CN-122023833-A - Method for determining tunnel face support parameters based on RGB-D fusion

CN122023833ACN 122023833 ACN122023833 ACN 122023833ACN-122023833-A

Abstract

The invention discloses a tunnel face support parameter determination method based on RGB-D fusion, which comprises the steps of inputting an RGB image and a depth map of a tunnel face into an RGB-D double-flow feature fusion network, outputting a structural face area semantic mask and a confidence map, throwing a pixel set corresponding to the structural face semantic mask back into a point cloud space to obtain a structural face three-dimensional point set, carrying out plane robust fitting on the structural face three-dimensional point set to obtain a structural face fitting plane, obtaining structural face attitude parameter data, distributing a virtual measuring line set in a three-dimensional structural face model, calculating RQD values of each virtual measuring line and RQD statistical total values of a plurality of virtual measuring lines, carrying out stability evaluation on surrounding rocks based on the RQD statistical total values and the structural face attitude parameter data, and obtaining a support parameter recommendation interval matched with a surrounding rock stability evaluation result based on a pre-constructed support parameter recommendation table. The invention can realize the integration of surrounding rock classification and support parameter output.

Inventors

MA SHUQI
HOU JUNHONG
WANG XIANGYU
LIU HONGLIN
WANG MENG
CHEN JIAZHENG

Assignees

安徽理工大学

Dates

Publication Date: 20260512
Application Date: 20260206

Claims (10)

