Landslide Monitoring Method and System Based on Hybrid Three-Dimensional Coordinate Solution

Abstract

This invention provides a landslide monitoring method and system based on hybrid three-dimensional coordinate calculation, relating to the field of landslide monitoring. The method accurately calculates the basic three-dimensional coordinates and local residual values ​​of the markers by means of the coordinate mapping relationship between the camera and the markers in the landslide monitoring area, and obtains the landslide monitoring results of the markers in real time by superimposing the basic three-dimensional coordinates and local residual values. This achieves high-precision calculation of the markers from image data to three-dimensional world coordinates, and greatly improves the measurement accuracy of close-range photography.

Description

Technical Field

[0001] This invention relates to the field of landslide monitoring, and in particular to a landslide monitoring method and system based on hybrid three-dimensional coordinate calculation. Background Technology

[0002] The main basis for landslide monitoring is the surface deformation of the landslide monitoring area, but existing landslide monitoring methods each have their limitations: While high-precision positioning of landslide monitoring areas based on satellite positioning systems can achieve high-precision landslide monitoring, it is costly and difficult to achieve high-density monitoring of target areas. Although spaceborne synthetic aperture radar interferometry can also achieve landslide monitoring, it is limited by the revisit cycle, has insufficient temporal resolution, and is also costly.

[0003] To address the issue of high costs, close-range photogrammetry, using cameras to measure landslide monitoring areas, has become a widely adopted landslide monitoring solution. This measurement process primarily relies on the parallax triangulation principle of binocular stereo vision. However, in areas with significant natural terrain undulations, the combined effects of observation distance limitations and camera resolution mean that relying solely on two frames acquired at a single moment and mapping their two-dimensional coordinates to three-dimensional spatial coordinates inevitably leads to inherent errors in displacement calculations. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a landslide monitoring method and system based on hybrid three-dimensional coordinate calculation. This method accurately calculates the basic three-dimensional coordinates and local residual values ​​of the markers by means of the coordinate mapping relationship between the camera and the markers in the landslide monitoring area, and obtains the landslide monitoring results of the markers in real time by superimposing the basic three-dimensional coordinates and local residual values. This achieves high-precision calculation of the markers from image data to three-dimensional world coordinates, and greatly improves the measurement accuracy of close-range photography.

[0005] In a first aspect, embodiments of the present invention provide a landslide monitoring method based on hybrid three-dimensional coordinate calculation, the method comprising: Once the marker in the landslide monitoring area is detected to have moved to the target location, the binocular four-dimensional pixel coordinates of the marker are collected, and the three-dimensional world coordinates of the marker are obtained. The feature mapping results of the marker are calculated using second-order polynomial regression. Based on the target position coordinates of the marker in the world coordinate system, the feature mapping results are fitted to the target position coordinates of the marker using the least squares method. Then, the regression transformation data and bias transformation data between the binocular four-dimensional pixel coordinates and the three-dimensional world coordinates are calculated iteratively. The regression coefficient matrix of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates is generated by the regression transformation data, and the bias vector of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates is generated by the bias transformation data. After performing Gaussian process regression calculations on the binocular four-dimensional pixel coordinates and target position coordinates using preset variance scale kernels and white noise variance kernels, residual values ​​are obtained. The nonlinear mapping data when the binocular four-dimensional pixel coordinates are mapped to the target position coordinates is obtained using the residual values. Preset Matrn and Kronecker functions are used to calculate the positional variation data and random variation data of the nonlinear mapping data under the residual values, respectively. The covariance value of the nonlinear mapping data is calculated by multiplying the positional variation data with the variance scale kernel, and the random variance value of the nonlinear mapping data is calculated by multiplying the white noise variance kernel with the random variation data. The basic three-dimensional coordinates of the marker are calculated using the regression coefficient matrix and the bias vector, and the local residual values ​​of the marker are calculated using the covariance and random variance values. The landslide monitoring results of the markers in the landslide monitoring area are determined by superimposing the basic three-dimensional coordinates and local residual values.

[0006] Optionally, after detecting that a marker in the landslide monitoring area has moved to the target location, the steps of acquiring the marker's binocular four-dimensional pixel coordinates and obtaining the marker's three-dimensional world coordinates include: Identify the binocular cameras installed in the foundation area of ​​the landslide monitoring area, and identify the markers deployed in the slope area of ​​the landslide monitoring area; The direction of movement of the marker and the target location are determined based on the slope direction in the sloping area; After moving the marker from the starting position to the target position according to the direction of movement, control the binocular camera to acquire first-eye and second-eye images of the marker in different directions; The binocular four-dimensional pixel coordinates of the marker are determined by using the first column and first row coordinates of the marker region in the first eye image and the second column and second row coordinates of the marker region in the second eye image. Using the coordinates of the starting position in the world coordinate system as a reference, obtain the three-dimensional world coordinates of the marker from the starting position to the target position.

[0007] Optionally, the step of obtaining the three-dimensional world coordinates of the marker from the starting position to the target position, using the coordinates of the starting position in the world coordinate system as a reference, includes: The differential positioning equipment installed in the foundation area of ​​the landslide monitoring area is used to determine the world coordinate system of the markers; the starting position is located at the origin of the world coordinate system. Once the marker is detected to have moved from its starting position to its target position, the differential positioning device is controlled to continuously collect the marker's coordinate positioning data multiple times according to the preset number of data collections. Calculate the average coordinate value of the coordinate positioning data under the acquisition quantity, and obtain the three-dimensional world coordinates of the marker at the target location by the coordinate difference between the average coordinate value and the origin.

[0008] Optionally, the underlying three-dimensional coordinates of the marker can be calculated using the regression coefficient matrix and the bias vector, including: After performing a second-order polynomial expansion on the real-time acquired binocular four-dimensional pixel coordinates, we obtain the first-order feature data, second-order feature data, and cross-feature data of the binocular four-dimensional pixel coordinates. Real-time feature mapping data of markers is determined based on first-order feature data, second-order feature data, and cross-feature data; Calculate the vector product of the transpose of the regression coefficient matrix and the real-time feature mapping data, and calculate the basic three-dimensional coordinates of the marker based on the vector product result and the vector summation result of the bias vector.

[0009] Optionally, the local residuals of the marker can be calculated using the covariance and random variance values, including: After performing Z-score processing on the real-time acquired binocular four-dimensional pixel coordinates, dimensionless coordinate data of the binocular four-dimensional pixel coordinates is obtained, and a sample vector of the marker is constructed based on the dimensionless coordinate data. The covariance value is used to calculate the position change value corresponding to the sample vector, and the random variance value is used to calculate the random noise value corresponding to the sample vector. The local residual value of the marker is calculated by superimposing the position change value and the random noise value.

[0010] Optionally, the step of obtaining the landslide monitoring results of the markers in the landslide monitoring area through the superposition of the basic three-dimensional coordinates and local residual values ​​includes: The real-time position coordinates of the marker in three-dimensional world coordinates are determined based on the superposition of the basic three-dimensional coordinates and the local residual values. The landslide monitoring results of the markers in the landslide monitoring area are determined based on the changes in their real-time location coordinates within the landslide monitoring area.

