Concrete defect segmentation and extraction method based on RGB-geometric information weighting

Abstract

The invention discloses a concrete defect segmentation and extraction method based on RGB-geometric information weighting, and relates to the field of defect detection of dam galleries. According to the method, color and geometric information are comprehensively utilized, and the accuracy and efficiency of dam concrete defect extraction are remarkably improved. More scientific and reasonable information fusion is realized by adaptively adjusting weights of colors and geometric information; a self-adaptive region growing strategy is adopted, dam concrete defects of different types, different scales and different distribution characteristics can be better adapted, and powerful support is provided for safety assessment and maintenance decision making of dams.

Description

Technical Field

[0001] The present invention relates to the field of defect detection of dam corridors, and specifically relates to a method for segmenting and extracting concrete defects based on weighted RGB-geometric information. Background Art

[0002] The statements in this section only provide background information related to the present disclosure and may not constitute prior art.

[0003] As a key infrastructure in water conservancy projects, the safe and stable operation of dams is crucial for ensuring the safety of people's lives and property and promoting regional economic development. As the main building material constituting the main body of the dam, the integrity and durability of the concrete structure directly affect the overall safety of the dam. However, during the construction and long-term operation of dam concrete, due to the influence of many complex factors, defects are very likely to occur. Timely and accurately detecting these dam concrete defects is of great significance for formulating scientific and effective repair and reinforcement measures and ensuring the safe and stable operation of the dam.

[0004] Traditional methods for detecting dam concrete defects mainly include visual inspection, ultrasonic testing, rebound method, etc. Visual inspection mainly relies on manual visual observation, which not only has low detection efficiency, but also the detection results are greatly affected by the subjective factors of the inspectors, and it is very difficult to detect some internal or relatively subtle defects. Although ultrasonic testing and the rebound method can detect internal defects to a certain extent, there are problems such as limited detection range and inaccurate judgment of complex defects. With the rapid development of computer vision and three-dimensional measurement technologies, methods for detecting dam concrete defects based on image or point cloud data have gradually become the forefront hotspots of research. Color and geometric information are extremely important features in the surface images or point cloud data of dam concrete, and they can effectively reflect the material characteristics and structural states of dam concrete. However, most current related detection methods only use color or geometric information alone, and fail to fully explore and utilize the complementary advantages of the two, resulting in the accuracy and reliability of dam concrete defect extraction being difficult to meet the actual engineering requirements. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for segmenting and extracting concrete defects based on weighted RGB-geometric information for the problems existing in the prior art, aiming to comprehensively and comprehensively utilize color and geometric information, and significantly improve the accuracy and efficiency of dam concrete defect extraction. By adaptively adjusting the weights of color and geometric information, more scientific and reasonable information fusion is achieved; an adaptive region growing strategy is adopted, which can better adapt to dam concrete defects of different types, different scales, and different distribution characteristics, providing strong support for the safety assessment and maintenance decision-making of dams.

[0006] The technical solution of the present invention is as follows: A method for segmenting and extracting concrete defects based on RGB-geometry information weighting, comprising: Step S1: Construct a spatial index for the acquired concrete point cloud data and perform noise filtering, and perform adaptive spatial division using an octree structure; Step S2: Extract geometric features and color features respectively in a local neighborhood, calculate the geometric deformation distance through robust plane fitting, and obtain color anomaly features based on RGB bin statistics; Step S3: Perform dynamic weight allocation through variance analysis of color and geometric features in a local window, and calculate a comprehensive score; Step S4: Use a double-threshold mechanism to screen high-confidence seed points, and perform region growing in combination with a density compensation factor; Step S5: Use a multi-resolution octree to implement hierarchical growth acceleration, and finally optimize the defect boundary through morphological closing operation and connected component area threshold.

[0007] Further, the geometric feature extraction in the step S2 specifically includes: Use the RANSAC algorithm to fit the reference plane of the local neighborhood, calculate the geometric distance from each point to the reference plane, and standardize the geometric distance into geometric deformation features through a formula.

[0008] Further, the color feature extraction in the step S2 specifically includes: Perform bin statistics on the mode of the reference plane area by RGB channels, take the midpoint value of the bin with the highest frequency in each channel as the color feature of the non-defect area, and calculate the difference degree between the measured point and the reference color feature through the Euclidean distance formula.

