Slope deformation monitoring method based on image point cloud modeling

Abstract

The invention provides a slope deformation monitoring method based on image point cloud modeling, and the method comprises the steps: 1, obtaining an image; 2, carrying out the point cloud conversion, and generating point cloud data through a shot picture; and step 3, carrying out point cloud processing, wherein the point cloud processing comprises preprocessing, point cloud filtering and point cloud fine processing. Step 4, DEM generation: generating a DEM according to the obtained slope point cloud elevation data; and step 5, visualization processing: reading multi-stage DEM data and difference data, and carrying out visualization processing on the DEM data after difference. The portable device is used for shooting the image and processing the image to generate the high-precision point cloud data, so that efficient, real-time and low-cost slope deformation monitoring is realized, comprehensive deformation information and an intuitive visual chart are provided, meanwhile, an automatic early warning function is achieved, and the accuracy and timeliness of slope safety management are remarkably improved.

Description

Technical Field

[0001] The present application relates to the technical field of geological disaster monitoring, and in particular to a slope deformation monitoring method based on image point cloud modeling. Background Art

[0002] Geological disasters, especially slope instability and landslides, are one of the most common natural disasters in the world. In recent years, the frequency and intensity of geological disasters have increased. Slope instability and landslides may bring serious consequences. Therefore, it is particularly important to monitor slope deformation in a timely and accurate manner and predict and prevent geological disasters.

[0003] Traditional slope monitoring methods mainly include manual observation, leveling, GNSS positioning and strain gauge monitoring. Although these methods can meet the monitoring needs to a certain extent, they also have obvious limitations. Manual observation relies on on-site investigation and visual inspection, which is inefficient and greatly affected by human factors, making it difficult to ensure the consistency and accuracy of the data. Leveling is complex and requires professional personnel to perform, and the monitoring frequency is limited, making continuous monitoring impossible. Although GNSS positioning can provide high-precision location information, it is expensive, and in areas with dense vegetation or complex terrain, signal reception may be affected, resulting in incomplete data. Sensors such as strain gauges and crack meters can be installed directly on the slope to monitor local deformation, but the monitoring effect on large-area or overall deformation is limited.

[0004] With the continuous advancement of science and technology, the maturity of technologies such as drones, satellite remote sensing, photogrammetry and computer vision has provided new possibilities for the acquisition of high-resolution images and the generation of three-dimensional point cloud models. These new technologies not only improve the efficiency and accuracy of data collection, but also provide more comprehensive and dynamic information for slope deformation monitoring, providing an efficient, accurate and real-time monitoring method.

[0005] Therefore, how to combine the acquisition of high-resolution images and three-dimensional point cloud models to monitor large-area or overall deformation in a timely manner has become the current research focus. Summary of the invention

[0006] The present application provides a slope deformation monitoring method based on image point cloud modeling, which can be used to solve the technical problem of untimely monitoring of large-area or overall deformation.

[0007] The present application provides a slope deformation monitoring method based on image point cloud modeling, the method comprising:

[0008] Step 1, image acquisition: using a portable device, take multiple photos around the slope in a uniform light environment, with the overlap between each adjacent photo controlled at 60%-80%, to capture a panoramic image of the slope;

[0009] Step 2: Perform point cloud conversion to generate point cloud data from the captured photos;

[0010] Step 3: perform point cloud processing, which includes preprocessing, point cloud filtering, and point cloud refinement.

[0011] Step 4, DEM generation: Generate DEM based on the slope point cloud elevation data;

[0012] Step 5: Visualization processing: read multiple DEM data and difference data, and perform visualization processing on the differenced DEM data.

[0013] This application uses portable devices such as mobile phones and tablets to obtain slope images. This method does not require any sensors or equipment to be installed directly on the slope, reducing potential interference with the slope structure. Compared with traditional professional equipment such as laser scanners, ordinary photography equipment is less expensive and easy to popularize and apply, reducing monitoring costs. This method uses a depth estimation algorithm to provide very accurate depth information, so that the generated point cloud model can reflect the subtle features of the slope surface in detail, which helps to find small signs of deformation.

[0014] Compared with new technologies such as remote sensing and drones, mobile phones and cameras are very easy to carry. Although drones also have certain flexibility, their operation is restricted by factors such as airspace control and weather conditions, and requires professional flight permits and training. In terms of image quality, this method can collect close-range, high-resolution images of specific areas, which is especially suitable for small areas or complex terrain areas where it is difficult to obtain clear images through remote sensing or drones. Since the shooting angle and distance can be controlled, the color and texture information of the object surface can be better preserved, which is very important for analyzing the properties and surface characteristics of slope materials.