[0011] Secondly, this invention provides a landslide monitoring system based on hybrid three-dimensional coordinate calculation, the system comprising: World coordinate acquisition module: used to collect the binocular four-dimensional pixel coordinates of the marker and obtain the three-dimensional world coordinates of the marker after it is detected that the marker in the landslide monitoring area has moved to the target position; The first coordinate mapping module is used to calculate the feature mapping results of the markers through second-order polynomial regression. Based on the target position coordinates of the world coordinate system where the markers are located, the module uses the least squares method to fit the feature mapping results to the target position coordinates of the markers. Then, it iteratively calculates the regression transformation data and bias transformation data between the binocular four-dimensional pixel coordinates and the three-dimensional world coordinates. The module generates a regression coefficient matrix for the transformation of binocular four-dimensional pixel coordinates to three-dimensional world coordinates through the regression transformation data, and generates a bias vector for the transformation of binocular four-dimensional pixel coordinates to three-dimensional world coordinates through the bias transformation data. The second coordinate mapping module is used to perform Gaussian process regression calculations on the binocular four-dimensional pixel coordinates and target position coordinates using preset variance scale kernels and white noise variance kernels to obtain residual values; it uses the residual values ​​to obtain nonlinear mapping data when the binocular four-dimensional pixel coordinates are mapped to the target position coordinates, and uses preset Matrn and Kronecker functions to calculate the position variation data and random variation data of the nonlinear mapping data under the residual values, respectively; it calculates the covariance value of the nonlinear mapping data by multiplying the position variation data with the variance scale kernel, and calculates the random variance value of the nonlinear mapping data by multiplying the white noise variance kernel with the random variation data; Coordinate prediction processing module: used to calculate the basic three-dimensional coordinates of the marker using the regression coefficient matrix and bias vector, and to calculate the local residual values ​​of the marker using the covariance and random variance values; Landslide monitoring execution module: used to determine the landslide monitoring results of markers in the landslide monitoring area by superimposing the basic three-dimensional coordinates and local residual values.

[0012] Thirdly, embodiments of the present invention also provide an electronic device, which includes a processor and a memory. The memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the steps of the landslide monitoring method based on hybrid three-dimensional coordinate calculation provided in the first aspect.

[0013] This invention provides a landslide monitoring method and system based on hybrid three-dimensional coordinate calculation. During landslide monitoring in a landslide monitoring area, when a marker in the monitoring area is detected to have moved to a target location, the binocular four-dimensional pixel coordinates of the marker are acquired, and the three-dimensional world coordinates of the marker are obtained. Then, using the target location coordinates in the world coordinate system of the marker as a reference, a regression coefficient matrix and bias vector mapping the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates are calculated using second-order polynomial regression. Furthermore, the covariance and random variance values ​​mapping the binocular four-dimensional pixel coordinates to the target location coordinates are calculated using Gaussian process regression. Subsequently, the basic three-dimensional coordinates of the marker are calculated using the regression coefficient matrix and bias vector, and the local residual values ​​of the marker are calculated using the covariance and random variance values. Finally, the landslide monitoring result of the marker in the landslide monitoring area is determined by the superposition of the basic three-dimensional coordinates and the local residual values. This method accurately calculates the basic three-dimensional coordinates and local residual values ​​of the markers by mapping the coordinates between the camera and the markers in the landslide monitoring area. It then obtains the landslide monitoring results of the markers in real time by superimposing the basic three-dimensional coordinates and local residual values. This achieves high-precision calculation of the markers from image data to three-dimensional world coordinates, which greatly improves the measurement accuracy of close-range photography.

[0014] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a landslide monitoring method based on hybrid three-dimensional coordinate calculation, provided as an embodiment of the present invention; Figure 2 This is a distribution map of markers in a landslide monitoring method based on hybrid three-dimensional coordinate calculation provided in an embodiment of the present invention; Figure 3A box plot of coordinate calculation error distribution in a landslide monitoring method based on hybrid three-dimensional coordinate calculation provided in an embodiment of the present invention; Figure 4 A box plot of coordinate positioning error distribution in a landslide monitoring method based on hybrid three-dimensional coordinate solution provided in an embodiment of the present invention; Figure 5 A comparison diagram of the spatial displacement trajectories of markers in a landslide monitoring method based on hybrid three-dimensional coordinate calculation, provided in an embodiment of the present invention; Figure 6 A schematic diagram of a landslide monitoring system based on hybrid three-dimensional coordinate calculation is provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0018] icon: 100 - World coordinate acquisition module; 200 - First coordinate mapping module; 300 - Second coordinate mapping module; 400 - Coordinate prediction and processing module; 500 - Landslide monitoring execution module; 101 - Processor; 102 - Memory; 103 - Bus; 104 - Communication interface. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Traditional close-range photogrammetry primarily relies on the parallax triangulation principle of binocular stereo vision. In field landslide measurements, the physical baseline length of the camera is limited by surrounding deployment conditions and cannot be extended proportionally with increasing monitoring distance. According to the depth calculation principle of stereo vision, the absolute error in the depth direction is directly proportional to the square of the observation distance and inversely proportional to the baseline length. As the observation distance extends, relying solely on the spatial triangulation of two frames instantaneously acquired by two cameras leads to a rapid amplification of depth calculation errors. To avoid the inherent systematic errors in binocular stereo vision, existing technologies often employ a strategy of measuring displacement separately with multiple monocular cameras and taking the arithmetic mean. This approach typically uses static background feature extraction and matching to correct camera extrinsic drift, calculating the spatial displacement of each marker separately and then averaging the results to reduce single-view observation bias. However, in areas with significant natural terrain undulations, the combined effects of observation distance and camera resolution mean that this monocular multi-view averaging strategy still struggles to avoid displacement calculation errors.

[0021] To address the aforementioned problems, a landslide monitoring method based on hybrid three-dimensional coordinate calculation, as disclosed in the embodiments of the present invention, is described below. The method is as follows: Figure 1 As shown, it includes: Step S101: After detecting that the marker in the landslide monitoring area has moved to the target position, collect the binocular four-dimensional pixel coordinates of the marker and obtain the three-dimensional world coordinates of the marker.

[0022] Monitoring markers were deployed on the surface of the landslide monitoring area, and binocular cameras were deployed on stable foundations outside the landslide area, ensuring that the cameras were relatively stationary relative to the monitoring area. Controlled local movement of the markers was then performed; once the markers reached the target location, the binocular close-range photography acquisition equipment was activated, simultaneously completing the acquisition of two sets of core data.

[0023] First, the binocular four-dimensional pixel coordinates corresponding to the target marker are acquired with high precision. These four-dimensional pixel coordinates integrate the pixel coordinates of the two-dimensional images acquired by the left and right cameras and the temporal parameters of the corresponding acquisition time, forming complete pixel data containing spatial and temporal dimensions. Secondly, relying on the unified world coordinate system pre-established in the monitoring area, and combining the camera's intrinsic and extrinsic parameter calibration results and equipment deployment location parameters, the original three-dimensional world coordinates corresponding to the current target position of the marker are initially calculated and obtained. This step provides complete and original basic monitoring data for subsequent coordinate model fitting, error calculation, and coordinate correction, and is a data prerequisite for achieving high-precision calculation.