1. The method for determining the tunnel face support parameters based on RGB-D fusion is characterized by comprising the following steps: Synchronously acquiring RGB image and three-dimensional point cloud data of a tunnel face, generating a depth map D aligned with the pixel level of the RGB image, and establishing a corresponding relation of the RGB image, the depth map D and the three-dimensional point cloud under the same coordinate system; inputting the RGB image and the depth map into a pre-trained RGB-D double-flow feature fusion network, and outputting a structural plane area semantic mask and a confidence map; The method comprises the steps of (1) projecting a pixel set corresponding to a semantic mask of a structural plane area back to a point cloud space to obtain a structural plane three-dimensional point set, carrying out plane robust fitting treatment on the structural plane three-dimensional point set to obtain a structural plane fitting plane, and obtaining structural plane attitude parameter data based on the fitting plane; virtual line sets are laid in the reconstructed three-dimensional structural surface model, RQD values of all the virtual lines are calculated, and RQD statistical total values of a plurality of virtual lines are obtained through statistics; Performing stability evaluation on surrounding rock based on the obtained RQD statistical summary value and structural surface attitude parameter data; Based on a pre-constructed support parameter recommendation table, a support parameter recommendation interval matched with the surrounding rock stability evaluation result is obtained, the support parameter recommendation interval is taken as a candidate, and the optimal support parameter combination is obtained through optimization solution.
2. The method for determining tunnel face support parameters based on RGB-D fusion as set forth in claim 1, wherein the constructing of the RGB-D dual-flow feature fusion network specifically comprises: RGB coding branch For inputting RGB images Extracting multi-scale texture semantic features ; Depth coding branch For depth map of input Extracting multi-scale geometric features ; Cross-modal interaction module For applying, at least one scale layer And (3) with Alignment and fusion are carried out to obtain fusion characteristics ; Decoding branch For upsampling the fusion features and skip-fusing the fusion features with the coding features to obtain decoding features ; Output head For being based on Generating structural planar region semantic masks ; For being based on Generating a confidence map Wherein The degree of confidence that the pixel belongs to the structural plane area is characterized.
3. The method for determining tunnel face support parameters based on RGB-D fusion as set forth in claim 1, wherein training the RGB-D dual-flow feature fusion network comprises: Collecting multiple groups of RGB images I and pixel level alignment depth maps on tunnel face And marking the truth value mask of the structural plane area by manual or semi-automatic mode Form training sample ; After data enhancement is carried out on the sample, the sample is input into an RGB-D double-flow characteristic fusion network for training; updating network parameters through back propagation iteration until the verification set index converges to obtain an RGB-D double-flow characteristic fusion network for structural face identification; In the online reasoning stage, the structural plane area semantic mask output by the RGB-D double-flow characteristic fusion network is subjected to And (3) performing post-treatment, including thresholding, connected domain screening, hole filling and small region removing, so as to obtain a structural plane area set.
4. The method for determining tunnel face support parameters based on RGB-D fusion according to claim 1, wherein the method is characterized by comprising the steps of throwing back a pixel set corresponding to a semantic mask of a structural face area into a point cloud space to obtain a structural face three-dimensional point set, carrying out plane robust fitting on the structural face three-dimensional point set to obtain a structural face fitting plane, and obtaining structural face attitude parameter data based on the fitting plane, and specifically comprises the following steps: Semantic mask for planar area of obtained structure Back-casting point cloud Obtaining the structure face point set And is opposite to Divided into a plurality of sub-point sets ; For each of Carrying out plane robust fitting by adopting RANSAC or weighted RANSAC to obtain a structural plane And its normal direction And extracting structural surface boundaries as finite patches based on point set projection, thereby Combining to form a three-dimensional structural surface model for intersecting the subsequent virtual measuring line with the structural surface; And calculating the trend, the tendency and the inclination angle of the structural surface according to the normal vector of the fitting plane, and counting the number of structural surface groups, the spacing between the structural surfaces and the extension scale parameters.
5. The method for determining tunnel face support parameters based on RGB-D fusion according to claim 1, wherein a virtual line set is laid in the reconstructed three-dimensional structure face model, and RQD values of each virtual line are calculated, specifically comprising: virtual survey lines are arranged in the projection range of the face according to preset intervals Uniformly distributed and at least comprises three directions, namely along the axial direction, the radial direction and the circumferential direction of the tunnel; Each measuring line Expressed as a three-dimensional straight line parameter equation , As the starting point of the measuring line, Measuring parameters as unit direction vector , The total length of the measuring line; To arbitrary survey line Calculating the structural surface of the three-dimensional structural surface model And according to the line parameters Ordering the intersection points to obtain an intersection point sequence The distance between adjacent intersections is defined as the equivalent core segment length of the survey line, namely: In the formula, Represent the first The k-th section equivalent core segment length on the strip measuring line; And (3) with Respectively represent the measuring lines The kth intersection point and the kth+1th intersection point obtained after the intersection with the three-dimensional structural surface model are sequenced from small to large according to the survey line parameter s; Representing the euclidean norm; wherein Is a measuring line The number of upper intersection points; Will be As a measuring line The physical meaning of the equivalent core segment length is that the equivalent core segment length obtained by dividing the virtual measuring line under the cutting action of the structural surface is used for simulating the whole core segment length statistical process in the drill core; Will satisfy Divided by the total length of the line Obtaining the measuring line A kind of electronic device In the formula, Represent the first Strip line Core sampling rate index; Is a measuring line Is a total length of (2); Is the effective segment threshold value, wherein Taking 1 when the condition is met, otherwise taking 0; Representation of the opposite line The segment lengths formed by all adjacent intersections are summed, wherein only the following conditions are satisfied Counting the segment length of the segment; For multiple measuring lines The calculated result is subjected to stability evaluation, and the stability evaluation index comprises variance Or confidence interval width Variance is calculated Or confidence interval width And correspond to threshold value 、 Comparing as convergence criterion, if the convergence criterion is not satisfied, increasing the number of the measuring lines or reducing the distance according to the self-adaptive encryption strategy Where the lay is preferentially encrypted in areas of high fracture density or large RQD fluctuations to converge the RQD results.