[0009] Further, the dynamic weight allocation in the step S3 specifically includes: Calculate the variance of color features and the variance of geometric features respectively in a local window, and perform weight normalization through the variance ratio.

[0010] Further, the seed point screening in the step S4 adopts a double-threshold mechanism: It is required that the comprehensive score of the candidate point is greater than the high-confidence threshold, and at the same time, it satisfies that any single-modal score of the color feature or geometric feature exceeds its corresponding threshold.

[0011] Further, the density compensation factor in the step S4 is specifically implemented as: Calculate the dynamic growth threshold according to the local density of the point to be grown.

[0012] Further, the hierarchical growth acceleration in the step S5 includes: Perform coarse-layer downsampling growth on the shallow nodes of the octree, perform fine-layer refinement growth on the deep nodes, and quickly cover the defect area through the center point of the bounding box and then restore the original accuracy.

[0013] Further, the morphological closing operation in step S5 specifically includes: First, perform a dilation operation on the segmented defect area to connect adjacent areas, and then perform an erosion operation to restore the boundary shape.

[0014] Further, it also includes: a defect area optimization step; The defect area optimization step removes isolated areas with an area smaller than a set threshold and abnormal depth based on the connected domain area threshold and the maximum point cloud depth.

[0015] Further, the spatial index construction specifically includes: Use an octree for multi-scale hierarchical partitioning, and remove outliers through local plane fitting residuals during the noise filtering stage.

[0016] Compared with the existing technology, the beneficial effects of the present invention are: The present invention effectively measures color and geometric information, and then adaptively adjusts the bimodal contribution weights of the feature information; uses an energy-driven strategy to convert the bimodal contribution weights into an energy comprehensive score, and then uses the energy score for region growing to obtain accurately segmented concrete defect data. Experimental results show that this method can significantly improve the accuracy and reliability of concrete defect extraction and has good engineering application prospects. Description of the Drawings

[0017] Figure 1 It is a principle block diagram of a concrete defect segmentation and extraction method based on RGB-geometric information weighting. Detailed Embodiments

[0018] It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variation thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0019] The features and performance of the present invention will be further described in detail below in conjunction with the embodiments.

[0020] Embodiment 1 Based on the concrete point cloud data of the dam corridor collected by a binocular structured light camera, aiming at the problem that it is difficult to segment and extract the defective areas of the dam corridor concrete, a method for segmenting and extracting concrete defects based on the weighted RGB-geometry information is proposed. Its main framework is as shown in Figure 1 shown below, and the main content is as follows: 1. Aiming at the complex surface characteristics of the dam corridor, first construct a spatial index and filter noise for the original point cloud. Establish a multi-scale hierarchical structure through octree adaptive partitioning. In the local neighborhood, use a robust plane fitting algorithm to calculate the normalized deformation distance of each point relative to the fitting reference plane, and quantify the geometric abnormality degree of the surface concave and convex defects. Synchronously extract color modality features: Based on the RGB color change information of the defective area and non-defective area of the dam corridor concrete, effectively quantify the color abnormality degree of the concrete defective area.

[0021] 2. Rely on the geometric abnormality degree in the previous step and the typical color characteristics of the dam corridor concrete defects to design a dynamic weight allocation strategy. Through the variance analysis of color and geometric features within a local window, adaptively adjust the bimodal contribution weights: strengthen color features in areas with stable illumination, and rely on geometric deformation indicators in shaded or reflective areas.

[0022] 3. According to the adaptive adjustment of the bimodal contribution weights in the previous step, implement precise segmentation of the defective area with an energy-driven strategy. First, screen high-confidence seed points through a double-threshold mechanism, and combine the point cloud density distribution characteristics (the density in the crack area is usually low). Introduce a density compensation factor during the region growth process, and dynamically adjust the growth threshold to suppress the mismerging of cracks. Use a multi-resolution octree to accelerate the growth process, layer by layer expand the defective boundary until the energy converges. Finally, fill the holes through morphological closing operations, and filter out small artifacts based on the connected domain area threshold, and output the defective vector boundary that meets the engineering specifications. The whole process takes into account both algorithm efficiency and accuracy, and can adapt to the dam corridor structure and the characteristics of point cloud data with an accuracy of 0.3 - 0.5 mm.