[0024] Step S102: Calculate the feature mapping result of the marker through second-order polynomial regression; using the target position coordinates of the world coordinate system where the marker is located as the reference, fit the feature mapping result to the target position coordinates of the marker using the least squares method, and then iteratively calculate the regression transformation data and bias transformation data between the binocular four-dimensional pixel coordinates and the three-dimensional world coordinates; generate the regression coefficient matrix of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates through the regression transformation data, and generate the bias vector of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates through the bias transformation data.

[0025] Step S103: After performing Gaussian process regression calculation on the binocular four-dimensional pixel coordinates and target position coordinates using a preset variance scale kernel and white noise variance kernel, the residual values ​​are obtained; the nonlinear mapping data when the binocular four-dimensional pixel coordinates are mapped to the target position coordinates is obtained using the residual values, and the position change data and random change data of the nonlinear mapping data under the residual values ​​are calculated using the preset Matern function and Kronecker function respectively; the covariance value of the nonlinear mapping data is calculated by multiplying the position change data with the variance scale kernel, and the random variance value of the nonlinear mapping data is calculated by multiplying the white noise variance kernel with the random change data.

[0026] Steps S102 and S103 use the target position coordinates of the world coordinate system where the marker is located as a reference, and calculate the regression coefficient matrix and bias vector of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates through second-order polynomial regression. They also calculate the covariance and random variance of the binocular four-dimensional pixel coordinates to the target position coordinates through Gaussian process regression.

[0027] Specifically, using the three-dimensional coordinates of the target position of the marker in the world coordinate system as the baseline truth, the model parameters and error parameters are solved in two ways: The first path corresponds to step S102, which uses a second-order polynomial regression algorithm to perform polynomial feature expansion on the binocular four-dimensional pixel coordinates and combines the least squares method to complete the fitting. Finally, the regression coefficient matrix and bias vector required to map the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates are calculated, and a global basic mapping model from pixel coordinates to spatial coordinates is built. The second path corresponds to step S103, which uses the Gaussian process regression algorithm and combines the Matern function and Kronecker function to construct the covariance function and random variance function respectively. Based on this, the covariance value and random variance value corresponding to the binocular four-dimensional pixel coordinate mapping process are calculated, which respectively characterize the spatial correlation systematic error and discrete random observation error of the coordinate mapping, and complete the parameter calibration of the local error model.

[0028] Step S104: Calculate the basic three-dimensional coordinates of the marker using the regression coefficient matrix and bias vector, and calculate the local residual values ​​of the marker using the covariance and random variance values.

[0029] The regression coefficient matrix and bias vector obtained in step S103 are used to perform feature transformation, matrix operations, and bias compensation on the real-time acquired binocular four-dimensional pixel coordinates to calculate the basic three-dimensional coordinates of the marker. These coordinates are the preliminary spatial position results output by the global polynomial mapping model, which initially avoids the problem of amplified depth calculation errors caused by the inability of the physical baseline to adapt and extend with the observation distance in traditional binocular stereo vision, and eliminates most of the structural systematic errors.

[0030] Simultaneously, by utilizing the obtained covariance and random variance values, the systematic positional deviation in the coordinate mapping and the random noise brought by the environment and equipment are separated, and the local residual value of the marker is obtained through comprehensive calculation. This residual value is used to quantify various residual errors in the basic three-dimensional coordinates that have not been eliminated by the global model, accurately quantify the local inherent displacement calculation errors caused by factors such as complex terrain, long-distance observation, and insufficient equipment resolution, and realize the refined and dynamic quantification of monitoring errors.

[0031] Step S105: Determine the landslide monitoring results of the markers in the landslide monitoring area by superimposing the basic three-dimensional coordinates and local residual values.

[0032] The basic three-dimensional coordinates of the marker obtained in step S104 are vector-superimposed with the real-time updated local residual values, and error compensation is performed. The local residual values ​​are used to comprehensively correct the inherent and random errors remaining in the basic three-dimensional coordinates, completely overcoming the displacement calculation deficiencies of traditional close-range photogrammetry schemes under complex field conditions. After error superposition and correction, real-time, high-precision three-dimensional spatial displacement coordinate data of the marker is obtained. Based on this accurate three-dimensional deformation data, the core monitoring parameters such as the displacement, deformation direction, and deformation rate of the marker within the landslide monitoring area can be analyzed in real time. Finally, complete and accurate landslide monitoring results are output, significantly improving the overall measurement accuracy and environmental adaptability of the close-range photogrammetric landslide monitoring scheme.

[0033] The distribution of markers in the above-mentioned landslide monitoring method based on hybrid three-dimensional coordinate solution is as follows: Figure 2 As shown, its spatial deployment mainly follows a strategy of denser coverage in active areas and sparse coverage in stable edge areas. Based on this, when a marker in the landslide monitoring area is detected to have moved to the target location, step S101, which involves acquiring the marker's binocular four-dimensional pixel coordinates and obtaining its three-dimensional world coordinates, includes the following steps: Step S201: Determine the binocular cameras installed in the foundation area of ​​the landslide monitoring area, and determine the markers deployed in the slope area of ​​the landslide monitoring area.

[0034] Beforehand, the binocular camera equipment is fixedly deployed in a stable foundation area within the landslide monitoring zone. The foundation area has undergone flatness and stability testing, ensuring no risk of landslide deformation. This guarantees that the camera's deployment position, shooting angle, and equipment posture remain fixed throughout the monitoring process, avoiding monitoring errors caused by equipment displacement. After deployment, the fixed-installation binocular synchronous imaging cameras are activated. Simultaneously, information on all deployed monitoring markers within the landslide monitoring slope area is comprehensively collected. Strictly adhering to the aforementioned differentiated deployment rules, all marker location data from active dense areas and stable sparse areas are compiled. The initial deployment location, number, and corresponding monitoring responsibilities of each marker are clearly defined, providing complete basic data for subsequent displacement monitoring, image acquisition, and coordinate calculation.

[0035] Step S202: Determine the direction of movement of the marker and the target location based on the slope direction in the slope area.

[0036] By combining the overall topographic parameters of the landslide monitoring slope area, core topographic feature data such as slope gradient, slope direction, and stratum slippage trend are accurately extracted. Based on the deformation law of natural landslides, the displacement direction of the landslide soil always conforms to the slope direction. Therefore, based on the measured slope topographic parameters, the natural displacement direction of each monitoring marker can be predicted and determined, and the target monitoring position after the marker displacement can be preset according to monitoring needs. This method closely matches the real deformation characteristics of landslides in the field, effectively simulates the natural displacement state of the landslide soil, ensures the authenticity of subsequent data collection and the validity of monitoring results, and avoids the problem of data distortion caused by randomly setting displacement positions.

[0037] Step S203: After moving the marker from the starting position to the target position according to the direction of movement, control the binocular camera to acquire the first and second eye images of the marker in different directions.

[0038] Using the movement direction determined in step S202 as the displacement reference, the markers within the landslide monitoring slope area are controlled to move smoothly from their initial placement position along a direction conforming to the natural deformation law of the landslide to the preset target monitoring position. Once the marker displacement is complete and the posture is stable without shaking, the binocular cameras fixed at the foundation are immediately triggered to simultaneously capture images, obtaining first-eye and second-eye images of the corresponding markers in different directions. A synchronous frame acquisition mode is used throughout the process to ensure that the acquisition sequence of the two images is completely consistent, eliminating errors in marker position shift caused by shooting time differences, accurately capturing complete imaging information of the markers at the target position, and providing a high-quality image data source for subsequent pixel coordinate extraction.