6. The method for determining tunnel face support parameters based on RGB-D fusion as claimed in claim 5, wherein the statistical obtaining of the RQD statistical summary values of the plurality of virtual lines comprises: setting n test lines in total, and taking the statistical total value of the results of the plurality of test lines by the total RQD, and obtaining the statistical total value through weighted averaging or arithmetic averaging; The calculation formula for weighted averaging is as follows: When the total length of each measuring line is the same or weighting is not needed, an arithmetic average calculation formula is adopted: Wherein RQD represents the statistical total value of the results of a plurality of measuring lines, n is the number of the measuring lines; is the first The total length of the strip line is used as a weighting coefficient; is the first Core sampling rate indexes obtained through calculation of the measuring lines.
7. The method for determining tunnel face support parameters based on RGB-D fusion as set forth in claim 1, wherein the stability evaluation of the surrounding rock is performed based on the obtained RQD statistical summary value and the structural face attitude parameter data, specifically including: determining a surrounding rock grade and an RMR score, wherein the RMR score is an alternative; wherein a preset rule function is adopted Determining the level of surrounding rock Wherein The method comprises the steps of providing an evaluation index vector, wherein the evaluation index vector comprises RQD statistical summary values, structural plane attitude parameter data and related indexes; A mapping model obtained by utilizing threshold segmentation rules, rule trees or training; The RMR score is obtained by summing the scores of the various terms according to a preset RMR score table/scoring rule, with the following formula: Wherein the method comprises the steps of Is the first The score of each sub-term is calculated, Is the number of the sub-items; Is obtained by the following steps of Input index of individual sub-items Performing interval mapping according to a preset scoring table to obtain a scoring score, or calculating according to a preset scoring rule function to obtain the scoring score, namely: Wherein, the Is the first Score mapping function for individual sub-items to be input into the index Mapping to corresponding sub-term scores , Implemented by presetting a scoring table or segment mapping rules The value range of (1) is divided into a plurality of intervals and corresponds to the score of the sub-item one by one when Outputting a corresponding score when the corresponding interval is fallen into; is the first The term entry includes RQD statistics summary values, structural plane attitude parameter data and related indexes.
8. The method for determining supporting parameters of tunnel face based on RGB-D fusion as set forth in claim 7, wherein the method for obtaining the supporting parameter recommendation interval matched with the surrounding rock stability evaluation result based on the pre-constructed supporting parameter recommendation table specifically comprises: Based on the determined level of surrounding rock And outputting a matched support parameter recommended section through a pre-constructed support parameter recommended table as an interval to which an alternative RMR score belongs, wherein the support parameters comprise the thickness of sprayed concrete Length of anchor rod Distance between anchor rods Spacing of steel arches Or the model of the steel arch, wherein the supporting parameter recommendation table is used for establishing the surrounding rock grade Or the mapping relation between the RMR scoring interval and the support parameter recommending interval, and two sets of support parameter recommending intervals of economy and conservation can be provided.
9. The method for determining the supporting parameters of the tunnel face based on RGB-D fusion as claimed in claim 8, wherein the optimal supporting parameter combination is obtained by optimizing and solving by taking the recommended section of the supporting parameters as a candidate, specifically comprising: And constructing a feasible region by using the support parameter recommendation interval given by the support parameter recommendation table, establishing an objective function with minimum cost/construction period, establishing deformation and deformation rate constraint, and obtaining an optimal support parameter combination meeting the deformation and deformation rate constraint by optimizing and solving in the feasible region.
10. The utility model provides a tunnel face support parameter determination system based on RGB-D fuses which characterized in that, the system includes: The data acquisition and alignment module is used for synchronously acquiring RGB image and three-dimensional point cloud data of the tunnel face, generating a depth map D aligned with the pixel level of the RGB image and establishing a corresponding relation of the RGB image, the depth map D and the three-dimensional point cloud under the same coordinate system; The structure surface facial recognition module is used for inputting the RGB image and the depth map into a pre-trained RGB-D double-flow feature fusion network and outputting a structure surface facial region semantic mask and a confidence map; The structural plane fitting module is used for projecting the pixel set corresponding to the semantic mask of the structural plane area back to the point cloud space to obtain a structural plane three-dimensional point set, carrying out plane robust fitting treatment on the structural plane three-dimensional point set to obtain a structural plane fitting plane, and acquiring structural plane attitude parameter data based on the fitting plane; the RQD value statistics module is used for laying a virtual line set in the reconstructed three-dimensional structural surface model, calculating RQD values of each virtual line and obtaining RQD statistics total values of a plurality of virtual lines through statistics; The stability evaluation module is used for evaluating the stability of the surrounding rock based on the obtained RQD statistical summary value and the structural surface attitude parameter data; The support parameter recommendation module is used for acquiring a support parameter recommendation interval matched with the surrounding rock stability evaluation result based on a pre-constructed support parameter recommendation table, taking the support parameter recommendation interval as a candidate, and obtaining an optimal support parameter combination through optimization solution.

