Slope displacement monitoring method based on optimal feature block adaptive identification and application

Abstract

This invention discloses a slope displacement monitoring method and its application based on adaptive identification of optimal feature blocks, belonging to the fields of engineering monitoring and computer vision. The method includes: acquiring time-series images of the slope; automatically identifying the "optimal tracking feature block" suitable for tracking in the image; adaptively deriving key parameters such as the query window size and search area required for LSPIV calculation based on the texture, stability, and other attributes of the feature block; performing high-precision displacement calculation using the adaptive parameters; and finally obtaining the full-field displacement field through interpolation. This invention completely changes the drawback of traditional LSPIV relying on manually set parameters, realizing intelligent and adaptive processes from image analysis to displacement calculation, significantly improving the automation, calculation accuracy, efficiency, and robustness to complex environments of slope displacement monitoring, and is suitable for automated early warning and long-term monitoring of geological disasters.

Description

Technical Field

[0001] This invention relates to the fields of engineering safety monitoring and computer vision technology, and in particular to a slope surface displacement monitoring method and application based on large-scale particle image velocimetry (LSPIV) technology, which integrates automatic image feature recognition and adaptive parameter optimization. Background Technology

[0002] LSPIV technology, as a non-contact, full-field measurement method, has been successfully applied in river surface velocity measurement. In recent years, it has been explored for use in monitoring surface displacement of geological bodies such as slopes and landslides. Its basic principle is to perform cross-correlation analysis on a series of continuously captured images to calculate the movement vectors of surface textures or natural landmarks, thereby obtaining full-field displacement information.

[0003] However, there are significant bottlenecks in directly applying LSPIV to slope monitoring: (1) Strong parameter dependence and low degree of automation. The core calculation steps of LSPIV (cross-correlation analysis) are highly dependent on manually set parameters, such as query window size, search area, step size and correlation threshold. These parameters require operators to repeatedly adjust them based on image texture, estimated displacement and scene experience. The process is cumbersome, subjective, and difficult to maintain optimal performance in different scenes or at different times in the same scene. (2) Contradiction between calculation efficiency and accuracy. To capture displacement, a sufficiently large search area needs to be set, but an excessively large area will increase the amount of calculation and noise interference; to accurately locate, a smaller query window needs to be used, but an excessively small window is prone to matching failure in areas with scarce texture. This contradiction is difficult to balance by fixing parameters. (3) Poor robustness to complex scenes: Slope environments have problems such as changes in lighting, partial occlusion by vegetation, and uneven texture. The calculation strategy with fixed parameters is prone to generating a large number of abnormal vectors, and subsequent data processing work is heavy.

[0004] Existing technologies lack an effective solution for automatically analyzing image content and intelligently setting calculation parameters to achieve high-precision, fully automatic, and adaptive monitoring of slope displacement. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the shortcomings of existing slope LSPIV monitoring technology, such as parameter setting relying on manual methods, low efficiency, unstable accuracy, and poor adaptability, and to provide a slope displacement monitoring method and application that can automatically identify the optimal tracking features, adaptively configure calculation parameters, and realize intelligent processing.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a slope displacement monitoring method based on optimal feature block adaptive identification, comprising the following steps: S1, acquire the time-series image sequence of the slope monitoring area; S2, automatically identify multiple optimal tracking feature blocks on the current frame image in the image sequence; S3, based on the image attributes of the optimal tracking feature block, adaptively derive the parameters for large-scale particle image velocimetry (LSPIV) calculation, the parameters including at least the cross-correlation query window size and the search area; S4. Using the adaptively derived parameters, perform LSPIV calculation on the time-series image sequence to obtain the displacement vector on the optimal tracking feature block; S5. Based on the displacement vector on the optimal tracking feature block, the full-field displacement field of the slope monitoring area is generated by spatial interpolation.

[0007] Optionally, step S2, which automatically identifies the optimal tracking feature block, includes: S21, Extract feature points or feature regions from the current frame image; S22, score the candidate block centered on the feature point or feature region based on a preset evaluation index system, the evaluation index system including local texture richness, multi-frame feature stability and spatial distribution rationality; S23, select a predetermined number of candidate blocks as the optimal tracking feature blocks based on the scoring results.