[0039] Step S204: Using the first column coordinates and first row coordinates of the marker region in the first visual image and the second column coordinates and second row coordinates of the marker region in the second visual image, determine the binocular four-dimensional pixel coordinates of the marker.

[0040] For the first and second view images acquired by the binocular camera, a sub-pixel-level feature extraction algorithm is used to accurately locate the complete imaging area of ​​the marker in both images, eliminating interference factors such as light and shadow, weeds, and terrain occlusion in the field environment. The column coordinates and row coordinates corresponding to the imaging area of ​​the marker in the first view image are extracted respectively, and denoted as the first column coordinates. coordinates of the first row Simultaneously extract the column and row coordinates corresponding to the imaging region of the same marker in the second image, and record them as the second column coordinates. With the second row coordinates By integrating and combining the row and column two-dimensional pixel coordinates of the two sets of images above, a binocular four-dimensional pixel coordinate system containing four-dimensional parameters for the marker is constructed. It can completely retain the original pixel position information of the marker in the binocular vision imaging system and realize the structured transformation of image data.

[0041] Step S205: Using the coordinates of the starting position in the world coordinate system as a reference, obtain the three-dimensional world coordinates of the marker from the starting position to the target position.

[0042] A unified world coordinate system for the landslide monitoring area is established beforehand, using the initial location of the markers as the reference origin. Coordinate system calibration and parameter adjustment are performed to ensure a unified and unbiased coordinate reference across the entire area. Based on this standardized world coordinate system, and combining the initial reference coordinates of the markers with the target location information after the markers' displacement in step S203, a spatial coordinate calculation algorithm is used to accurately calculate and obtain the real-time three-dimensional world coordinates corresponding to the markers' movement from the initial location to the target location. This coordinate system contains precise location parameters in the three spatial dimensions of X, Y, and Z, which can fully characterize the spatial displacement state of the marker and provide accurate three-dimensional coordinate reference data for subsequent regression matrix calculation, residual value solving, and high-precision landslide deformation result calculation.

[0043] Optionally, step S205, which uses the coordinates of the starting position in the world coordinate system as a reference to obtain the three-dimensional world coordinates of the marker from the starting position to the target position, includes the following steps: Step S301: Obtain the differential positioning device set up in the foundation area of ​​the landslide monitoring area, and use the coordinate system of the differential positioning device to determine the world coordinate system of the marker; wherein, the starting position is located at the origin of the world coordinate system.

[0044] RTK (Real-Time Kinematic) differential positioning equipment was fixedly deployed in the landslide monitoring area in a structurally stable foundation area without deformation risk. This equipment can achieve high-precision spatial positioning and coordinate system calibration, unaffected by landslide soil deformation or slope displacement, maintaining the stability of the equipment position and positioning benchmark throughout the process. Based on the high-precision positioning benchmark of the differential positioning equipment, a dedicated world coordinate system adapted to this landslide monitoring scenario was constructed, unifying the spatial coordinate measurement standard across the entire area and avoiding the problems of poor adaptability and large coordinate transformation deviations of general coordinate systems. At the same time, the initial starting position of the marker without displacement was marked as the origin of this world coordinate system, serving as the reference zero point for all displacement monitoring and coordinate calculations, simplifying the subsequent displacement difference calculation logic and achieving relatively accurate measurement of marker displacement data.

[0045] Step S302: After detecting that the marker has moved from the starting position to the target position, control the differential positioning device to continuously collect the coordinate positioning data of the marker multiple times according to the preset number of data collections.

[0046] The displacement status of the marker is monitored in real time. Once the marker is detected to have moved smoothly from its initial position to the preset target position, and its posture is stable without shaking or instantaneous displacement fluctuations, the differential positioning device is activated to perform batch data acquisition. A fixed number of data acquisitions is pre-set based on the accuracy requirements of the field monitoring environment and the equipment's error characteristics. The differential positioning device is controlled to continuously and uninterruptedly acquire multi-point coordinate positioning data of the marker at the current target location according to the preset frequency and acquisition quantity, obtaining multiple sets of raw coordinate positioning data. By using multiple batches of continuous sampling (e.g., five consecutive times), the distortion of single-shot positioning data caused by wind disturbance, light and shadow interference, and instantaneous signal fluctuations in the field environment is avoided, providing sufficient sample data for subsequent data noise reduction and optimization.

[0047] Step S303: Calculate the average coordinate value of the coordinate positioning data under the acquisition quantity, and obtain the three-dimensional world coordinates of the marker at the target position by the coordinate difference between the average coordinate value and the origin.

[0048] For the multiple sets of raw coordinate positioning data collected in step S302, unified computational processing is performed. By calculating the average coordinates of all collected data, random observation errors, instantaneous signal errors, and environmental interference errors are filtered and denoised to remove abnormal deviation data, resulting in high-precision average coordinate data of the marker target position. Furthermore, using the origin of the starting position coordinates set in step S301 as a reference, the three-dimensional coordinate difference between this average coordinate and the origin is calculated. This coordinate difference accurately represents the spatial displacement change of the marker from the starting position to the target position, ultimately obtaining high-precision and high-stability three-dimensional world coordinates of the marker at the target position. This provides an accurate true coordinate benchmark for subsequent pixel coordinate regression fitting, residual calculation, and landslide deformation calculation.

[0049] Optionally, using the target position coordinates in the world coordinate system where the marker is located as a reference, the regression coefficient matrix and bias vector mapping the binocular four-dimensional pixel coordinates to three-dimensional world coordinates are calculated through second-order polynomial regression, including the following steps: Step S401: Calculate the feature mapping results of the markers using second-order polynomial regression.

[0050] The binocular four-dimensional pixel coordinate system integrates the row and column pixel information of the marker in the images captured by the left and right cameras, which can completely reflect the imaging characteristics of the marker in the binocular vision system. Considering that the field landslide monitoring scene is susceptible to interference from nonlinear factors such as imaging distortion, terrain obstruction, changes in observation distance, and lens perspective deviation, the accuracy of traditional linear mapping methods is insufficient. This step performs second-order polynomial fitting calculation on the acquired binocular four-dimensional pixel coordinates. By using the second-order polynomial model, the intrinsic correlation between pixel data and real spatial location is explored, while invalid interference information such as imaging noise and instantaneous observation errors is filtered out. Finally, a standardized and highly adaptable marker feature mapping result is output, providing a high-precision feature dataset for subsequent coordinate fitting modeling.

[0051] Step S402: Based on the target position coordinates of the world coordinate system where the marker is located, use the least squares method to fit the feature mapping result to the target position coordinates of the marker, and then iteratively calculate the regression transformation data and offset transformation data between the binocular four-dimensional pixel coordinates and the three-dimensional world coordinates.

[0052] This step uses the true 3D coordinates of the target location obtained after multiple samplings and mean denoising by the differential positioning device as a unified fitting benchmark. Leveraging the global optimal fitting property of the least squares method, the feature mapping results obtained in the previous step are fitted to converge to the true 3D coordinates. By minimizing the sum of squared residuals, systematic biases, nonlinear distortion biases, and random observation biases between the pixel imaging space and the real geographic space are eliminated to the greatest extent possible. After fitting, a cyclical calculation is used to sequentially solve for the regression transformation base data between the binocular four-dimensional pixel coordinates and the 3D world coordinates, as well as the bias transformation base data used to compensate for various fixed biases, thus achieving accurate matching between pixel features and spatial coordinates.