[0023] In this embodiment, specifically, a method for segmenting and extracting concrete defects based on the weighted RGB-geometry information specifically includes the following steps: Step S1: Construct a spatial index and filter noise for the obtained concrete point cloud data, and perform adaptive spatial partitioning using an octree structure; that is, use an octree structure to perform adaptive spatial partitioning on the input point cloud, and extract color features and geometric features from the processed point cloud data respectively in the subsequent steps; Step S2: Extract geometric features and color features respectively in the local neighborhood, calculate the geometric deformation distance through robust plane fitting, and obtain the color abnormality features based on RGB binning statistics; Step S3: Perform dynamic weight allocation through the variance analysis of color and geometric features within a local window, and calculate the comprehensive score; Step S4: Use a double-threshold mechanism to screen high-confidence seed points and perform region growing in combination with a density compensation factor; Step S5: Use a multi-resolution octree to implement hierarchical growth acceleration, and finally optimize the defect boundary through morphological closing operation and connected domain area threshold.

[0024] In this actual example, specifically, the geometric feature extraction in step S2 specifically includes: Use the RANSAC algorithm to fit the reference plane of the local neighborhood, calculate the geometric distance from each point to the reference plane, and standardize the geometric distance into geometric deformation features through a formula.

[0025] That is, perform the best plane or surface fitting on the input dam gallery concrete point cloud data. Since the data collection area is a local small area when collecting the dam gallery point cloud data, the data can also be regarded as a plane when collecting non-planar areas.

[0026] Use RANSAC to fit the best robust plane of the point cloud data , let the deformation point to be calculated be , calculate 's geometric distance to the plane , as follows:

[0027] Select the neighborhood of the deformation point , assume the geometric distance set within this neighborhood , and finally standardize the geometric distance into geometric deformation features , as follows:

[0028] In the formula: , is a non-zero small constant.

[0029] In this actual example, specifically, the color feature extraction in step S2 specifically includes: Bin and count the mode of the reference plane area by RGB channels, take the midpoint value of the bin with the highest frequency in each channel as the color feature of the non-defect area, and calculate the difference degree between the measured point and the reference color feature through the Euclidean distance formula.

[0030] It should be noted that from the collected point cloud data of the dam corridor, it can be obtained that the RGB values of the point cloud in the defect area and the RGB values of the surrounding non-defect area will change significantly. Generally, the RGB values of the point cloud in the defect area are relatively small, while those in the non-curved area are the opposite. Based on this conclusion, color features are added as one of the conditions for defect segmentation and extraction in this paper.

[0031] In the dam corridor environment, the lighting conditions are complex and variable, and the point clouds at different positions are affected by light to different degrees. Due to the change of the surface structure in the defect area, the reflection and absorption characteristics of light are different from those in the non-defect area, resulting in obvious changes in RGB values. Usually, the phenomenon of brightness attenuation will occur in the defect area. For the best-fitted plane in the extracted geometric features, assume the geometry of the points on the fitted plane is , where the point , bin and count the mode according to the RGB channels. The number of bins for each channel is , and the bin width is . Count the three bins with the highest frequency in each channel, and take the midpoint value as the color feature of the non-defect area of the dam corridor concrete .

[0032] Let the set of points to be measured be , and the point with abnormal color feature to be measured be . Use the Euclidean distance formula to calculate the color difference between the point and as follows. The color anomaly feature calculation of the point is :

[0033] In this embodiment, specifically, the dynamic weight allocation in step S3 specifically includes: Calculate the variance of color features and the variance of geometric features respectively within the local window, and perform weight normalization through the variance ratio.

[0034] That is, because the dam corridor environment is relatively complex and the confidence levels of color and geometric features are significantly different in different regions, in order to adaptively balance the reliability of color features and geometric features in different local regions, by quantifying the local consistency of the two features, the contribution weights of color features and geometric features are automatically adjusted to achieve dynamic weights and ensure accurate defect recognition in different environments.