[0015] The generated image point cloud is processed by secondary filtering to more finely distinguish ground points and non-ground points, and obtain a more accurate DEM model that can better reflect the real terrain characteristics. Accurate DEM is essential for various terrain analyses, such as slope, aspect, watershed division, etc. High-quality DEM can provide more reliable data support and improve the accuracy of analysis results. Especially for those areas with a large amount of vegetation cover or other complex landforms, traditional single filtering may be difficult to effectively distinguish ground points, while secondary filtering can better cope with these challenges and ensure the quality of DEM under complex terrain. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a slope deformation monitoring method based on image point cloud modeling provided in an embodiment of the present application;

[0017] Figure 2An improved surface fitting filtering flow chart provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0019] The following first introduces the embodiments of the present application in conjunction with the accompanying drawings.

[0020] This application includes image acquisition, point cloud conversion, point cloud processing, DEM generation, and visualization processing. Figure 1 A schematic diagram of the method.

[0021] Step 1, obtain the image: use a mobile phone, tablet computer, camera or other portable device to take multiple photos around the slope in a uniform light environment, with the overlap of each adjacent photo controlled at 60%-80%, to capture a panoramic image of the slope.

[0022] Step 2: Perform point cloud conversion to generate point cloud data from the captured photos.

[0023] The photos taken are 2D images, which are first cleaned and standardized, including denoising, contrast enhancement, color balancing, etc.

[0024] Then the image is depth estimated. The depth estimation uses a 3D convolutional neural network, which takes the slope images taken from multiple perspectives as input, and extracts spatial and temporal features through multiple 3D convolutional layers and activation functions. In the deep layer of the network, the extracted features are fused and the spatial dimension is restored through the pooling layer and upsampling layer; in the last layer of the network, the fused feature map is mapped to a depth map with the same spatial dimension as the input image. The depth map contains the depth value of each pixel location;

[0025] According to the camera internal parameter matrix K, each pixel (u,v) in the depth map is normalized and calculated to obtain the coordinates (x',y'), and the depth value is used to convert the normalized coordinates into points (X,Y,Z) in the world coordinate system; the 3D coordinates corresponding to all valid depth values ​​are stored as point cloud data in LAS format, and color information is added, that is, the color value of the corresponding pixel in the original 2D image; the generated LAS file containing 3D point cloud data contains X, Y, Z coordinates and RGB color information for each point.

[0026] Step 3: perform point cloud processing, which includes preprocessing, point cloud filtering, and point cloud refinement.

[0027] Step 31, preprocessing, that is, denoising the point cloud data. Since the noise points are generally small and relatively isolated in the elevation space distribution, this paper combines the elevation distribution histogram and the KD tree to denoise the point cloud data.

[0028] Step 32: point cloud filtering includes primary filtering and secondary filtering.

[0029] Step 321, using a triangulated progressive encryption algorithm to perform initial filtering on the denoised point cloud, gridding the pre-processed point cloud data, taking the lowest elevation point in each grid as a ground seed point, and constructing an initial TIN model with the seed point;

[0030] Traverse all the points to be judged in the area, make threshold judgments on distance and angle, incorporate those that meet the conditions into the initial TIN model, and update the TIN model. Perform iterative judgments on the remaining points and update the TIN model to initially obtain ground points.

[0031] Since the progressive triangulation encryption in this step is to provide a certain number of ground seed points for the next step of secondary filtering, and considering that the progressive triangulation encryption algorithm is time-consuming, two iterative calculations are performed and the obtained ground point set is used as the seed point for the next step of secondary filtering;

[0032] Step 322, secondary filtering uses an improved surface fitting algorithm to distinguish ground points and non-ground points by judging the height difference between the elevation fitting value and the actual elevation value of the corresponding plane point. The height difference of the ground point is smaller, while the height difference of the non-ground point is larger. Figure 2 This is the flow chart of the method, and the specific steps are:

[0033] Step 3221, a grid is established based on the ground points obtained by the progressive triangulation encryption algorithm. In order to meet the condition that there are at least 6 ground points in each grid required by the surface fitting, the side length of the established grid must be greater than or equal to n, and n is determined by the following method:

[0034]

[0035] Where: N is the total number of ground seed points; S is the survey area;

[0036] Step 3222, judging the number of ground points in each grid in order, if the number of ground points in the grid is greater than or equal to 6, proceeding to step 3224, otherwise proceeding to step 3223;