[0053] Step S403: Generate a regression coefficient matrix for converting binocular four-dimensional pixel coordinates to three-dimensional world coordinates using regression transformation data, and generate an offset vector for converting binocular four-dimensional pixel coordinates to three-dimensional world coordinates using offset transformation data.

[0054] The regression transformation data obtained in step S402 undergoes structured integration, dimension matching, and parameter normalization. Following the mapping dimensional rules from binocular four-dimensional pixel coordinates to three-dimensional world coordinates, a standardized regression coefficient matrix is ​​generated. This matrix serves as the core weight parameter of the coordinate mapping model, characterizing the influence weight and mapping rules of each dimension of four-dimensional pixel data on three-dimensional spatial coordinates. Simultaneously, the bias transformation data is integrated and calculated to generate a corresponding dimension bias vector. This vector uniformly compensates for binocular camera installation deviations, coordinate system calibration errors, inherent lens distortion, and various systematic fixed interferences from the field environment. The regression coefficient matrix and bias vector, used together, complete the mapping model for the transformation from pixel coordinates to three-dimensional world coordinates, providing core parameter support for subsequent real-time calculation of the basic three-dimensional coordinates of markers.

[0055] Optionally, the covariance and random variance values ​​of the binocular four-dimensional pixel coordinates mapped to the target position coordinates are calculated using Gaussian process regression, including: Step S501: After performing Gaussian process regression calculation on the binocular four-dimensional pixel coordinates and target position coordinates using preset variance scale kernel and white noise variance kernel, the residual values ​​are obtained.

[0056] Using the binocular four-dimensional pixel coordinates acquired by the binocular camera and the true three-dimensional coordinates of the target location after calibration by the differential positioning device as the calculation objects, and selecting pre-configured variance scaling kernel and white noise variance kernel as the core calculation basis, Gaussian process regression is performed on the two sets of matched data to finally calculate the residual value generated in the process of mapping pixel coordinates to spatial coordinates. This residual value summarizes the overall deviation caused by various interferences such as imaging distortion, terrain occlusion, observation distance differences, and lens perspective deviation in the field landslide scene, and serves as the basis for subsequent error data extraction and error feature classification analysis.

[0057] Step S502: Use the residual value to obtain the nonlinear mapping data when the binocular four-dimensional pixel coordinates are mapped to the target position coordinates, and use the preset Matrn function and Kronecker function to calculate the position change data and random change data of the nonlinear mapping data under the residual value.

[0058] Based on the residual values ​​obtained in the previous step, nonlinear mapping data generated during the mapping of binocular four-dimensional pixel coordinates to target position coordinates is further extracted. This type of data includes various nonlinear error components such as dimensional mapping bias, spatial distortion error, and fitting residuals, comprehensively covering various mapping interference factors under complex field conditions. Subsequently, the pre-defined Matrn and Kronecker functions are called to classify the nonlinear mapping data based on the distribution characteristics of the residuals, obtaining the corresponding positional variation data and random variation data. Among them, the positional variation data is a continuous, spatially correlated systematic deviation data formed by the influence of terrain slope, observation distance, camera perspective distortion, and spatial distribution of points; the random variation data is an irregular, discrete random deviation data caused by factors such as field wind disturbance, light and shadow fluctuations, equipment signal jitter, and pixel sampling noise.

[0059] Step S503: Calculate the covariance value of the nonlinear mapping data by multiplying the location change data with the variance scale kernel, and calculate the random variance value of the nonlinear mapping data by multiplying the white noise variance kernel with the randomly changing data.

[0060] The location change data is multiplied by the variance scale kernel to quantify the spatially correlated systematic errors in the coordinate mapping process, ultimately calculating the covariance value corresponding to the nonlinear mapping data. Simultaneously, the white noise variance kernel is multiplied by the randomly changing data to quantify various instantaneous discrete random errors during monitoring, thus obtaining the random variance value corresponding to the nonlinear mapping data. The covariance and random variance values ​​obtained through this step can completely distinguish and characterize the systematic and random errors in the coordinate mapping system, providing reliable data support for subsequent accurate calculation of local residual values, error compensation, and high-precision calculation of the three-dimensional coordinates of markers.

[0061] Optionally, the underlying three-dimensional coordinates of the marker can be calculated using the regression coefficient matrix and the bias vector, including: Step S601: After performing second-order polynomial expansion on the real-time acquired binocular four-dimensional pixel coordinates, the first-order feature data, second-order feature data, and cross-feature data of the binocular four-dimensional pixel coordinates are obtained.

[0062] During real-time monitoring of the landslide site, the binocular four-dimensional pixel coordinates corresponding to the markers in the current image are continuously collected. To fully analyze the variation patterns of the pixel data and mitigate the nonlinear interference caused by the field environment, a comprehensive second-order polynomial expansion calculation is performed on the collected binocular four-dimensional pixel coordinates.

[0063] This operation yields three types of feature data: first-order feature data, which reflects the basic linear variation of pixel coordinates; second-order feature data, which reflects the nonlinear distortion and deviation characteristics of pixel coordinates; and cross-feature data, which characterizes the coupling and interference between the various dimensions of the four-dimensional pixel. This step fully preserves the effective information in the binocular imaging process, deeply analyzes the nonlinear errors of pixels under complex conditions, and prepares a complete high-dimensional feature dataset for subsequent coordinate mapping operations.

[0064] Step S602: Determine the real-time feature mapping data of the marker based on the first-order feature data, second-order feature data, and cross-feature data.

[0065] Based on the first-order, second-order, and cross-feature data obtained in the previous step, feature selection, noise reduction, dimensional alignment and weighted fusion, and normalization are performed sequentially to remove invalid noise and redundant interference features from the data, and to select core effective features that are suitable for the on-site camera imaging parameters, monitoring distance, and terrain environment. The processed features are then integrated to finally determine the real-time feature mapping data of the marker. This data integrates multiple types of information, including linear changes, nonlinear distortion, and dimensional coupling, and can accurately represent the true pixel distribution of the marker under a binocular vision imaging system, completing the transformation from raw pixel data to high-dimensional feature data.

[0066] Step S603: Calculate the vector product of the transpose of the regression coefficient matrix and the real-time feature mapping data, and calculate the basic three-dimensional coordinates of the marker based on the vector product result and the vector summation result of the bias vector.

[0067] First, the regression coefficient matrix from the previous training is retrieved and transposed to obtain the transpose matrix. Then, this transpose matrix is ​​multiplied with the real-time feature mapping data to perform a preliminary mapping transformation from high-dimensional pixel features to spatial coordinates. Subsequently, the resulting vector product is accumulated dimension-by-dimensionally with a pre-fitted and calibrated bias vector. This bias vector is used to compensate for various fixed systematic errors, such as binocular camera installation deviations, coordinate system calibration residuals, and inherent lens distortion. After these matrix and vector operations, the basic three-dimensional coordinates of the marker are finally calculated. These coordinates eliminate most of the inherent systematic errors of binocular vision, providing reliable basic data for subsequent local residual correction and landslide monitoring result calculation.