[0035] Let the variance of color features be . Within the selected local neighborhood , the color feature The smaller the variance, the stronger the color consistency within the local neighborhood, and the higher the credibility of the color feature; conversely, the confidence level is low. Similarly for the geometric feature variance, let the geometric feature variance be , when within the local neighborhood is smaller, it indicates that the geometric surface is flat, and the credibility of the geometric feature decreases; conversely, the credibility is high. Based on the characteristics that different feature methods will bring different credibilities, the weight difference between features is changed according to the magnitudes of different feature methods within the local neighborhood. The normalization calculation of the color feature and geometric feature weights is as follows, is a non-zero small constant:

[0036] Based on the above dynamic weights, the comprehensive score of the point to be measured is as follows:

[0037] In this embodiment, specifically, in step S4, the seed point screening adopts a double-threshold mechanism: It is required that the comprehensive score of the candidate point is greater than the high-confidence threshold, and at the same time, either the single-modal score of the color feature or the geometric feature exceeds its corresponding threshold.

[0038] That is, after the dynamic weight allocation is realized, the probability that each point belongs to the defect area is quantified by defining an energy function, and by integrating the region growing and density compensation strategies, the defect boundary is gradually expanded until the energy is stable.

[0039] To screen out the seed points most likely to belong to the defect from the point cloud as the starting point of region growing, and from the dynamic weights of different features, it can be known that the seed points must satisfy that the comprehensive score is greater than a high-confidence threshold. Assuming the comprehensive score of this point is , then to ensure that the seed points have significant abnormality, it is required that , where is the high-confidence threshold, taking points above 95% of the scores of all points. At the same time, the color feature or the geometric feature respectively satisfies its corresponding threshold. This double-threshold design can prevent misjudgment of a single modality.

[0040] In this embodiment, specifically, the density compensation factor in step S4 is specifically implemented as: Calculate the dynamic growth threshold according to the local density of the point to be grown.

[0041] That is, after the growth point is selected, the condition for in the neighborhood to be selected and merged into the current defect area is as follows, where is the growth threshold, which is used to control the looseness of growth. And through the local density for Make corrections to prevent small defect areas from not truly growing.

[0042]

[0043] Set the point to be grown , calculate the radius the number of points within the domain , obtain the local density maximum value through calculation , based on this value, obtain the dynamic growth threshold :

[0044] where, is the density compensation intensity coefficient, generally 0.5, is the originally set growth threshold. By repeating the above steps, new growth points can be added to the growth queue until no new points are added.

[0045] In this embodiment, specifically, the hierarchical growth acceleration in step S5 includes: Perform coarse-layer downsampling growth on the shallow nodes of the octree, and perform fine-layer refinement growth on the deep nodes. After quickly covering the defect area through the center point of the bounding box, restore the original accuracy.

[0046] To achieve accelerated growth, utilize the multi-scale characteristics of the octree to set a hierarchical growth strategy: 1. Coarse-layer growth, perform regional growth on the shallow nodes of the octree, and the point cloud within the node is downsampled to the center point of the bounding box for quick coverage of the defect area; 2. Fine-layer refinement, perform growth on the deep nodes corresponding to the points in the defect area covered by the coarse-layer growth, and retain the original point cloud accuracy. In this way, the number of coarse-layer nodes is much less than the original point cloud, and the fine-layer fast neighborhood query is used to avoid global search.

[0047] In this embodiment, specifically, the morphological closing operation in step S5 specifically includes: First perform a dilation operation on the segmented defect area to connect adjacent areas, and then perform an erosion operation to restore the boundary shape.

[0048] In this embodiment, specifically, it further includes: a defect area optimization step; The defect area optimization step removes isolated areas with an area smaller than the set threshold and abnormal depth based on the connected domain area threshold and the maximum point cloud depth.

[0049] That is, morphological operations and abnormal area removal are performed on the growing defect areas in sequence. The small cavities of the defects are filled using morphological closing operations, and adjacent areas are connected. Then, the over-filled areas are eroded to ensure the integrity of the edges of the defect areas. For the point clouds of isolated small defect areas, by taking the number of points and the maximum depth value of the points as references, when both the number of point clouds and the current maximum depth value are less than the set threshold, the defect areas are removed from the growing areas and regarded as the complete concrete surface.

