Tunnel point cloud registration method fusing color and geometric features

Abstract

The invention relates to the technical field of point cloud registration and splicing, and discloses a rapid color point cloud feature histogram registration method. According to the technical scheme, a color point cloud to be registered is obtained, a calculation radius is set, and a point cloud normal is calculated; calculating geometric features of the point clouds, and obtaining key points of the source point clouds and the target point clouds; performing neighborhood search on the key points, calculating geometric and color features of the key points, and constructing a color fast feature histogram of the key points; performing feature matching to obtain an initial feature matching pair; obtaining an optimal transformation matrix by using a random sampling consistency algorithm; and the obtained optimal transformation matrix is used as an initial transformation matrix to complete coarse registration of the source point cloud and the target point cloud, and an iterative nearest point algorithm is used to complete fine registration. The method has the beneficial effects that through the steps, the purpose of point cloud fine registration can be achieved. The method has the advantages of high speed, small error, high efficiency and the like. And an important technical support is provided for dynamic data acquisition in a tunnel construction process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of point cloud registration splicing, in particular to a fast color point cloud feature histogram registration method, which is suitable for point cloud registration with color information and non-obvious geometric features, and is mainly applied to the registration of tunnel point clouds. BACKGROUND

[0002] In tunnel engineering, it is often difficult to detect detailed and accurate geological conditions through geological exploration. Therefore, during the tunneling process, it is necessary to quickly update the exposed geological conditions of excavation in order to guide the subsequent construction and support of the tunnel. Data acquisition based on laser scanners has been widely used in recent years due to its accuracy and efficiency. When processing the data collected by the scanner, the problem of point cloud registration and splicing often cannot be avoided. Common high-precision point cloud splicing requires placing a target ball in advance, maintaining the position of the target ball constant, erecting the next station and performing a laser scanning task, and completing the splicing of the point cloud data of the two stations by aligning the position of the target ball. However, under the condition of tunnel blasting and excavation, this method is difficult to implement. The time-consuming placement of the target ball, as well as the volume and shape of the target ball, also limit the flexibility of placement. Most importantly, the target ball cannot be guaranteed not to be affected by blasting during tunnel excavation. Therefore, in the task of tunnel point cloud splicing, the point cloud registration algorithm is particularly important.

[0003] A commonly used fine registration algorithm is the Iterative Closest Point (ICP) algorithm, which, however, relies on initial alignment and is prone to local optimization. Therefore, it is necessary to perform coarse registration of the point cloud before fine registration. In the coarse registration task, the Fast Point Feature Histogram (FPFH) algorithm performs well, but this algorithm only describes features based on geometric information, and in scenes with similar geometric structures, the FPFH algorithm cannot provide good registration capability. Therefore, a method is needed that can capture the diversified features of point clouds to accurately and efficiently register the point clouds. The present application is proposed based on this need. SUMMARY

[0004] To solve the technical problems existing in the prior art, the present application provides a tunnel point cloud registration method that fuses color and geometric features, which can accurately and efficiently register tunnel point clouds.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions: A tunnel point cloud registration method that fuses color and geometric features, comprising the following steps: S1, using a laser scanner equipped with a color perception module to obtain color point clouds to be registered, setting a calculation radius and calculating point cloud normals; S2, calculating point cloud geometric features and obtaining key points of the source point cloud and the target point cloud; S3, performing neighborhood search on the key points, calculating key point geometric and color features and constructing a color fast feature histogram of the key points; S4, performing feature matching according to the feature histogram of the key points to obtain an initial feature matching pair; S5, using a random sample consensus algorithm to obtain an optimal transformation matrix through random sampling and iterative optimization; S6, using the optimal transformation matrix as an initial transformation matrix to complete coarse registration of the source point cloud and the target point cloud, and finally using an iterative closest point algorithm to complete fine registration.

[0006] Further, the method for obtaining key points in step S2 is an intrinsic shape feature key point extraction method, including calculating a neighborhood covariance matrix of each point in the point cloud and its eigenvalue, and screening key points through eigenvalue ratio and non-maximum suppression.

[0007] Further, the construction of the color fast feature histogram of the key points in step S3 includes: calculating a geometric feature vector between the key point and the neighborhood point ; calculating a color feature vector of the key point and the neighborhood point ; tensor splicing the geometric feature vector and the color feature vector to form a six-dimensional joint feature vector .

