3D modeling method for foundation pit deformation monitoring

Abstract

The invention relates to the technical field of foundation pit monitoring, and particularly discloses a 3D modeling method for foundation pit deformation monitoring, which comprises the following steps: S1, performing gridding processing on a monitoring area to obtain a sub-area, and constructing a 3D model of the sub-area as a sub-model; s2, obtaining a target sequence, and recording the similarity between the first target model in the target sequence and the target model at the ith sequence position; s3, generating coordinate points according to the similarity degree, and fitting the coordinate points to obtain a fitted curve; calculating a judgment score based on the fitted curve, and determining whether the curve is an abnormal curve according to the judgment score; and determining an abnormal time point based on the abnormal curve, and sending early warning information for prompting. Possible foundation pit deformation can be found in advance, and the safety and reliability of the foundation pit are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of foundation pit monitoring, and particularly relates to a 3D modeling method for foundation pit deformation monitoring. Background Art

[0002] A foundation pit refers to a temporary or permanent pit excavated in construction projects for underground construction, usually used in the construction of building basements, subways, etc. The excavation depth of the foundation pit can vary from several meters to dozens of meters according to needs. During the construction of the foundation pit, the pit wall needs to be supported to prevent soil collapse or landslide and ensure the safety of the surrounding environment.

[0003] Foundation pit deformation monitoring refers to the real-time monitoring of the deformation of the soil body, structures and equipment around the foundation pit during the excavation and support process of the foundation pit to ensure safety during construction, prevent the deformation of the foundation pit and the surrounding environment from exceeding the design range, and avoid damage to surrounding buildings, roads, underground pipelines, etc.

[0004] In the existing methods for monitoring foundation pit deformation through 3D modeling technology, most of them issue alarms when the deformation reaches a certain level. However, at this time, the deformation of the foundation pit may already be relatively serious, and there is a certain lag in dealing with it at this time. Therefore, how to detect possible abnormalities in advance has become an urgent problem to be solved. Summary of the Invention

[0005] The purpose of the present invention is to provide a 3D modeling method for foundation pit deformation monitoring to solve the following technical problems:

[0006] In the existing methods for monitoring foundation pit deformation through 3D modeling technology, most of them issue alarms when the deformation reaches a certain level. However, at this time, the deformation of the foundation pit may already be relatively serious, and there is a certain lag in dealing with it at this time.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] A 3D modeling method for foundation pit deformation monitoring includes the following steps:

[0009] S1: Set a monitoring area, perform grid processing on the monitoring area to obtain sub-areas, the length and width of the sub-areas are both preset values, collect the point cloud data of the sub-areas, and perform 3D modeling based on the point cloud data to obtain sub-models;

[0010] S2: Set time nodes at preset time intervals within a preset monitoring period. At these time nodes, obtain the sub - models corresponding to the same sub - region, denoted as target models. Sort the target models in chronological order to obtain a target sorting. Obtain the similarity degree, denoted as Ki, between the target model at the first position in the target sorting and the target model at the i - th sorting position.

[0011] Among them, in the process of obtaining the target sorting, the following steps are also included:

[0012] Take the target model corresponding to time node a as a pending model. At time node a, collect an image of the sub - region and identify whether there is a preset entity target in the image. If so, delete the pending model from the target sorting to obtain a new target sorting; if not, retain the pending model in the target sorting.

[0013] S3: Generate coordinate points (ti, Ki), where ti represents the time point when obtaining the target model at the i - th sorting position in the target sorting. Fit the coordinate points to obtain a fitting curve f(t), where t represents time.

[0014] Obtain the tangent slope of each point on the fitting curve, calculate the average tangent slope and the minimum tangent slope, and perform a weighted sum of the average tangent slope and the minimum tangent slope to obtain a judgment score P. When the judgment score P is less than or equal to a preset judgment score threshold, regard the corresponding curve as an abnormal curve.