Description

Method for determining tunnel face support parameters based on RGB-D fusion Technical Field The invention relates to the technical field of surrounding rock control and face digital survey in tunnel engineering, in particular to a method for determining tunnel face supporting parameters based on RGB-D fusion. Background The classification of tunnel surrounding rock and the determination of supporting parameters usually depend on manual compiling, manual survey and experience interpretation of a tunnel face, and are obviously affected by light reflection, water stain dust and personnel difference, and have low efficiency and insufficient consistency. The existing partial technology can realize the crack trace identification or point cloud crack extraction of the face image, but the existing partial technology is mostly remained on the identification/display layer, so that a three-dimensional model of a structural face and parameters such as the occurrence, the spacing and the like which can be directly used for engineering evaluation are difficult to form. On the other hand, the quality indexes of the rock mass such as RQD/RMR have universality in engineering, but the traditional calculation depends on manual survey lines or a small number of drilling holes, is difficult to link with the digital data of the face, has large fluctuation of the evaluation result, and is difficult to be directly converted into executable suggestions of supporting parameters such as shotcrete, anchor rods, steel arches and the like. Disclosure of Invention The invention provides a tunnel face support parameter determination method based on RGB-D fusion, which aims to solve the problems that face surrounding rock evaluation is high in subjectivity, RQD calculation is not repeatable, support parameters are difficult to direct guide in classification in the prior art, and integration of surrounding rock classification and support parameter output is realized. According to a first aspect, in one embodiment, a method for determining tunnel face support parameters based on RGB-D fusion is provided, where the method includes: Synchronously acquiring RGB image and three-dimensional point cloud data of a tunnel face, generating a depth map D aligned with the pixel level of the RGB image, and establishing a corresponding relation of the RGB image, the depth map D and the three-dimensional point cloud under the same coordinate system; inputting the RGB image and the depth map into a pre-trained RGB-D double-flow feature fusion network, and outputting a structural plane area semantic mask and a confidence map; The method comprises the steps of (1) projecting a pixel set corresponding to a semantic mask of a structural plane area back to a point cloud space to obtain a structural plane three-dimensional point set, carrying out plane robust fitting treatment on the structural plane three-dimensional point set to obtain a structural plane fitting plane, and obtaining structural plane attitude parameter data based on the fitting plane; virtual line sets are laid in the reconstructed three-dimensional structural surface model, RQD values of all the virtual lines are calculated, and RQD statistical total values of a plurality of virtual lines are obtained through statistics; Performing stability evaluation on surrounding rock based on the obtained RQD statistical summary value and structural surface attitude parameter data; Based on a pre-constructed support parameter recommendation table, a support parameter recommendation interval matched with the surrounding rock stability evaluation result is obtained, the support parameter recommendation interval is taken as a candidate, and the optimal support parameter combination is obtained through optimization solution. Further, an RGB-D double-flow characteristic fusion network is constructed, which comprises the following steps: RGB coding branch For inputting RGB imagesExtracting multi-scale texture semantic features; Depth coding branchFor depth map of inputExtracting multi-scale geometric features; Cross-modal interaction moduleFor applying, at least one scale layerAnd (3) withAlignment and fusion are carried out to obtain fusion characteristics; Decoding branchFor upsampling the fusion features and skip-fusing the fusion features with the coding features to obtain decoding features; Output headFor being based onGenerating structural planar region semantic masks; For being based onGenerating a confidence mapWhereinThe degree of confidence that the pixel belongs to the structural plane area is characterized. Further, training the RGB-D dual stream feature fusion network, comprising: Collecting multiple groups of RGB images I and pixel level alignment depth maps on tunnel face And marking the truth value mask of the structural plane area by manual or semi-automatic modeForm training sample; After data enhancement is carried out on the sample, the sample is input into an RGB-D double-flow characteristi