[0008] Further, the color feature vector is obtained by converting the point cloud color from an RGB color space to a YIQ color space, wherein the Y component represents luminance information, and the I and Q components represent chroma information.

[0009] Further, the color fast feature histogram is a 57-dimensional feature histogram, and the geometric feature vector is divided into 11 bins, and the color feature vectors Y, I and Q are divided into 4, 10 and 10 bins respectively, and the feature histograms are combined.

[0010] Further, the feature matching in S4 uses a bidirectional nearest neighbor matching method, including: S41, forward matching: for each point in the source point cloud key point set P , find a candidate point set in the target point cloud key point set Q that satisfies , wherein F(·) represents the feature histogram of the point; S42, reverse verification: for each point in the candidate point set , verify whether it is a nearest neighbor; S43, retaining the point pairs with bidirectional consistency to form an initial matching set .

[0011] Further, the random sample consensus algorithm used in S5 includes: S51, randomly selecting matching pairs from the matching set , ensuring that the minimum distance between the points in the matching pairs is greater than the threshold value d; S52, estimating the rigid transformation matrix between the matching pairs using the coarse ICP, and calculating the Huber error between the matching pair sets;

[0012] Further, in step S6, the source point cloud is rigidly transformed using the best transformation matrix in the coarse registration, and the iterative closest point algorithm is used for iterative optimization in the fine registration to obtain the registered point cloud.

[0013] Further, in step S1, the calculation radius is set to 5-10 times the resolution of the point cloud.

[0014] Further, in step S3, constructing the color fast feature histogram of the key points further includes: For each key point, calculate the color fast feature histogram of its neighborhood points, and perform weighted summation according to the distance between the neighborhood points and the key point; normalizing the weighted and summed feature histogram to make the histogram sum a fixed value.

[0015] Advantages of the present application: Compared with the prior art, the tunnel point cloud registration method fusing color and geometric features has the following technical features and advantages: (1) Multi-modal feature fusion advantage: by tensor fusion of geometric spatial relationship and chrominance space features, a 57-dimensional composite feature histogram is constructed, effectively overcoming the matching ambiguity problem of traditional FPFH and other pure geometric descriptors in similar structure areas of tunnel rock mass. The introduction of color features significantly enhances the discriminability of the feature descriptor, especially in scenes with color differences such as rock layer interfaces and different lithology areas.

[0016] (2) Engineering environmental adaptability: The designed bidirectional nearest neighbor matching mechanism (S41-S43) combined with the reverse verification strategy can effectively suppress the false matching caused by local interference such as blasting debris, temporary support components, etc. in the tunnel point cloud. The dynamic threshold adjustment strategy can adapt to the changes of point cloud density in different sections, ensuring stable matching of the point cloud.

[0017] (3) Registration process optimization: The two-stage architecture of "color enhancement coarse registration + ICP fine registration" is adopted, which not only retains the global convergence of feature matching, but also inherits the high precision characteristics of ICP algorithm. Compared with the traditional target ball registration method, this scheme does not need to lay physical markers on site, and the single station registration preparation time is shortened by about 85%, which is especially suitable for the working condition that needs to be scanned quickly after blasting.

[0018] (4) Geological engineering applicability: The feature extraction method combined with YIQ color space is robust to common illumination interference such as water leakage and rock powder adhesion in tunnels.

[0019] The present application provides a tunnel point cloud registration method combining color and geometric features, which effectively solves the problem of point cloud registration in the blasting construction environment through an adaptive registration method that combines geometric features and color information, and realizes high-precision and fast splicing of complex rock mass structure. The target-free, anti-interference and high-efficiency technical characteristics greatly improve the real-time updating ability of geological information, provide a reliable point cloud data processing foundation for tunnel intelligent construction, reduce the construction safety risk and operation and maintenance cost, and promote the development of underground engineering survey technology towards automation and intelligence. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the present application will be described in detail below in combination with the drawings and detailed embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor. Among them: Figure 1 The technical solution flowchart of the present application; Figure 2(a) is a photograph of the tunnel side spraying in the field of embodiment 1 of the present application; Figure 2(b) is a photograph of the field of embodiment 1 of the present application; Figure 3 The key point extraction result graph in embodiment 1 of the present application; Figure 4 The schematic diagram of the multi-dimensional joint feature calculation method of the present application; Figure 5 The schematic diagram of the geometric feature calculation coordinate system establishment method of the present application; Figure 6This refers to the multidimensional feature histogram established in Embodiment 1 of the present invention; Figure 7 (a) is the cloud map of the points to be registered in Embodiment 1 of the present invention; Figure 7 (b) is a registration result diagram in Embodiment 1 of the present invention; Figure 7 (c) A magnified view of the registration result of the conventional method in Example 1; Figure 7 (d) is a magnified view of the registration result of the present invention in Example 1. Detailed Implementation