[0068] Optionally, the local residuals of the marker can be calculated using the covariance and random variance values, including the following steps: Step S701: After performing Z-score processing on the real-time acquired binocular four-dimensional pixel coordinates, dimensionless coordinate data of the binocular four-dimensional pixel coordinates is obtained, and a sample vector of the marker is constructed based on the dimensionless coordinate data.

[0069] During real-time monitoring, raw binocular four-dimensional pixel coordinate data of markers is collected. This data includes multi-dimensional pixel parameters corresponding to the left and right cameras, which generally suffer from problems such as inconsistent dimensions, large differences in numerical scales, and susceptibility to extreme value noise. Directly using this data in calculations can cause dimensional weight imbalances and distorted error calculations. Therefore, the Z-score standardization algorithm is used to normalize the real-time pixel coordinates. By normalizing the mean to zero and the variance to one, the dimensional and scale deviations between dimensions are eliminated, resulting in standardized dimensionless coordinate data. Then, the dimensionless coordinate data is dimensionally normalized and vectorized to construct marker sample vectors adapted to the error calculation scenario, thereby ensuring the accuracy and stability of subsequent error calculation processes.

[0070] Step S702: Calculate the position change value corresponding to the sample vector using the covariance value, and calculate the random noise value corresponding to the sample vector using the random variance value.

[0071] Using the standardized sample vector generated in the previous step as input, error calculations are performed separately. The covariance value is used to calculate the positional variation caused by factors such as terrain undulation, observation distance offset, and nonlinear deviation of coordinate mapping. This value represents the systematic local offset error with spatial correlation. Simultaneously, the random variance value is used to extract the random noise value caused by fluctuations in field lighting and shadows, airflow disturbances, equipment sampling noise, and signal jitter. This value corresponds to the irregular instantaneous observation error. This step achieves separate calculation of systematic and random errors, effectively avoiding the problem of insufficient quantization accuracy caused by the mixing of these two types of errors.

[0072] Step S703: Calculate the local residual value of the marker by superimposing the position change value and the random noise value.

[0073] The obtained position change value is superimposed with the random noise value to integrate the systematic local deformation deviation and instantaneous random observation deviation of the monitoring point. This comprehensively summarizes all kinds of subtle error components in the basic 3D coordinates that have not yet been corrected, and finally calculates the local residual value of the marker's current position. This local residual value can fully characterize the comprehensive residual error of a single frame and a single measurement point, and can be used as a basis for error compensation to correct residual deviations in the basic 3D coordinates, providing support for achieving high-precision calculation from pixel coordinates to 3D world coordinates.

[0074] Optionally, step S104, which obtains the landslide monitoring results of the markers in the landslide monitoring area through the superposition of the basic three-dimensional coordinates and local residual values, includes the following steps: Step S801: Determine the real-time position coordinates of the marker in three-dimensional world coordinates based on the superposition result of the basic three-dimensional coordinates and the local residual values.

[0075] The calculated base 3D coordinates are vector-superimposed with the real-time calculated local residual values. The local residual values ​​are used to comprehensively compensate for and correct the spatial correlation errors and random noise errors remaining in the base 3D coordinates, thus overcoming the accuracy shortcomings of traditional close-range photogrammetry. After superposition and correction, the real-time position coordinates of the marker in a unified 3D world coordinate system can be determined. These coordinates effectively eliminate various systematic biases and random errors, and can accurately and objectively reflect the marker's current actual spatial position.

[0076] Step S802: Determine the landslide monitoring results of the marker in the landslide monitoring area based on the change value of the real-time location coordinates in the landslide monitoring area.

[0077] Throughout the monitoring period, real-time location coordinates of the markers are continuously collected, forming complete time-series coordinate data. The real-time location coordinates at different times are compared with the initial baseline coordinates and historical monitoring coordinates to calculate the three-dimensional coordinate changes of the markers. Based on these changes, core deformation indicators such as total three-dimensional displacement, displacement direction, vertical settlement, and deformation rate are further calculated. Simultaneously, considering the differentiated layout characteristics of markers within the region, a comprehensive analysis of the deformation distribution, soil sliding trend, and regional activity level in the landslide area is conducted. Ultimately, complete landslide monitoring results are determined, providing accurate data for landslide disaster early warning and slope stability assessment.

[0078] The landslide monitoring method based on hybrid three-dimensional coordinate solution in the above embodiments can be implemented by constructing a hybrid three-dimensional coordinate solution model, specifically including the following steps: (1) Model data acquisition.

[0079] Markers were placed on the landslide surface, and a binocular camera was deployed on a stable foundation outside the landslide. The close-range photogrammetry system remained relatively stationary with respect to the landslide body. Controlled local movement of the markers was then performed, i.e., manually moving each marker in a specific direction. After the movement was completed, images were captured using the binocular camera, and the two-dimensional pixel coordinates of the markers in the left and right cameras were extracted. These images were then stitched together to extract the binocular four-dimensional pixel coordinates. ,specific (Where u and v correspond to column coordinates and row coordinates respectively, and L and R are the left and right camera coordinates respectively).

[0080] Simultaneously, for each marker after each movement, coordinate data is continuously and repeatedly acquired five times using RTK (Real-Time Kinematic) and the average value is calculated, which is then used as the true 3D world coordinates at that location. ,specific The above operations are used to obtain multiple sets of corresponding binocular four-dimensional pixel coordinates and real three-dimensional world coordinates, which are then used as training data for building subsequent models.

[0081] (2) Construct a hybrid three-dimensional coordinate solution model.

[0082] Based on the training data collected in step (1), a three-dimensional coordinate solution model is constructed, which includes global multinomial regression and local residual Gaussian process regression. The three-dimensional coordinate solution process of this model is divided into two steps: basic trend fitting and local residual compensation.

[0083] Basic trend fitting steps: First, a global 3D mapping benchmark is extracted using second-order polynomial regression to perform basic trend fitting. Then, the input binocular 4D pixel coordinates are... Perform a second-order polynomial expansion, and let the eigenmap after the expansion be... This mapping includes all first-order, second-order, and cross terms of the input features. The expanded features are then input into a second-order polynomial regression model to calculate the global fundamental 3D coordinates, denoted as... The mathematical expression for this basic prediction model is: ;in, This is the regression coefficient matrix. This is the bias vector. During the model training phase, the acquired binocular four-dimensional pixel coordinates are fitted using the least squares method. Corresponding real three-dimensional world coordinates Solve for the parameters and This establishes a global mapping relationship and outputs the global basic three-dimensional coordinates. .

[0084] Local residual compensation step: Calculate the true 3D world coordinates With global basic 3D coordinates The residual between them is denoted as For the input binocular four-dimensional pixel coordinates Z-score standardization is performed to eliminate dimensional differences; the mathematical formula for its calculation is as follows: ;in, For the input binocular four-dimensional pixel coordinate vector, the first... The original coordinate values ​​of each dimension, and These are the mean and standard deviation of this dimension in the training data, respectively. These are the standardized dimensionless eigenvalues.