[0015] Based on the abnormal curve, determine the abnormal time point and send a warning message to prompt the preset management personnel that there is an abnormality in the sub - region corresponding to the abnormal curve at the abnormal time point.

[0016] As a further solution of the present invention: In step S3, the process of determining the abnormal time point specifically includes:

[0017] Set a similarity degree threshold Kys, substitute the similarity degree threshold Kys into the function relation of the abnormal curve, and solve to obtain the abnormal time point.

[0018] As a further solution of the present invention: In step S2, the target entities include construction workers and construction vehicles.

[0019] As a further solution of the present invention: In step S2, based on a pre - trained target entity recognition model, identify whether there is a target entity in the image.

[0020] As a further solution of the present invention: The process of training the target entity recognition model specifically includes:

[0021] A database is established, in which images with labeled tags are stored. The tags are the names of target entities and 0. A tag of 0 indicates that the target entity does not exist in the image.

[0022] Based on a deep learning model, a target entity recognition model is established, and the target entity recognition model is trained and verified through the database to obtain a pre-trained target entity recognition model.

[0023] As a further solution of the present invention: in step S3, the judgment score P = η1*A1 + η1*A2, where η1 and η1 are preset first weight and second weight, and 0 < η1 < η1, and A1 and A2 respectively represent the average tangent slope and the minimum tangent slope.

[0024] As a further solution of the present invention: in step S2, the following steps are further included:

[0025] When Ki ≤ Kys, the subsequent steps are stopped, and a warning message is sent to the management staff for prompting.

[0026] As a further solution of the present invention: in step S2, when the number of target models in the target sorting is less than a preset number threshold, a prompt message is sent to the management staff.

[0027] As a further solution of the present invention: in step S3, the following steps are further included:

[0028] When the average tangent slope is less than a preset average tangent slope threshold and / or the minimum tangent slope is less than a preset minimum tangent slope threshold, it is determined that the corresponding curve is not an abnormal curve.

[0029] As a further solution of the present invention: in step S3, if the average tangent slope of the curve is less than a preset average tangent slope threshold and / or the minimum tangent slope is less than a preset minimum tangent slope threshold, then in the next monitoring period adjacent to the current time, this sub-region is not monitored.

[0030] Advantages of the present invention: In this solution, first, by setting up a monitoring area and dividing it into a grid, the monitoring area can be accurately divided into multiple sub-areas, each with clear length and width boundaries; this makes subsequent data collection and modeling more systematic, facilitating separate modeling of each sub-area, thereby improving the modeling accuracy and efficiency. The collected point cloud data can help generate sub-models, which provides reliable basic data for subsequent comparison and analysis; through 3D modeling based on the point cloud data, an accurate virtual representation of the sub-area can be obtained, which helps to judge the similarity between models in the subsequent stage; then, by setting time nodes at preset time intervals within a preset monitoring period, the sub-models corresponding to the same sub-area are gradually obtained, forming time series data, establishing the dynamic characteristics of the target model over time, and providing time series data support for subsequent detection of model changes; in the sorted target models, by calculating the similarity degree Ki, the changes between models at different time nodes can be quantified, thereby monitoring the speed of model changes and providing more accurate positioning for abnormal situations; the similarity degree Ki can be determined based on point cloud comparison. Specifically, the similarity between models is evaluated by calculating the distance between corresponding points in the models. Common calculation methods include calculating the shortest distance (e.g., the distance to the nearest point) or the point-to-point error (such as the Hausdorff distance); it should be noted that in the detection of the target entity, by identifying whether there are entity targets (such as people, vehicles, etc.) in the image, unnecessary interference can be avoided, and the undetermined models in the target sorting can be removed in a timely manner, further improving the accuracy and reliability of similarity judgment, which plays a very important role in improving the true reflection of the model and ensuring that subsequent judgments and analyses are based on pure environmental data rather than errors; then, by generating coordinate points and fitting them, a mathematical expression (fitting curve) of the model change can be obtained, and then the tangent slope is used to reflect the trend of the model change. The weighted sum of the average tangent slope and the minimum tangent slope is used as the judgment score P, which can effectively reflect the stability and abnormality of the model change. If the judgment score P is less than or equal to the preset threshold, an abnormal curve can be identified, and then the abnormal time point can be determined; by quantifying the changes in the target area through a mathematical model, the deviation of subjective judgment is avoided, potential abnormalities can be discovered in a timely manner, and accurate warning information can be provided to the management personnel; the present invention effectively avoids interference factors (such as people, vehicles, etc.) in the monitoring area, ensuring the purity and accuracy of data collection and model construction; at the same time, by combining time series data with model similarity analysis, the accuracy of the monitoring process is further improved; and through the analysis of the tangent slope of the fitting curve, abnormal changes can be detected at an earlier stage and warnings can be issued in a timely manner, thereby ensuring the real-time and reliability of the monitoring work. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present invention will be further described below in conjunction with the accompanying drawings.