[0021] 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 and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The following description, in conjunction with the accompanying drawings... Figures 1 to 7 The method for registering tunnel point clouds by integrating color and geometric features is further explained.

[0022] Example 1 This invention provides a method for tunnel point cloud registration that integrates color and geometric features, the technical approach of which is as follows: Figure 1 As shown, the specific steps include: S1: Use a laser scanner equipped with a color perception module to acquire the color point cloud to be registered, set the calculation radius, and calculate the point cloud normal.

[0023] S2 calculates the geometric features of the point cloud and obtains the key points of the source and target point clouds.

[0024] S3 performs a neighborhood search on the keypoints, calculates the geometric and color features of the keypoints, and constructs a fast color feature histogram for the keypoints.

[0025] S4. Perform feature matching based on the feature histogram of key points to obtain initial feature matching pairs.

[0026] S5 uses the RANSAC (Random Sample Consensus) algorithm to obtain the optimal transformation matrix through random sampling and iterative optimization, so that more matching pairs have an error less than the threshold after transformation.

[0027] S6. The obtained optimal transformation matrix is ​​used as the initial transformation matrix to complete the coarse registration of the source point cloud and the target point cloud. Finally, the Iterative Closest Point Algorithm (ICP) is used to complete the fine registration.

[0028] To make the color features in this invention more obvious, markings can be sprayed on the tunnel sidewalls in advance to improve the registration rate (Figure 2(a)). These markings are not affected by tunnel blasting and are easy to operate. As shown in Figure 2(b), a laser scanner equipped with a color perception module is used to perform at least two station scans in the tunnel. The point clouds of the two station scans must have overlapping areas. The higher the overlap, the better the stitching effect. It is necessary to balance and adjust according to the actual situation. For the obtained point clouds of the two stations, filtering, downsampling and other preprocessing operations are performed. Then, the normal vector calculation radius is set. It is recommended to set the radius to 5-10 times the resolution size. In this embodiment, the calculation radius is set to 10cm. The kd-tree algorithm is used to search the point cloud within a 10cm neighborhood of each point. For the target point The set of points in the neighborhood of (Total of n points), calculate its centroid (mean): (1) in Let represent each point in the neighborhood. Construct a 3×3 covariance matrix using the deviations of each point in the neighborhood relative to the centroid: (2) Here, each The covariance matrix is ​​a column vector, and its outer product yields a 3×3 matrix. This matrix reflects the dispersion of points in the neighborhood across various directions. Eigenvalue decomposition is performed on the covariance matrix , and the eigenvalues ​​and corresponding eigenvectors are obtained: (3) The three eigenvalues ​​are obtained and sorted from smallest to largest, satisfying the following: (4) Fit the point cloud to a plane and calculate the plane's normal vector. Since the variance of a local point set is minimized in the direction of the normal vector, it corresponds to the minimum eigenvalue. eigenvectors This is the estimated normal vector: (5) Normalize the direction of the point cloud normal vectors to ensure that the normal vectors satisfy the following conditions: (6) Where O is the coordinate of the origin, otherwise make .

[0029] Preferably, the eigenvalue ratio is calculated: (7) Set two thresholds and These are used to determine the eigenvalue ratio of the local covariance matrix, and only when... and When a point is considered sufficiently salient in its local structure, it is selected as a candidate keypoint. To avoid an overly dense pool of candidate keypoints, non-maximum suppression (NSS) is applied to the candidate set. The salientity of each candidate point within its local region is compared, and only the point with the highest salientity within that region is retained. After NSS, the remaining points are the final ISS keypoints. The extracted ISS keypoints in this embodiment are as follows: Figure 3 As shown.

