current construction area, the geometric features of the excavation machinery's image semantic segmentation output in the two-dimensional range are obtained by using the image-point cloud coupling matrix. Through spatial mapping transformation, the image range is compared with the point cloud dataset P to obtain the point cloud elements of the target object (excavation machinery object) in the machine vision results and store them as the target point cloud dataset T.

[0017] S4: Data Settlement Based on the detection scenario parameters preset in the parameter planning stage, the point cloud dataset P is processed in the scene coordinate system so that the spatial vector of the working condition change direction in the scene is consistent with the preset detection scenario parameters. This step transforms the professional business problem of identifying complex working condition changes into a mathematical solution mode for extracting key features through spatial direction mapping, and improves the detection accuracy by utilizing prior direction information.

[0018] [Construction Target / Area Identification] The location of the target point cloud dataset T within the full point cloud dataset P is calculated, and the construction object and construction area are identified based on the set semantic label parameters (including but not limited to excavation machinery, support objects, etc.). This step can clearly identify the object / area currently under construction, or infer the object / area under construction based on construction elements (excavation machinery, personnel, etc.), enabling sensitive capture of the location of excavation conditions and locating the construction target before abrupt changes in the working condition (existing methods or technologies lack this sensitive capture and use fixed area / point or full area average calculation methods to obtain the data).

[0019] Specifically, the position of the target point cloud dataset T within the full dataset P is calculated, and construction production elements are identified based on the excavator machinery labels. This step, through the localization of the excavator machinery point cloud, is used for sensitive capture of the location of excavation conditions, locating the construction target before abrupt changes in the working conditions.

[0020] [Regional Feature Index Extraction] Based on the identified construction objects, construction areas, and other elements, feature indicators of the focused area are extracted. For example, the elevation information of the object / area under construction is directly extracted after identification; in the mode of inferring the current construction object / area through construction elements (excavation machinery, personnel, etc.), the object height and elevation are converted and extracted.

[0021] Specifically, based on the identified excavation machinery elements, the actual excavation depth of the current foundation pit is estimated using the empirical formula of "target point cloud dataset T bounding box center elevation - standard data of center height of this type of excavator equipment". The actual excavation depth of the foundation pit is then used as a feature indicator of the construction area for extraction.

[0022] S5: Operating Condition Sensing and Continuous Application: [Automatic Working Condition Judgment] Based on dynamic monitoring of the construction area and construction objects, a preset rule model is used. The characteristic indicators of the construction area are substituted into the rule model for calculation to determine whether different sequences of working conditions have been reached, thus achieving automatic evolution and identification of working conditions. Specifically, the rule model is a calculation model that sets an excavation depth threshold within a specified time period. The actual excavation depth of the foundation pit is used as a characteristic indicator of the construction area and substituted into the rule model for calculation. By comparing the excavation depth with the threshold, it is determined whether over-excavation or completion of the current stage's excavation standard has been achieved. This achieves the effect of automatic evolution and identification of working conditions. This step, on the one hand, understands changes in regional working conditions through sensitive perception, thus enhancing timeliness, and on the other hand, provides early warning, location, and detection of high-risk and sensitive working conditions in a timely manner.

[0023] [Working Condition Data Analysis] In addition, the acquired working condition characteristic data can also be used for subsequent specialized analysis.

[0024] Repeat the above steps to implement full-process monitoring of foundation pit construction.

[0025] The beneficial effects of this embodiment are as follows: (1) In view of the problems of automated identification and intelligent capture of dynamic working conditions in the analysis of working conditions of foundation pit engineering, combined with the complex characteristics of underground engineering environment such as visual occlusion, continuous dynamic changes, uncertain identification features, and low sensitivity and timeliness of existing solutions, a real-time identification method for foundation pit working conditions based on holographic perception and semantic recognition is proposed. By collecting machine vision elements, effective strategies and methods are designed for analysis, calculation and data processing, and the conversion from real-scene features of underground engineering to working condition index data is completed, providing support for construction monitoring and automated analysis of working conditions of underground engineering.

[0026] (2) It enables parameter planning, real-time calculation and efficient analysis of the characteristics of underground engineering construction conditions. At the same time, it can automatically "understand" the object elements in the project and keenly detect changes in the working conditions at the moment of "movement before movement" so as to achieve early warning. It provides methods and tools for efficient and accurate foundation pit engineering management.

Claims

1. A method for real-time analysis of foundation pit working conditions based on holographic perception and semantic recognition, characterized in that... The real-time analysis method includes the following steps: S1: Feature and Parametric Programming Select and configure the features of interest for the foundation pit, including the acquisition parameters, detection scene parameters, target area and semantic label parameters. S2: Data Input: Based on the preset acquisition parameters, machine vision data acquisition is carried out through a combination of camera and point cloud radar; the acquired machine vision data is converted into single-frame image data I, point cloud dataset P, and image-point cloud coupling matrix; S3: Semantic recognition: A large semantic segmentation model is used to perform image segmentation and basic semantic extraction on each single frame image data I. A two-dimensional thematic semantic recognition model is used to analyze the image segmentation results and take the union of the basic semantic recognition and thematic semantic recognition. The union is processed by a verification function to obtain the target object dataset, which is the image semantic segmentation output result. Using the image-point cloud coupling relationship matrix, the geometric features of the image semantic segmentation output in the two-dimensional range are transformed by spatial mapping, and the image range is compared with the point cloud dataset P to obtain the target object and store it as the target point cloud dataset T. S4: Data Settlement The point cloud dataset P is processed according to the preset detection scene parameters to make the spatial vector of the working condition change direction in the scene consistent with the preset detection scene parameters. Calculate the position of the target point cloud dataset T in the full point cloud dataset P, and identify the construction object and construction area elements according to the set semantic label parameters; Based on the identified construction objects and construction areas, feature indicators of the construction areas are extracted. S5: Operating Condition Sensing and Continuous Application: Based on the dynamic monitoring of construction areas and construction objects, a preset rule model is adopted to input the characteristic indicators of the construction area into the rule model for calculation, and to determine whether different sequences of working conditions are reached, so as to realize the automatic evolution and identification of working conditions. Repeat the above steps to implement full-process monitoring of foundation pit construction; In step S1, the configuration acquisition parameters are the data acquisition frequency, the detection scene parameters are the scene registration parameters with the geoid screen normal direction as the scene registration parameter, the target area is the excavation surface of the foundation pit, and the semantic label parameters are the excavation machinery labels used. In step S3, the semantic segmentation large model is the SAM semantic segmentation large model; the image segmentation refers to segmenting personnel elements, mechanical parts elements, and scene elements; the two-dimensional thematic semantic recognition model is the U-Net two-dimensional thematic semantic recognition model. In step S4, the semantic label parameter is the excavation machinery label. The location of the excavation operation is captured by the positioning of the excavation machinery point cloud, and the construction target is located before the change in the operation state. Based on the identified excavation machinery elements, the actual excavation depth of the current foundation pit is estimated by the empirical formula of "target point cloud dataset T bounding box center elevation - standard data of center height of this type of excavator equipment". The actual excavation depth of the foundation pit is used as the feature index of the construction area for extraction.

2. The method for real-time analysis of foundation pit working conditions based on holographic perception and semantic recognition according to claim 1, characterized in that... In step S5, the rule model is a calculation model that sets an excavation depth threshold within a specified time period. The actual excavation depth of the foundation pit is used as a characteristic indicator of the construction area and substituted into the rule model for calculation. The excavation depth and the threshold are compared to determine whether over-excavation or completion of the excavation standard for this stage has been achieved.

Images