[0032] Figure 1 It is a schematic flowchart of a 3D modeling method for foundation pit deformation monitoring according to the present invention. Specific embodiments

[0033] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0034] Please refer to Figure 1 As shown, the present invention is a 3D modeling method for foundation pit deformation monitoring, including the following steps:

[0035] S1: Set a monitoring area, perform grid processing on the monitoring area to obtain sub-areas, the length and width of the sub-areas are both preset values, collect point cloud data of the sub-areas, and perform 3D modeling based on the point cloud data to obtain sub-models;

[0036] S2: Set time nodes at preset time intervals within a preset monitoring period. Obtain the sub-models corresponding to the same sub-area at the time nodes, denoted as target models. Sort the target models in the order of the time axis to obtain a target sorting. Obtain the similarity degree between the target model at the first position in the target sorting and the target model at the i-th sorting position, denoted as Ki;

[0037] Among them, in the process of obtaining the target sorting, the following steps are further included:

[0038] Take the target model corresponding to time node a as a to-be-determined model. Collect an image of the sub-area at time node a, and identify whether there is a preset entity target in the image. If so, delete the to-be-determined model from the target sorting to obtain a new target sorting; if not, retain the to-be-determined model in the target sorting;

[0039] S3: Generate coordinate points (ti, Ki), where ti represents the time point when obtaining the target model at the i-th sorting position in the target sorting. Fit the coordinate points to obtain a fitting curve f(t), where t represents time;

[0040] Obtain the tangent slope of each point on the obtained fitting curve, calculate the average tangent slope and the minimum tangent slope, perform weighted summation on the average tangent slope and the minimum tangent slope to obtain a judgment score P. When the judgment score P is less than or equal to a preset judgment score threshold, the corresponding curve is regarded as an abnormal curve;

[0041] Based on the abnormal curve, determine the abnormal time point, and send a warning message to prompt the preset management personnel that there is an abnormality in the sub-region corresponding to the abnormal curve at the abnormal time point.