[0008] Optionally, step S3 is based on the adaptive derivation of parameters for the optimal tracking feature block: The derivation of the cross-correlation query window size is based on the texture scale of each optimal tracking feature block; The derivation process of the search region is as follows: First, estimate the initial displacement of the optimal tracking feature block between adjacent frames, and then determine a dynamic search region range centered on the target position pointed to by the initial displacement.

[0009] Optionally, the method for estimating the initial displacement is either the sparse optical flow method or the fast feature matching method.

[0010] Optionally, the spatial interpolation in step S5 is Kriging interpolation or plate spline interpolation.

[0011] A computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the above-described method.

[0012] A computer device includes a memory, a processor, and a program stored in the memory and executable thereon, the program being executed by the processor to implement the steps of the method described above.

[0013] The beneficial effects of this invention are: It achieves full-process automation and intelligence, replacing manual parameter trial and error with automatic image content analysis, which greatly improves the level of automation and engineering practical value of monitoring.

[0014] It improves computational accuracy and robustness. The computational strategy based on optimal feature blocks ensures high confidence of the displacement vector. Adaptive parameter setting enables the computational strategy to best match the local image content, effectively suppressing noise and mismatches.

[0015] It significantly improves computational efficiency. The dynamic search region technology concentrates the computational load on the location where the target is most likely to appear, avoiding global redundant computation, and is particularly suitable for long-term, high-frequency real-time monitoring scenarios.

[0016] It enhances the monitoring's adaptive capabilities, enabling the monitoring system to automatically adjust its calculation strategies based on changes in sunlight, season, and vegetation cover, thus exhibiting greater adaptability to complex and ever-changing outdoor slope environments. Attached Figure Description

[0017] Figure 1 This is an overall flowchart of the method of the present invention.

[0018] Figure 2 This is a diagram of the calculated slope surface displacement field. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] This embodiment provides a slope displacement monitoring method based on optimal feature block adaptive identification. The execution flow of this method is as follows: Figure 1 As shown, it includes the following steps: S1, acquire the time-series image sequence of the slope monitoring area; S2, automatically identify multiple optimal tracking feature blocks on the current frame image in the image sequence; Specifically, step S2, which automatically identifies the optimal tracking feature block, includes: S21, Extract feature points or feature regions from the current frame image; S22, score the candidate block centered on the feature point or feature region based on a preset evaluation index system, the evaluation index system including local texture richness, multi-frame feature stability and spatial distribution rationality; S23, select a predetermined number of candidate blocks as the optimal tracking feature blocks based on the scoring results.

[0021] S3, based on the image attributes of the optimal tracking feature block, adaptively derive the parameters for large-scale particle image velocimetry (LSPIV) calculation, the parameters including at least the cross-correlation query window size and the search area; Specifically, step S3 is based on the adaptive derivation of parameters for the optimal tracking feature block: The derivation of the cross-correlation query window size is based on the texture scale of each optimal tracking feature block; The derivation process of the search region is as follows: First, estimate the initial displacement of the optimal tracking feature block between adjacent frames, and then determine a dynamic search region range centered on the target position pointed to by the initial displacement.

[0022] The method for estimating the initial displacement is the sparse optical flow method.

[0023] S4. Using the adaptively derived parameters, perform LSPIV calculation on the time-series image sequence to obtain the displacement vector on the optimal tracking feature block; S5. Based on the displacement vector on the optimal tracking feature block, the full-field displacement field of the slope monitoring area is generated by spatial interpolation.

[0024] Specifically, the spatial interpolation in step S5 is Kriging interpolation.

[0025] A computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the above-described method.

[0026] A computer device includes a memory, a processor, and a program stored in the memory and executable thereon, the program being executed by the processor to implement the steps of the method described above.