[0030] For each keypoint, search for other points in the keypoint's neighborhood and calculate the multidimensional features relating the neighboring points to the keypoint. The relationships are calculated as follows: Figure 4 As shown, a local coordinate system is established between the key points and the neighboring points to be calculated. The method for determining the three-axis orientation is as follows: Figure 5 As shown.

[0031] (8) (9) (10) in , These represent key points and neighborhood points, respectively. After establishing a local coordinate system, the PFH feature representation is calculated: (11) (12) (13) in Let represent the normal vector of the neighboring points. The geometric feature vectors of the positional relationship are obtained through formulas (11)-(13). .

[0032] Furthermore, the color space feature vectors of keypoints and their neighbors are calculated. Since lighting conditions within the tunnel are affected by various factors, the impact of brightness variations in the color features should be reduced. The RGB color space clearly represents the three elements, but it is insufficient for representing the brightness component. Therefore, it needs to be converted to the YIQ color space. In the YIQ color space, the Y component represents brightness information, the I component represents the orange-blue related component in the hue information, and the Q component represents the green-purple related component in the hue information. The RGB to YIQ conversion formula is: (14) After converting the colors of neighboring points to YIQ representations, the color features of the three components of the neighboring points are calculated to obtain the color feature vector. Geometric features and color features are concatenated using tensors to form a six-dimensional joint feature vector. .

[0033] Will Each region is defined as a 30-degree interval, thus classifying each geometric feature vector into an 11-dimensional feature vector. The YIQ value ranges are as follows: The data was divided into 4, 10, and 10 bins respectively. To reduce the weight of the luminance feature component and decrease the discriminative sensitivity of the Y component, only four partitions were set. The number of bins for each of the six features was counted to decouple the features and simplify the feature histograms. The feature histograms of the independent features were merged to obtain the Tunnel Colored Simplified Point Feature Histogram (TCSPFH). In this embodiment, the feature histogram of a certain key point is shown below. Figure 6 As shown. This histogram has a total of 57 dimensions, which are high-dimensional feature vectors that can accurately describe the geometric and color features of the point cloud. The neighboring points are set as query points, and the steps for building TCSPFH are repeated. Figure 4 The TCSPFH weights of the neighboring points are added to the neighborhood feature vector of the keypoint: (15) in Indicates query point The tunnel color point cloud feature histogram, where k represents the number of neighboring points. For query point With neighboring points The distance between them. This ultimately yields the key points. A complete histogram of the tunnel color point cloud features is generated. Histogram normalization is then performed; since there are 6 features, the sum of the histogram values ​​is set to 600. The value on each bin represents the normalized frequency of the points within the corresponding range of that bin. After normalization, the sum of the 57 bins is 600.

[0034] Furthermore, after obtaining the tunnel key points TCFPFH, a bidirectional nearest neighbor matching is performed on the source point cloud key point set P and the target point cloud key point set Q. The specific steps are as follows: Forward matching: for each Search in Q for a set of candidate points that satisfy equation (18). : (16) (17) in Point The feature histogram. express The value of the i-th bin in the feature histogram.

[0035] Reverse verification: verify if is the nearest neighbor of in P.

[0036] Keep the pairs of points that are bidirectional consistent to form the initial matching set .

[0037] Preferably, randomly select matching pairs from the matching set , while to avoid the amplification of registration error caused by the dense matching points, it should be ensured that the minimum distance between each other is greater than the set threshold d. Estimate the rigid transformation matrix between using the coarse ICP and calculate the Huber error between the matching pair set: (18) where denotes the mean square error MSE: (19) is the position of in the target point coordinate system after the action of the transformation matrix T.

[0038] Further, save the Huber error and the corresponding transformation matrix. The loop of selecting matching pairs and calculating matching error terminates, and the cycle threshold is the number of times, if the number of times reaches the threshold, the loop is terminated. Record the transformation matrix corresponding to the minimum Huber error, and the transformation matrix at this time is the best initial transformation matrix.

[0039] Use the obtained best transformation matrix as the initial transformation matrix to complete the coarse registration of the source point cloud and the target point cloud.