[0042] It should be noted that by first setting the monitoring area and performing grid processing on it, the monitoring area can be accurately divided into multiple sub-areas, each of which has clear length and width boundaries; this makes subsequent data collection and modeling more systematic, facilitating individual modeling of each sub-area, thereby improving the modeling accuracy and efficiency. The collected point cloud data can help generate sub-models, which provides reliable basic data for subsequent comparison and analysis; through 3D modeling based on the point cloud data, an accurate virtual representation of the sub-area can be obtained, which helps to judge the similarity between models in the subsequent stage; then, by setting time nodes at preset time intervals within a preset monitoring period, the sub-models corresponding to the same sub-area are gradually obtained, forming time series data, and the dynamic characteristics of the target model changing over time are established, providing time series data support for subsequent detection of model changes; in the sorted target models, by calculating the similarity degree Ki, the changes between models at different time nodes can be quantified, thereby monitoring the speed of model changes and providing more accurate positioning for abnormal situations; the similarity degree Ki can be determined based on point cloud comparison. Specifically, the similarity between models is evaluated by calculating the distance between corresponding points in the models. Common calculation methods include calculating the shortest distance (e.g., the distance to the nearest point) or the point-to-point error (such as the Hausdorff distance); it is worth noting that in the detection of target entities, by identifying whether there are entity targets (such as personnel, vehicles, etc.) in the image, unnecessary interference can be avoided, and the undetermined models in the target sorting can be removed in a timely manner, further improving the accuracy and reliability of similarity judgment, which plays a very important role in ensuring that the model truly reflects the situation and ensuring that subsequent judgments and analyses are based on pure environmental data rather than errors; then, by generating coordinate points and performing fitting, a mathematical expression (fitting curve) of model changes can be obtained, and then the tangent slope is used to reflect the trend of model changes. The weighted sum of the average tangent slope and the minimum tangent slope is used as the judgment score P, which can effectively reflect the stability and abnormality of model changes. If the judgment score P is less than or equal to the preset threshold, an abnormal curve can be identified, and then the abnormal time point can be determined; by quantifying the changes in the target area through a mathematical model, the deviation of subjective judgment is avoided, potential abnormalities can be discovered in a timely manner, and accurate early warning information can be provided for management personnel; the present invention effectively avoids interference factors (such as personnel, vehicles, etc.) in the monitoring area, ensuring the purity and accuracy of data collection and model construction; at the same time, by combining time series data with model similarity analysis, the accuracy of the monitoring process is further improved; and through the analysis of the tangent slope of the fitting curve, abnormal changes can be detected at an earlier stage and early warnings can be issued in a timely manner, thus ensuring the real-time and reliability of the monitoring work.

[0043] In another preferred embodiment of the present invention, in step S3, the process of determining the abnormal time point specifically includes:

[0044] Set a similarity threshold Kys, substitute the similarity threshold Kys into the functional relationship of the abnormal curve, and solve to obtain the abnormal time point.

[0045] It should be noted that taking the functional relationship as F(t) as an example, that is, let F(t)=Ky to solve, and the solving process belongs to the prior art and will not be elaborated here.

[0046] In another preferred embodiment of the present invention, in step S2, the target entities include construction workers and construction vehicles.

[0047] In another preferred embodiment of the present invention, in step S2, based on a pre-trained target entity recognition model, it is identified whether there are target entities in the image.

[0048] It can be understood that the pre-trained model can automatically and accurately identify specific targets (such as people, vehicles, animals, tools, etc.); if these "target entities" appear in the image, it means that the current sub-model may be interfered by temporary and external factors (rather than changes in the environment itself). Removing the 3D models of such target entities can avoid interference in the subsequent analysis of environmental changes and reduce the misjudgment rate; traditional methods may require manual frame-by-frame inspection to check whether there are specific interfering objects in the image; the pre-trained model can perform recognition automatically in batches, greatly reducing the labor input. Moreover, the pre-trained model is usually trained based on a large number of sample data, has strong robustness and generalization ability, and can maintain good recognition performance under different lighting, angle, and background conditions, ensuring the consistency and stability of data screening.

[0049] In another preferred embodiment of the present invention, the process of training the target entity recognition model specifically includes:

[0050] Establish a database, which stores images with labeled tags. The tag is the name of the target entity and 0. The tag of 0 indicates that there is no such target entity in the image;

[0051] Based on a deep learning model, establish a target entity recognition model, and train and verify the target entity recognition model through the database to obtain a pre-trained target entity recognition model.