[0037] Step 3223: When the number of ground points in a grid is less than 6, merge the ground points in the next grid into the current grid until the number of ground points is greater than or equal to 6; if the number of ground points is still less than 6 when merging to the last grid in a row, then merge the ground points in all subsequent grids into the previous grid block that meets the surface fitting conditions, and re-perform surface fitting;

[0038] Step 3224, performing least square surface fitting using the ground points in the grid block to calculate quadratic surface parameters;

[0039] Step 3225, traverse all points to be determined and calculate the fitting elevation difference of each point;

[0040] Step 3226, mark the points whose fitted elevation difference is less than the set threshold as ground points and include them in the ground point set.

[0041] Step 33, point cloud refinement processing: After point cloud denoising and point cloud filtering, the original point cloud data is separated into two parts: a ground point set and a non-ground point set. For the ground point set, it also contains a small number of low vegetation points. The height variance σ2 of each point and the points in the K neighborhood is traversed and calculated. The height variance value is higher in the vegetation and smaller at the top of the ground. By setting the height variance threshold, the low vegetation points mixed in the ground point set are further eliminated to obtain an accurate slope point cloud data set.

[0042] Step 4, DEM generation: Generate DEM based on the obtained slope point cloud elevation data.

[0043] The slope point cloud data is used to construct a triangulated network, and natural neighbor interpolation is used on the constructed triangulated network to generate regular raster data to obtain the slope DEM data.

[0044] Multi-period DEM data of the slope at different time points are obtained, the coordinate system of each period of DEM data is unified, and professional GIS software is used to perform difference calculations on the multi-period DEM data to obtain the elevation change of the slope within the time series.

[0045] Step 5: Visualization processing: read multiple DEM data and difference data, and perform visualization processing on the differenced DEM data.

[0046] The elevation change map is generated using the difference data to visually display the elevation change of each location in a color gradient manner; based on the elevation change map, a deformation distribution map is further generated to display the overall deformation trend and local deformation characteristics of the slope; the elevation change rate is calculated based on the elevation change value and the time interval between two DEM acquisitions;

[0047] Use the rate of change data to generate a rate of change map, which shows the rate of change of each part of the slope and helps identify areas of rapid deformation;

[0048] Combined with the elevation change map and the change rate map, a threshold is set to determine the potential danger zone with poor slope stability. If the threshold is exceeded, an alarm will be triggered. The alarm method is a prompt on the software system interface and a text message sent to the manager's mobile phone.

[0049] To implement the above-mentioned visualization processing steps, the present invention also proposes a slope monitoring and early warning platform, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps corresponding to the above-mentioned method when executing the computer program.

[0050] In potential danger areas, formulate targeted monitoring plans, such as increasing the monitoring frequency and adopting more accurate monitoring technology, so as to better understand the stability of the slope.

[0051] The above-described embodiments of the present application do not constitute a limitation on the protection scope of the present application.

Claims

1. A slope deformation monitoring method based on image point cloud modeling, characterized in that: The method comprises: Step 1, image acquisition: using a portable device, take multiple photos around the slope in a uniform light environment, with the overlap between each adjacent photo controlled at 60%-80%, to capture a panoramic image of the slope; Step 2: Perform point cloud conversion to generate point cloud data from the captured photos; Step 3: perform point cloud processing, which includes preprocessing, point cloud filtering, and point cloud refinement processing; Step 4, DEM generation: Generate DEM based on the obtained slope point cloud elevation data; Step 5: Visualization processing: read multiple DEM data and difference data, and perform visualization processing on the differenced DEM data.

2. The method according to claim 1, characterized in that Step 2: Perform point cloud conversion to generate point cloud data from the captured photos, including: Clean and normalize photos, including denoising, contrast enhancement, and color balancing; Then the image is depth estimated. The depth estimation uses a 3D convolutional neural network, which takes the slope images taken from multiple perspectives as input, and extracts spatial and temporal features through multiple 3D convolutional layers and activation functions. In the deep layer of the network, the extracted features are fused and the spatial dimension is restored through the pooling layer and upsampling layer; in the last layer of the network, the fused feature map is mapped to a depth map with the same spatial dimension as the input image. The depth map contains the depth value of each pixel location; According to the camera internal parameter matrix K, each pixel (u,v) in the depth map is normalized and calculated to obtain the coordinates (x',y'), and the depth value is used to convert the normalized coordinates into points (X,Y,Z) in the world coordinate system; the 3D coordinates corresponding to all valid depth values ​​are stored as point cloud data in LAS format, and color information is added, that is, the color value of the corresponding pixel in the original 2D image; the generated LAS file containing 3D point cloud data contains X, Y, Z coordinates and RGB color information for each point.