[0085] The standardized binocular four-dimensional pixel coordinates are input into a Gaussian process regression model containing a composite kernel function. Nonlinear mappings between pixel features and residuals Δy in the three dimensions are fitted to achieve local residual compensation. The composite kernel function consists of a constant kernel and a smoothness parameter of 2.5. The combination of a nucleus and a white noise nucleus is specifically as follows: ;in, The variance scaling parameter is a constant kernel. The variance of white noise. For the Kronecker function, and Let p and q be the input training sample vectors of the Gaussian process regression model. The Gaussian process regression model outputs the predicted local residuals based on the aforementioned composite kernel function, denoted as . .

[0086] (3) Three-dimensional coordinate calculation based on the constructed model.

[0087] In practical close-range photogrammetric landslide monitoring, real-time binocular four-dimensional pixel coordinates of markers are continuously acquired. When the marker's trajectory lies within the calibrated field-of-view mapping space, it is input into a pre-trained three-dimensional coordinate calculation model that incorporates global multinomial regression and local residual Gaussian process regression. The basic three-dimensional coordinates output by the second-order multinomial regression model are then used. Local residual predictions from the Gaussian process regression model output By superimposing the coordinates, the high-precision three-dimensional world coordinates of the marker are obtained, denoted as y. The final solution equation is: .

[0088] Through the above steps, the conversion from binocular four-dimensional pixel coordinates to high-precision three-dimensional world coordinates in the field environment was achieved.

[0089] To verify the effectiveness of the above model, an accuracy verification experiment has been conducted at the landslide site. Figure 3 The coordinate solution error distribution of this experiment is shown. The results show that the average fitting deviation of this model in the three independent dimensions of north, east and elevation is 0.31cm, 0.58cm and 0.40cm respectively, and the average deviation of the combined three-dimensional Euclidean distance is 0.82cm.

[0090] The data demonstrates that the model establishes a high-precision coordinate mapping relationship within the local spatial range defined by the known data.

[0091] Figure 4The error distribution of RTK positioning results when repeated measurements are taken at the same location is shown. The results indicate that, affected by factors such as the field environment, the three-dimensional spatial error of a single RTK positioning varies from 0.1 cm to 1.2 cm. Based on this, the error benchmark boundary of RTK measurements under normal observation conditions can be established as approximately 1 cm.

[0092] The three-dimensional coordinate points calculated by the model are spatially superimposed with the scatter set obtained from multiple consecutive RTK measurements at the corresponding time points. Figure 5 The comparison results of the spatial displacement trajectory of the marker are presented. It can be seen that more than 90% of the coordinate points output by the solution model of this invention fall within or on the edge of a reference circle with the average value of RTK measurements as the center and a radius of 1 cm; and the displacement path calculated by the model is consistent with the spatial orientation of the RTK measured displacement path.

[0093] As can be seen from the above-mentioned landslide monitoring method based on hybrid three-dimensional coordinate calculation, this method accurately calculates the basic three-dimensional coordinates and local residual values ​​of the markers by means of the coordinate mapping relationship between the camera and the markers in the landslide monitoring area, and obtains the landslide monitoring results of the markers in real time by superimposing the basic three-dimensional coordinates and local residual values. This achieves high-precision calculation of the markers from image data to three-dimensional world coordinates, which greatly improves the measurement accuracy of close-range photography.

[0094] Corresponding to the above-described embodiment of the landslide monitoring method based on hybrid three-dimensional coordinate calculation, this embodiment of the invention also provides a landslide monitoring system based on hybrid three-dimensional coordinate calculation, such as... Figure 6 As shown, the system includes: World coordinate acquisition module 100: used to acquire the binocular four-dimensional pixel coordinates of the marker and obtain the three-dimensional world coordinates of the marker after it is detected that the marker in the landslide monitoring area has moved to the target position; First coordinate mapping module 200: used to calculate the feature mapping result of the marker through second-order polynomial regression; based on the target position coordinates of the world coordinate system where the marker is located, the feature mapping result is fitted to the target position coordinates of the marker using the least squares method, and then the regression transformation data and bias transformation data between the binocular four-dimensional pixel coordinates and the three-dimensional world coordinates are calculated iteratively; the regression coefficient matrix of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates is generated through the regression transformation data, and the bias vector of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates is generated through the bias transformation data. The second coordinate mapping module 300 is used to perform Gaussian process regression calculations on the binocular four-dimensional pixel coordinates and target position coordinates using preset variance scale kernels and white noise variance kernels to obtain residual values; it uses the residual values ​​to obtain nonlinear mapping data when the binocular four-dimensional pixel coordinates are mapped to the target position coordinates, and uses preset Matrn and Kronecker functions to calculate the position change data and random change data of the nonlinear mapping data under the residual values, respectively; it calculates the covariance value of the nonlinear mapping data by multiplying the position change data with the variance scale kernel, and calculates the random variance value of the nonlinear mapping data by multiplying the white noise variance kernel with the random change data. Coordinate prediction processing module 400: used to calculate the basic three-dimensional coordinates of the marker using the regression coefficient matrix and the bias vector, and to calculate the local residual values ​​of the marker using the covariance value and the random variance value; Landslide monitoring execution module 500: Used to determine the landslide monitoring results of markers in the landslide monitoring area by superimposing the basic three-dimensional coordinates and local residual values.

[0095] As can be seen from the above landslide monitoring system based on hybrid three-dimensional coordinate calculation, the system accurately calculates the basic three-dimensional coordinates and local residual values ​​of the markers by means of the coordinate mapping relationship between the camera and the markers in the landslide monitoring area, and obtains the landslide monitoring results of the markers in real time by superimposing the basic three-dimensional coordinates and local residual values. Thus, it realizes high-precision calculation of the markers from image data to three-dimensional world coordinates, which greatly improves the measurement accuracy of close-range photography.

[0096] The landslide monitoring system based on hybrid three-dimensional coordinate calculation provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned landslide monitoring method based on hybrid three-dimensional coordinate calculation. For the sake of brevity, any parts not mentioned in the system embodiment can be referred to the corresponding content in the aforementioned landslide monitoring method based on hybrid three-dimensional coordinate calculation.

[0097] This embodiment also provides an electronic device, the structural schematic diagram of which is shown below. Figure 7 As shown, the device includes a processor 101 and a memory 102; wherein, the memory 102 is used to store one or more computer instructions, which are executed by the processor to implement the steps of the landslide monitoring method based on hybrid three-dimensional coordinate calculation described above.

[0098] Figure 7 The electronic device shown also includes a bus 103 and a communication interface 104, with the processor 101, communication interface 104 and memory 102 connected via the bus 103.

[0099] The memory 102 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. The bus 103 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 7 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0100] The communication interface 104 is used to connect to at least one user terminal and other network units through a network interface, and to send encapsulated IPv4 packets or IPv4 packets to the user terminal through the network interface.

[0101] Processor 101 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 101 or by instructions in software form. The processor 101 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this disclosure can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 102. The processor 101 reads the information in memory 102 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments.

[0102] This invention also provides a storage medium storing a computer program, which, when run by a processor, executes the steps of the landslide monitoring method based on hybrid three-dimensional coordinate calculation in the foregoing embodiments.