[0052] It should be noted that by storing the labeled positive examples (including target entities) and negative examples (with a label of 0, indicating no target entity in the image) in the database, clear and rich samples can be provided for the model, thus ensuring data diversity and accuracy during training, effectively improving the generalization ability and robustness of the model; since the database contains both images with target entities and images without target entities (negative examples), the model can learn distinct differential features, enabling the trained model to accurately identify target entities and reducing the probability of missed detection and false detection; with the help of a deep learning model, the training process can be highly automated, and the recognition effect is also automatically evaluated and improved during the verification stage to reduce manual intervention and time costs and improve the processing efficiency of the overall system; by collecting multi-dimensional samples such as various scenarios, angles, lighting, and target sizes in the database, the trained model can adapt to different actual monitoring environments, enhancing the stability and reliability of the recognition model, and thus having a more accurate entity recognition performance in subsequent monitoring scenarios; the trained model can be used as a "pre-filter" to identify images with specific entities (people, vehicles, equipment, etc.), thereby removing the corresponding 3D models, making the subsequent analysis of environmental change trends more "clean" and ensuring more focused and effective monitoring of anomalies.

[0053] In another preferred embodiment of the present invention, in step S3, the judgment score P = η1*A1 + η1*A2, where η1 and η1 are preset first and second weights, and 0 < η1 < η1, and A1 and A2 respectively represent the average tangent slope and the minimum tangent slope.

[0054] In another preferred embodiment of the present invention, in step S2, the following steps are further included:

[0055] When Ki ≤ Kys, the subsequent steps are stopped, and a warning message is sent to the management staff for prompting.

[0056] In another preferred embodiment of the present invention, in step S2, when the number of target models in the target ranking is less than a preset number threshold, a prompt message is sent to the management staff.

[0057] It is worth noting that if the number of target models is too small, it may indicate that insufficient data has been collected in this area during certain periods. When analyzing at this time, due to the lack of a sufficient data basis, there may be problems with a large deviation from the actual situation. Therefore, a prompt message needs to be sent for prompting to avoid safety issues.

[0058] In another preferred embodiment of the present invention, in step S3, the following steps are further included:

[0059] When the average tangent slope is less than a preset average tangent slope threshold and / or the minimum tangent slope is less than a preset minimum tangent slope threshold, it is determined that the corresponding curve is not an abnormal curve.

[0060] It can be understood that if the average tangent slope or the minimum tangent slope does not reach a certain threshold, it indicates that the overall change amplitude or mutation degree is small and is not sufficient to be recognized as abnormal. In some scenarios, there may be slight fluctuations in the environmental change itself (such as light changes, subtle vibrations, etc.). If frequent alarms are given for these normal fluctuations, it will lead to an increase in the workload and interference. By screening out those curves with insignificant average or local changes through this step, the system can focus resources or attention on the truly abnormal regions or time points.

[0061] In another preferred embodiment of the present invention, in step S3, if the average tangent slope of the curve is less than a preset average tangent slope threshold and / or the minimum tangent slope is less than a preset minimum tangent slope threshold, then in the next monitoring period adjacent to the current time, this sub-region is not monitored.

[0062] It should be noted that if it is confirmed that the change amplitude of the current sub-region is extremely small, it means that the value of continuing to monitor frequently is limited. By appropriately reducing the acquisition and analysis of this sub-region, the computational pressure and data processing cost can be reduced; leaving more monitoring resources, computing power, and attention to those sub-regions where abnormalities have occurred or are more likely to occur, enabling the system to conduct more frequent and in-depth monitoring in more critical or more significantly changing places; avoiding "over - monitoring" in regions with extremely small changes, reducing unnecessary data acquisition and model comparison, improving the overall work efficiency, and reducing manpower and time consumption.

[0063] The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be considered as used to limit the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the application of the present invention should still fall within the patent coverage scope of the present invention.