[0040] In this embodiment, the point clouds to be registered are as shown in Figure 7 (a), and the initial registration result in this embodiment is as shown in Figure 7 (b). Finally, the iterative closest point algorithm (ICP) is used to complete fine registration. The enlarged view of the local details after registration is as shown in Figure 7 (c)-(d), wherein Figure 7 (c) is an enlarged view of the details of the traditional registration method, Figure 7 (d) is an enlarged view of the details of the registration method of the present application.

[0041] Embodiment 2 A tunnel point cloud registration method fusing color and geometric features, the specific steps are as follows: ​​S1, acquire the color point cloud to be registered using a laser scanner equipped with a color perception module, set the calculation radius and calculate the point cloud normal.

[0042] S2, calculate the geometric features of the point cloud, and acquire the key points of the source point cloud and the target point cloud.

[0043] S3, perform neighborhood search on the key points, calculate the geometric and color features of the key points, and construct the color quick feature histogram of the key points.

[0044] S4, perform feature matching according to the feature histogram of the key points to obtain the initial feature matching pairs.

[0045] S5, use the RANSAC (Random Sample Consensus) algorithm to obtain the best transformation matrix through random sampling and iterative optimization, so that more matching pairs have an error less than a threshold value after transformation.

[0046] S6, use the obtained best transformation matrix as the initial transformation matrix to complete the coarse registration of the source point cloud and the target point cloud, and finally use the Iterative Closest Point algorithm (ICP) to complete the fine registration.

[0047] Step S1, most existing point clouds have color information, which can be easily obtained through laser emitter and synchronous camera integrated data. Use a laser scanner equipped with a color perception module or an RGB-D camera to acquire the color point cloud to be registered, set the calculation radius and calculate the point cloud normal. The size of the radius setting is related to the resolution of the point cloud. According to the geometric structure characteristics of the tunnel, it is recommended to set the radius to be 5-10 times the resolution size.

[0048] Step S2, calculate the distribution of points in the neighborhood of each point, and extract the intrinsic shape feature (Intrinsic Shape Signature, ISS) key points by calculating the covariance matrix and its eigenvalues.

[0049] Preferably, the specific steps of S3 are as follows: for each key point, search for other points in the neighborhood of the key point and calculate the position relationship geometric feature vector of the neighborhood points and the key point , then convert the point cloud color RGB to YIQ color space, where Y component represents luminance information, and I, Q components represent chrominance information; calculate the YIQ mean value of each point in the neighborhood of the key point to obtain the color feature vector . The geometric feature and the color feature are spliced into a six-dimensional joint feature vector Then the feature histogram is constructed, and the value range of the feature vector is divided into (11, 11, 11, 4, 10, 10) for a total of 57 bins. The neighborhood points are set as the query points, and the above steps are repeated. The feature six-tuple of the query point is calculated again, and the histogram is constructed. The feature histogram of the neighborhood point is weighted to the feature histogram of the key point according to the distance attenuation. The feature distribution of each subspace is independently counted and the histogram is drawn. The feature histograms are combined to finally generate a 57-dimensional fast feature histogram with space-color coupling characteristics (Tunnel Colored Fast Point Feature Histogram, TCFPFH).

[0050] Preferably, the specific steps of S4 are as follows: after calculating the 57-dimensional fast feature histogram TCFPFH in S3, bidirectional nearest neighbor matching is performed on the source point cloud key point set P and the target point cloud key point set Q, and the specific steps are as follows: S41, forward matching: for each , find the candidate point set in Q that satisfies .Wherein represents the feature histogram of the point S42, reverse verification: for in P, verify whether is the nearest neighbor in P.

[0051] S43, reserve the point pairs consistent in both directions to form the initial matching set .

[0052] Preferably, the specific steps of S5 are as follows: S51, randomly select matching pairs from the matching set , while ensuring that the minimum distance between each other is greater than the set threshold d.

[0053] S52, estimate the rigid transformation matrix between using the coarse ICP and calculate the Huber error between the matching pair set, and save both.

[0054] S53, repeat steps S51 and S52, and use the rigid transformation matrix corresponding to the minimum Huber error as the initial transformation matrix.

[0055] S6, use the initial transformation matrix to perform rigid transformation on the original point cloud to complete the initial registration. Perform ICP registration on the source point cloud and the target point cloud after the initial transformation to obtain the fine registration transformation matrix and the registered point cloud.