3. The method according to claim 1, characterized in that: Step 3: perform point cloud processing, which includes preprocessing, point cloud filtering, and point cloud refinement, including: Step 31, preprocessing, i.e. denoising the point cloud data; denoising the point cloud data by combining the elevation distribution histogram and the KD tree; Step 32, point cloud filtering includes primary filtering and secondary filtering; wherein, the secondary filtering adopts an improved surface fitting algorithm to distinguish ground points and non-ground points by judging the height difference between the elevation fitting value and the actual elevation value of the corresponding plane point. The height difference of the ground point is smaller, while the height difference of the non-ground point is larger; Step 33, point cloud refinement processing: After point cloud denoising and point cloud filtering, the original point cloud data is separated into two parts: a ground point set and a non-ground point set. For the ground point set, it also contains a small number of low vegetation points. The height variance σ2 of each point and the points in the K neighborhood is traversed and calculated. The height variance value is higher in the vegetation and smaller at the top of the ground. By setting the height variance threshold, the low vegetation points mixed in the ground point set are further eliminated to obtain an accurate slope point cloud data set.

4. The method according to claim 3, characterized in that The initial filtering includes: The denoised point cloud is initially filtered using a triangulated progressive encryption algorithm, and the pre-processed point cloud data is gridded. The lowest elevation point in each grid is used as the ground seed point, and the initial TIN model is constructed using the seed point. Traverse all the points to be judged in the area, make threshold judgments on distance and angle, incorporate those that meet the conditions into the initial TIN model, and update the TIN model. Perform iterative judgments on the remaining points and update the TIN model to initially obtain ground points. Since the progressive triangulation encryption in this step is to provide a certain number of ground seed points for the next step of secondary filtering, and considering that the progressive triangulation encryption algorithm is time-consuming, two iterative calculations are performed and the obtained ground point set is used as the seed points for the next step of secondary filtering.

5. The method according to claim 3, characterized in that: Secondary filtering includes: Step 3221, a grid is established based on the ground points obtained by the progressive triangulation encryption algorithm. In order to meet the condition that there are at least 6 ground points in each grid required by the surface fitting, the side length of the established grid must be greater than or equal to n, and n is determined by the following method: Where: N is the total number of ground seed points; S is the survey area; Step 3222, judging the number of ground points in each grid in order, if the number of ground points in the grid is greater than or equal to 6, proceeding to step 3224, otherwise proceeding to step 3223; Step 3223: When the number of ground points in a grid is less than 6, merge the ground points in the next grid into the current grid until the number of ground points is greater than or equal to 6; if the number of ground points is still less than 6 when merging to the last grid in a row, then merge the ground points in all subsequent grids into the previous grid block that meets the surface fitting conditions, and re-perform surface fitting; Step 3224, performing least square surface fitting using the ground points in the grid block to calculate quadratic surface parameters; Step 3225, traverse all points to be determined and calculate the fitting elevation difference of each point; Step 3226, mark the points whose fitted elevation difference is less than the set threshold as ground points and include them in the ground point set.

6. The method according to claim 1, characterized in that Step 4, DEM generation: Generate DEM based on the slope point cloud elevation data, including: Use the slope point cloud data to construct a triangulated network. Use natural neighbor interpolation on the constructed triangulated network to generate regular raster data and obtain the slope DEM data. The multi-period DEM data of the slope at different time points are obtained, the coordinate system of each period of DEM data is unified, and the multi-period DEM data are subjected to difference calculation to obtain the elevation change of the slope within the time series.

7. The method according to claim 1, characterized in that Step 5: Visualization processing: read multiple DEM data and difference data, and perform visualization processing on the differenced DEM data, including: Use the difference data to generate an elevation change map, which visually displays the elevation changes at each location in a color gradient manner; Based on the elevation change map, a deformation distribution map is further generated to show the overall deformation trend and local deformation characteristics of the slope; the elevation change rate is calculated based on the elevation change value and the time interval between two DEM acquisitions; Use the rate of change data to generate a rate of change graph, which shows the rate of change of each part of the slope; Combined with the elevation change map and the change rate map, a threshold is set to determine the potential danger zone with poor slope stability. If the threshold is exceeded, an alarm is triggered.

Images