Claims

1. A 3D modeling method for foundation pit deformation monitoring, characterized in that: The following steps are involved: S1: Setting a monitoring area, gridding the monitoring area to obtain a sub-area, where the length and width of the sub-area are both preset values, collecting point cloud data of the sub-area, and performing 3D modeling based on the point cloud data to obtain a sub-model; S2: setting time nodes at preset time intervals within a preset monitoring period, obtaining sub-models corresponding to the same sub-region at the time nodes, recorded as target models, sorting the target models in the order of the time axis to obtain target sorting, and obtaining the similarity between the first target model in the target sorting and the target model at the i-th sorting position, recorded as Ki; The process of obtaining the target ranking also includes the following steps: The target model corresponding to the time node a is taken as the pending model, and an image of the sub-area is collected at the time node a to identify whether there is a preset physical target in the image. If so, the pending model is deleted from the target sorting to obtain a new target sorting; if not, the pending model is retained in the target sorting; S3: Generate a coordinate point (ti, Ki), where ti represents the time point of obtaining the target model of the i-th ranking position in the target ranking, fit the coordinate point, and obtain a fitting curve f(t), where t represents time; Obtaining the tangent slope of each point on the fitting curve, calculating the average tangent slope and the minimum tangent slope, performing weighted summation on the average tangent slope and the minimum tangent slope to obtain a judgment score P, and when the judgment score P is less than or equal to a preset judgment score threshold, treating the corresponding curve as an abnormal curve; The abnormal time point is determined based on the abnormal curve, and an early warning message is issued to remind a preset manager that an abnormality exists in the sub-area corresponding to the abnormal curve at the abnormal time point.

2. A 3D modeling method for foundation pit deformation monitoring according to claim 1, characterized in that: In step S3, the process of determining the abnormal time point specifically includes: A similarity threshold Kys is set, and the similarity threshold Kys is substituted into the functional relationship of the abnormal curve to obtain the abnormal time point.

3. A 3D modeling method for foundation pit deformation monitoring according to claim 1, characterized in that: In the step S2, the target entities include construction personnel and construction vehicles.

4. A 3D modeling method for foundation pit deformation monitoring according to claim 1, characterized in that: In the step S2, it is identified whether a target entity exists in the image based on a pre-trained target entity recognition model.

5. A 3D modeling method for foundation pit deformation monitoring according to claim 4, characterized in that: The process of training the target entity recognition model specifically includes: Establishing a database, wherein the database stores images with labeled labels, wherein the labels are the name of the target entity and 0, and the label 0 indicates that the target entity does not exist in the image; A target entity recognition model is established based on the deep learning model, and the target entity recognition model is trained and verified through the database to obtain a pre-trained target entity recognition model.

6. A 3D modeling method for foundation pit deformation monitoring according to claim 1, characterized in that: In the step S3, the judgment score P=η1*A1+η1*A2, η1 and η1 are the preset first weight and second weight, and 0<η1<η1, A1 and A2 represent the average tangent slope and the minimum tangent slope respectively.

7. A 3D modeling method for foundation pit deformation monitoring according to claim 2, characterized in that: The step S2 further includes the following steps: When Ki≤Kys, the subsequent steps are stopped and an early warning message is sent to the management personnel for prompting.

8. The 3D modeling method for foundation pit deformation monitoring according to claim 1, characterized in that: In the step S2, when the number of target models in the target sorting is less than a preset number threshold, a prompt message is sent to the management personnel.

9. The 3D modeling method for foundation pit deformation monitoring according to claim 1, characterized in that: The step S3 further includes the following steps: When the average tangent slope is less than a preset average tangent slope threshold and / or the minimum tangent slope is less than a preset minimum tangent slope threshold, it is determined that the corresponding curve is not an abnormal curve.

10. A 3D modeling method for foundation pit deformation monitoring according to claim 9, characterized in that: In step S3, if the average tangent slope of the curve is less than the preset average tangent slope threshold and / or the minimum tangent slope is less than the preset minimum tangent slope threshold, the sub-area will not be monitored in the next monitoring cycle adjacent to the current time.

Images