[0056] Obviously, the above embodiments are merely example for clearly illustrating but not limitation to the embodiments. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, all the embodiments need not and can not be enumerated. The obvious changes or variations derived from this still fall within the protection scope of the present application.

Claims

1. A method for registering tunnel point clouds by integrating color and geometric features, characterized in that, Includes the following steps: S1. Use a laser scanner equipped with a color perception module to acquire the color point cloud to be registered, set the calculation radius and calculate the point cloud normal; S2. Calculate the geometric features of the point cloud to obtain the key points of the source and target point clouds; S3. Perform neighborhood search on key points, calculate the geometric and color features of key points and construct a fast color feature histogram of key points; S4. Perform feature matching based on the feature histogram of key points to obtain initial feature matching pairs; S5. Use the random sampling consensus algorithm to obtain the optimal transformation matrix through random sampling and iterative optimization; S6. Use the optimal transformation matrix as the initial transformation matrix to complete the coarse registration of the source point cloud and the target point cloud, and finally use the iterative nearest point algorithm to complete the fine registration.

2. The tunnel point cloud registration method for fusing color and geometric features according to claim 1, characterized in that, The method for obtaining key points in step S2 is the inherent shape feature key point extraction method, which includes calculating the neighborhood covariance matrix and its eigenvalues ​​of each point in the point cloud, and filtering key points by eigenvalue ratio and non-maximum suppression.

3. The tunnel point cloud registration method for fusing color and geometric features according to claim 1, characterized in that, The construction of the color fast feature histogram of key points in step S3 includes: Calculate the geometric feature vector between key points and their neighboring points. ; Calculate the color feature vectors of key points and their neighbors. ; The geometric feature vector and the color feature vector are concatenated using tensors to form a six-dimensional joint feature vector. .

4. The tunnel point cloud registration method for fusing color and geometric features according to claim 3, characterized in that, The color feature vector The point cloud color is obtained by converting it from the RGB color space to the YIQ color space, where the Y component represents luminance information and the I and Q components represent chromaticity information.

5. The tunnel point cloud registration method for fusing color and geometric features according to claim 3, characterized in that, The color fast feature histogram is a 57-dimensional feature histogram, which is obtained by using geometric feature vectors. Each feature vector is divided into 11 bins, and the color feature vectors Y, I, and Q are divided into 4, 10, and 10 bins respectively. The feature histograms are then merged.

6. The tunnel point cloud registration method for fusing color and geometric features according to claim 1, characterized in that, The feature matching in S4 employs a bidirectional nearest neighbor matching method, including: S41. Forward Matching: Matching keypoint sets from the source point cloud. Each point in Search for the key point set Q in the target point cloud that satisfies candidate point set , where F(·) represents the feature histogram of the point; S42, Reverse Verification: [This part is incomplete and likely refers to a separate, unrelated section.] Each point in Verify in reverse P Is it its nearest neighbor? S43. Retain bidirectional consistent point pairs to form the initial matching set. .

7. The tunnel point cloud registration method for fusing color and geometric features according to claim 1, characterized in that, The random sampling consensus algorithm used in S5 includes: S51, from the matching set Random selection A matching pair, ensuring The minimum distance between any two points in a matching pair is greater than the threshold d; S52. Use a coarse ICP estimate The rigid transformation matrix between each matching pair is calculated, and the Huber error between the matching pair sets is computed. S53. Repeat steps S51 and S52 to select the rigid transformation matrix corresponding to the minimum Huber error as the optimal transformation matrix.

8. The tunnel point cloud registration method for fusing color and geometric features according to claim 1, characterized in that, In step S6, coarse registration uses the optimal transformation matrix to perform a rigid transformation on the source point cloud, and fine registration uses the iterative nearest point algorithm for iterative optimization to obtain the registered point cloud.

9. The tunnel point cloud registration method for fusing color and geometric features according to claim 1, characterized in that, In step S1, the calculation radius is set to 5-10 times the point cloud resolution.

10. The tunnel point cloud registration method for fusing color and geometric features according to claim 1, characterized in that, The step S3 of constructing the color fast feature histogram of key points also includes: For each keypoint, calculate the color fast feature histogram of its neighboring points, and perform a weighted sum based on the distance between the neighboring points and the keypoint; The weighted sum of the feature histograms is normalized to make the sum of the histograms a fixed value.

Images