[0103] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, devices, and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0104] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0105] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0106] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0107] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A landslide monitoring method based on hybrid three-dimensional coordinate solution, characterized in that, The method includes: Once a marker in the landslide monitoring area is detected to have moved to the target location, the binocular four-dimensional pixel coordinates of the marker are collected, and the three-dimensional world coordinates of the marker are obtained. The feature mapping result of the marker is calculated by second-order polynomial regression; based on the target position coordinates of the world coordinate system where the marker is located, the feature mapping result is fitted to the target position coordinates corresponding to the marker using the least squares method, and then the regression transformation data and bias transformation data between the binocular four-dimensional pixel coordinates and the three-dimensional world coordinates are calculated iteratively; the regression coefficient matrix of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates is generated by the regression transformation data, and the bias vector of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates is generated by the bias transformation data. After performing Gaussian process regression calculations on the binocular four-dimensional pixel coordinates and the target position coordinates using a preset variance scale kernel and a white noise variance kernel, residual values ​​are obtained. The residual values ​​are then used to obtain nonlinear mapping data when the binocular four-dimensional pixel coordinates are mapped to the target position coordinates. Preset Matrn and Kronecker functions are used to calculate the positional variation data and random variation data of the nonlinear mapping data under the residual values, respectively. The covariance value of the nonlinear mapping data is calculated using the product of the positional variation data and the variance scale kernel, and the random variance value of the nonlinear mapping data is calculated using the product of the white noise variance kernel and the random variation data. The basic three-dimensional coordinates of the marker are calculated using the regression coefficient matrix and the bias vector, and the local residual values ​​of the marker are calculated using the covariance value and the random variance value. The landslide monitoring results of the marker in the landslide monitoring area are determined by superimposing the basic three-dimensional coordinates and the local residual values.

2. The landslide monitoring method based on hybrid three-dimensional coordinate solution according to claim 1, characterized in that, The steps of acquiring the binocular four-dimensional pixel coordinates of the marker and obtaining the three-dimensional world coordinates of the marker after detecting that the marker in the landslide monitoring area has moved to the target location include: Identify the binocular cameras installed in the foundation area of ​​the landslide monitoring area, and identify the markers deployed in the slope area of ​​the landslide monitoring area; The direction of movement and target location of the marker are determined based on the slope direction in the slope area. After moving the marker from the starting position to the target position according to the moving direction, the binocular camera is controlled to acquire first and second eye images of the marker in different directions; Using the first column and first row coordinates of the marker region in the first visual image and the second column and second row coordinates of the marker region in the second visual image, the binocular four-dimensional pixel coordinates of the marker are determined; Using the coordinates of the starting position in the world coordinate system as a reference, obtain the three-dimensional world coordinates of the marker from the starting position to the target position.

3. The landslide monitoring method based on hybrid three-dimensional coordinate solution according to claim 2, characterized in that, The step of obtaining the three-dimensional world coordinates of the marker from the starting position to the target position, using the coordinates of the starting position in the world coordinate system as a reference, includes: The differential positioning device installed at the foundation area of ​​the landslide monitoring area is obtained, and the world coordinate system of the marker is determined using the coordinate system of the differential positioning device; wherein, the starting position is located at the origin of the world coordinate system; Once the marker is detected to have moved from the starting position to the target position, the differential positioning device is controlled to continuously collect the coordinate positioning data of the marker multiple times according to a preset number of data collections. Calculate the average coordinate value of the coordinate positioning data under the specified number of acquisitions, and obtain the three-dimensional world coordinates of the marker at the target location by the coordinate difference between the average coordinate value and the origin.

4. The landslide monitoring method based on hybrid three-dimensional coordinate solution according to claim 1, characterized in that, Calculating the basic three-dimensional coordinates of the marker using the regression coefficient matrix and the bias vector includes: After performing a second-order polynomial expansion on the real-time acquired binocular four-dimensional pixel coordinates, the first-order feature data, second-order feature data, and cross-feature data of the binocular four-dimensional pixel coordinates are obtained. The real-time feature mapping data of the marker is determined based on the first-order feature data, the second-order feature data, and the cross-feature data. Calculate the vector product of the transpose of the regression coefficient matrix and the real-time feature mapping data, and calculate the basic three-dimensional coordinates of the marker based on the vector product result and the vector summation result of the bias vector.

5. The landslide monitoring method based on hybrid three-dimensional coordinate solution according to claim 1, characterized in that, Calculating the local residuals of the marker using the covariance and the random variance includes: After performing Z-score processing on the real-time acquired binocular four-dimensional pixel coordinates, dimensionless coordinate data of the binocular four-dimensional pixel coordinates is obtained, and a sample vector of the marker is constructed based on the dimensionless coordinate data. The position change value corresponding to the sample vector is calculated using the covariance value, and the random noise value corresponding to the sample vector is calculated using the random variance value. The local residual value of the marker is calculated by superimposing the position change value and the random noise value.

6. The landslide monitoring method based on hybrid three-dimensional coordinate solution according to claim 1, characterized in that, The steps for obtaining the landslide monitoring results of the marker in the landslide monitoring area by superimposing the basic three-dimensional coordinates and the local residual values ​​include: The real-time position coordinates of the marker in the three-dimensional world coordinates are determined based on the superposition result of the basic three-dimensional coordinates and the local residual values. The landslide monitoring results of the marker in the landslide monitoring area are determined based on the change value of the real-time location coordinates in the landslide monitoring area.

7. A landslide monitoring system based on hybrid three-dimensional coordinate solution, characterized in that, The system includes: World coordinate acquisition module: used to collect the binocular four-dimensional pixel coordinates of the marker and obtain the three-dimensional world coordinates of the marker after it is detected that the marker in the landslide monitoring area has moved to the target position; The first coordinate mapping module is used to calculate the feature mapping result of the marker through second-order polynomial regression; based on the target position coordinates of the world coordinate system where the marker is located, the feature mapping result is fitted to the target position coordinates corresponding to the marker using the least squares method, and then the regression transformation data and bias transformation data between the binocular four-dimensional pixel coordinates and the three-dimensional world coordinates are calculated iteratively; the regression coefficient matrix of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates is generated through the regression transformation data, and the bias vector of the binocular four-dimensional pixel coordinates to the three-dimensional world coordinates is generated through the bias transformation data. The second coordinate mapping module is used to perform Gaussian process regression calculations on the binocular four-dimensional pixel coordinates and the target position coordinates using a preset variance scale kernel and a white noise variance kernel to obtain residual values; it uses the residual values ​​to obtain nonlinear mapping data when the binocular four-dimensional pixel coordinates are mapped to the target position coordinates, and uses preset Matrn and Kronecker functions to calculate the position variation data and random variation data of the nonlinear mapping data under the residual values, respectively; it calculates the covariance value of the nonlinear mapping data by multiplying the position variation data with the variance scale kernel, and calculates the random variance value of the nonlinear mapping data by multiplying the white noise variance kernel with the random variation data; Coordinate prediction processing module: used to calculate the basic three-dimensional coordinates of the marker using the regression coefficient matrix and the bias vector, and to calculate the local residual value of the marker using the covariance value and the random variance value; Landslide monitoring execution module: used to determine the landslide monitoring results of the marker in the landslide monitoring area by superimposing the basic three-dimensional coordinates and the local residual values.

8. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing computer-executable instructions that can be executed by the processor, the processor executing the computer-executable instructions to implement the steps of the landslide monitoring method based on hybrid three-dimensional coordinate calculation as described in any one of claims 1 to 6.

Images