[0027] The implementation process and effects of the method of the present invention will be described in detail below with reference to a practical application example of the technical solution of the present invention. Specifically, the implementation process of the method includes the following steps: S1, acquire the time-series image sequence of the slope monitoring area; S2, automatically identify multiple optimal tracking feature blocks on the current frame image in the image sequence; Specifically, step S2, which automatically identifies the optimal tracking feature block, includes: S21. The SIFT algorithm is used to extract full-map feature points from the reference image. To avoid interference from unstable areas such as vegetation, the DSM and NDVI index maps generated from the reference perimeter are used. By setting slope thresholds and NDVI thresholds, densely vegetated areas and overly flat non-interest areas (such as platforms) are pre-masked, and feature detection is concentrated on the exposed rock and soil slope surface, initially obtaining about 5,000 feature points. S22: Generate candidate square image patches with a side length of 128 pixels (approximately 2.56 meters in real-world conditions) centered on each SIFT feature point. Construct an evaluation system comprising the following three metrics to score each candidate patch on a 100-point scale: ① Local texture richness (weight 40%): Calculate the sum of the Sobel gradient magnitudes of all pixels within the block and normalize them. The higher the gradient value, the clearer the texture, the richer the edge information, and the stronger the uniqueness of the match.

[0028] ② Multi-frame feature stability pre-assessment (weight 40%): Perform rapid ORB feature matching between the candidate block's features in the baseline image and the image from week 2 (this step is lightweight and fast). Calculate the matching success rate (number of successfully matched feature pairs / number of feature points in the baseline block) and the average reprojection error of the matched point pairs. A higher success rate and smaller error indicate better short-term stability of the region, making it a strong predictive indicator for long-term reliable tracking.

[0029] ③ Spatial distribution rationality (weight 20%): Calculate the minimum Euclidean distance between the candidate block and the selected (temporary) best block. The larger the distance, the higher the score, to ensure that the feature blocks are evenly distributed on the slope surface, in the middle, lower and left and right regions, and to avoid excessive gaps in displacement field interpolation.

[0030] S23. Sort all candidate blocks according to the weighted total score. Set the total score threshold to 70 points. Select blocks sequentially from high score to low score, and check the spatial distance between each block and the selected blocks in real time. If the distance to any selected block is less than 300 pixels (to prevent overcrowding), skip it. Repeat until 50 "optimal tracking feature blocks" are selected. Store the coordinates, image content, and score files of these 50 blocks in the database.

[0031] S3, based on the image attributes of the optimal tracking feature block, adaptively derive the parameters for large-scale particle image velocimetry (LSPIV) calculation, the parameters including at least the cross-correlation query window size and the search area; ① The derivation of the cross-correlation query window size is based on analyzing the texture scale of each optimal block. The contrast value of the gray-level co-occurrence matrix (GLCM) for each block is calculated. Blocks with high contrast values ​​have fine textures and drastic local variations, so a smaller query window (e.g., 64×64 pixels) is sufficient for accurate localization; blocks with low contrast values ​​have smooth or uniform textures, requiring a larger window (e.g., 128×128 pixels) to obtain sufficient matching information. Setting... Where C is the contrast of that block, and C_avg is the average contrast of all blocks. This formula dynamically maps the window size between approximately 64 and 128 pixels.

[0032] ② Method for estimating preliminary displacement: In order to predict the approximate position of these optimal blocks in the second week image in the third week image, the LK sparse optical flow method is used to calculate the preliminary displacement vector (du,dv) of these 50 blocks from the first week image to the second week image.

[0033] ③ Search Region Generation: For displacement calculations from week 2 to week 3, a large fixed search region (such as the traditional ±256 pixels) is no longer set. Instead, the prediction center is the position (x, y) of each block in the week 2 image, and the position (x+du, y+dv) pointed to by its initial displacement vector (du, dv). Considering that the displacement may accelerate or decelerate, a dynamic radius R containing uncertainty is set: α is the scaling factor (set to 1.5), and β is the minimum radius (set to 20 pixels to handle static areas). Finally, the search area for this block is defined as a square with a side length of 2R centered at the prediction center.

[0034] ④ Cross-correlation peak ratio threshold: Calculate the ratio of the primary peak to the secondary peak for each optimal block during week 1-2 matching. Statistically analyze the distribution of these ratios and take the 25th percentile as the global threshold for that week. If the cross-correlation peak ratio of a block is lower than this threshold, its calculation result will be marked as "low confidence".