[0050] In this embodiment, the construction of the spatial index specifically includes: An octree is used for multi-scale hierarchical division, and outlier points are removed through local plane fitting residuals during the noise filtering stage.

[0051] The above-described embodiments only represent the specific implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the protection scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the technical solution of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application.

[0052] This background art section is provided to generally present the context of the present invention. The work of the currently named inventors, to the extent described in this background art section, and aspects of the work described in this section that do not constitute prior art at the time of filing this application are neither expressly nor impliedly admitted to be prior art to the present invention.

Claims

1. A concrete defect segmentation and extraction method based on RGB-geometric information weighting, characterized in that: include: Step S1: construct a spatial index for the acquired concrete point cloud data and perform noise filtering, and use an octree structure for adaptive spatial division; Step S2: extracting geometric features and color features in the local neighborhood respectively, calculating the geometric deformation distance by robust plane fitting, and obtaining color anomaly features based on RGB bin statistics; Step S3: Perform dynamic weight allocation and calculate the comprehensive score through variance analysis of color and geometric features in the local window; Step S4: using a double threshold mechanism to screen high confidence seed points and combining density compensation factors to perform region growing; Step S5: Use multi-resolution octree to implement hierarchical growth acceleration, and finally optimize the defect boundary through morphological closing operation and connected domain area threshold.

2. The concrete defect segmentation and extraction method based on RGB-geometric information weighting according to claim 1 is characterized in that: The geometric feature extraction in step S2 specifically includes: The RANSAC algorithm is used to fit the reference plane of the local neighborhood, and the geometric distance from each point to the reference plane is calculated. The geometric distance is standardized into a geometric deformation feature through the formula.

3. The concrete defect segmentation and extraction method based on RGB-geometric information weighting according to claim 1 is characterized in that: The color feature extraction in step S2 specifically includes: The mode of the reference plane area is counted by RGB channel binning, and the midpoint value of the highest frequency box of each channel is taken as the color feature of the non-defective area. The difference between the test point and the reference color feature is calculated using the Euclidean distance formula.

4. The concrete defect segmentation and extraction method based on RGB-geometric information weighting according to claim 1 is characterized in that: The dynamic weight allocation in step S3 specifically includes: The color feature variance and geometric feature variance are calculated separately in the local window, and the weights are normalized by the variance ratio.

5. The concrete defect segmentation and extraction method based on RGB-geometric information weighting according to claim 1 is characterized in that: In step S4, seed point screening adopts a dual threshold mechanism: The comprehensive score of the candidate point is required to be greater than the high confidence threshold, and at the same time, the score of any single modality of color feature or geometric feature exceeds its corresponding threshold.

6. The concrete defect segmentation and extraction method based on RGB-geometric information weighting according to claim 1 is characterized in that: The density compensation factor in step S4 is specifically implemented as: The dynamic growth threshold is calculated according to the local density of the points to be grown.

7. The concrete defect segmentation and extraction method based on RGB-geometric information weighting according to claim 1 is characterized in that: The layered growth acceleration in step S5 includes: The coarse layer downsampling growth is performed on the shallow nodes of the octree, and the fine layer refinement growth is performed on the deep nodes. The original accuracy is restored after the defect area is quickly covered by the center point of the bounding box.

8. The concrete defect segmentation and extraction method based on RGB-geometric information weighting according to claim 1 is characterized in that: The morphological closing operation in step S5 specifically includes: The segmented defective area is first dilated to connect the adjacent areas, and then eroded to restore the boundary shape.

9. The method for concrete defect segmentation and extraction based on RGB-geometric information weighting according to claim 1, characterized in that: Also includes: Defective area optimization steps; The defect area optimization step removes isolated areas whose areas are smaller than the set threshold and whose depths are abnormal based on the connected domain area threshold and the maximum value of the point cloud depth.

10. The method for concrete defect segmentation and extraction based on RGB-geometric information weighting according to claim 1, characterized in that: The spatial index construction specifically includes: Octree is used for multi-scale hierarchical division, and outliers are removed through local plane fitting residuals in the noise filtering stage.

Images