[0035] ⑤ Spatial Consistency Filtering: After calculation, check the displacement vectors of all blocks. Use the RANSAC algorithm to fit a global affine motion model, and discard vectors that deviate from the model by more than 3 pixels of median absolute deviation (MAD) as outliers.

[0036] S4. Using the adaptively derived parameters, perform LSPIV calculation on the time-series image sequence to obtain the displacement vector on the optimal tracking feature block; ① Using the personalized parameters determined in step S3 (a unique query window and search area for each block), perform standardized cross-correlation calculations on the orthophotos of week 2 and week 3.

[0037] ② The accuracy of correlation peak localization is improved to the 0.1 pixel level by using a sub-pixel fitting algorithm (such as two-dimensional Gaussian surface fitting).

[0038] ③ The final output is the high-precision, high-confidence displacement vector (dx) of these 50 optimal blocks in two time periods (weeks 2-3). i ,dy i ), where i = (1, 2, 3... 50).

[0039] S5, Based on the displacement vector on the optimal tracking feature block, generate the full-field displacement field of the slope monitoring area through spatial interpolation; Specifically, the spatial interpolation in step S5 is the Kriging interpolation method. Using the displacement vectors of 50 optimal blocks as observed data, and considering the spatial continuity of slope displacement, a spherical model is selected as the semi-variogram function to interpolate and generate a two-dimensional displacement field on the entire projection plane, as shown below. Figure 2 As shown.

[0040] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented, in whole or in part, as a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape) or an optical medium.

[0041] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A slope displacement monitoring method based on optimal feature block adaptive identification, characterized in that, Includes the following steps: S1, acquire the time-series image sequence of the slope monitoring area; S2, automatically identify multiple optimal tracking feature blocks on the current frame image in the image sequence; S3, Based on the image attributes of the optimal tracking feature block, adaptively derive the parameters used for large-scale particle image velocimetry (LSPIV) calculation; S4. Using the adaptively derived parameters, perform LSPIV calculation on the time-series image sequence to obtain the displacement vector on the optimal tracking feature block; S5. Based on the displacement vector on the optimal tracking feature block, the full-field displacement field of the slope monitoring area is generated by spatial interpolation.

2. The slope displacement monitoring method based on optimal feature block adaptive identification according to claim 1, characterized in that, The step S2, which automatically identifies the optimal tracking feature block, includes: S21, Extract feature points or feature regions from the current frame image; S22, score the candidate block centered on the feature point or feature region based on the preset evaluation index system; S23, select a predetermined number of candidate blocks as the optimal tracking feature blocks based on the scoring results.

3. The slope displacement monitoring method based on optimal feature block adaptive identification according to claim 2, characterized in that, The evaluation index system in S22 includes local texture richness, multi-frame feature stability, and spatial distribution rationality.

4. The slope displacement monitoring method based on optimal feature block adaptive identification according to claim 1, characterized in that, The parameters in step S3 include at least the cross-correlation query window size and the search area.

5. The slope displacement monitoring method based on optimal feature block adaptive identification according to claim 4, characterized in that, In step S3, the parameters are adaptively derived based on the optimal tracking feature block: The derivation of the cross-correlation query window size is based on the texture scale of each optimal tracking feature block; The derivation process of the search region is as follows: First, estimate the initial displacement of the optimal tracking feature block between adjacent frames, and then determine a dynamic search region range centered on the target position pointed to by the initial displacement.

6. The slope displacement monitoring method based on optimal feature block adaptive identification according to claim 5, characterized in that, The method for estimating the initial displacement is either the sparse optical flow method or the fast feature matching method.

7. The slope displacement monitoring method based on optimal feature block adaptive identification according to claim 1, characterized in that, The spatial interpolation in step S5 is either Kriging interpolation or thin-plate spline interpolation.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the method as described in any one of claims 1-7.

9. A computer device, characterized in that, The computer device includes a memory, a processor, and a program stored in and executable on the memory, the program being executed by the processor to implement the steps of the method as described in any one of claims 1-7.

Images