A tunnel point cloud high-precision splicing method based on rotation translation constraint

Abstract

The application discloses a tunnel point cloud high-precision splicing method based on rotation and translation constraints, and belongs to the field of engineering measurement and three-dimensional laser scanning technology. The method comprises the following steps: in the data acquisition stage, the absolute coordinates and relative point cloud data of each station are acquired through the cooperative operation of a total station and a station scanner, and the scanning of a target tunnel section is completed; in the point cloud data preliminary conversion stage, the relative point cloud data of each station is subjected to translation transformation by using the absolute coordinates of the station scanner, and is preliminarily transformed into a global coordinate system; in the two-stage registration stage based on rotation and translation constraints, the relative point cloud data of adjacent stations after preliminary transformation is subjected to two-stage registration; and in the global point cloud model generation stage, the relative point cloud data of all stations is unified to the tunnel construction global coordinate system through two-stage registration, and a complete tunnel three-dimensional point cloud model is generated. According to the application, the rotation and translation constraints and the two-stage registration strategy are adopted, so that the millimeter-level splicing precision is ensured, and the cost and operation difficulty are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of engineering surveying and three-dimensional laser scanning technology, and in particular to a tunnel point cloud high-precision splicing method based on rotation and translation constraints. BACKGROUND

[0002] In the process of tunnel engineering construction, quickly and accurately obtaining the three-dimensional spatial information (i.e. three-dimensional point cloud data) of the excavation profile is crucial for controlling overbreak and underbreak, ensuring construction quality, and optimizing construction progress. Three-dimensional laser scanning technology has become an important means of tunnel engineering surveying and digital modeling due to its high-density and high-efficiency data acquisition capabilities. However, current point cloud data acquisition and splicing schemes applied in tunnel environments have obvious limitations, making it difficult to achieve an ideal balance between precision, cost, and efficiency.

[0003] Currently, the mainstream technical solutions mainly include the following categories:

[0004] The first category is high-precision total station scanners. This type of device integrates a total station and a laser scanning module, allowing direct acquisition of high-precision point cloud data with absolute coordinates. However, it has significant drawbacks: the device procurement cost is extremely high, making it difficult to popularize in small and medium-sized projects; precise coordinate station setting is required before scanning, the operation process is complex, and the overall scanning efficiency is low, making it difficult to meet the needs of rapid measurement in tunnel construction.

[0005] The second category is handheld or mobile scanners based on SLAM (Simultaneous Localization and Mapping). This type of device is lightweight and flexible, allowing scanning without the need for station setting, and is efficient and easy to operate. However, its core defect is insufficient precision, typically only reaching centimeter-level accuracy, which cannot meet the strict requirements of millimeter-level precision in tunnel construction monitoring. Especially in environments with unobvious surface features and high structural symmetry, such as tunnel secondary lining, SLAM algorithms are prone to drift, leading to further increased point cloud splicing errors and reduced data reliability.

[0006] The third category is station-mounted scanners. This type of device has fast scanning speed and high point cloud precision at a single station. However, it is essentially a relative measurement device and cannot perform absolute coordinate station setting like a total station. Each station's acquired point cloud is in an independent instrument coordinate system. Traditional point cloud splicing methods (such as those relying on targets or automatic registration based on scene features) in long-distance, multi-station tunnel scanning accumulate errors with increasing stations, leading to global splicing deviations of up to decimeters, which cannot guarantee the absolute precision of the overall model. Although algorithms such as Iterative Closest Point (ICP) can be used for point cloud fine registration, in the absence of high-precision initial poses, they are prone to local optimal solutions and have high computational complexity, limiting their ability to control global precision in long-distance tunnel scanning.

[0007] In summary, the prior art has a clear contradiction: high-precision solutions (total station scanners) are high in cost and low in efficiency; while low-cost, high-efficiency solutions (SLAM, traditional splicing of frame station scanners) cannot meet the requirements of high precision. Therefore, the field urgently needs an innovative technical solution that can integrate high-precision absolute coordinate control with low-cost, high-efficiency relative scanning technology, while ensuring millimeter-level splicing accuracy, reducing equipment cost and operation difficulty, to meet the actual needs of tunnel engineering construction. SUMMARY

[0008] The purpose of the present application is to overcome the shortcomings of the prior art and provide a tunnel point cloud high-precision splicing method based on rotational translation constraints, which combines the high-precision absolute coordinate measurement capability of a total station with the flexible scanning capability of a frame station scanner, introduces innovative rotational translation constraints and a two-stage registration strategy, and significantly reduces equipment cost and operation difficulty while ensuring millimeter-level splicing accuracy.

[0009] The purpose of the present application is achieved by the following technical solution: a tunnel point cloud high-precision splicing method based on rotational translation constraints, comprising the following steps:

[0010] Data acquisition stage: through the cooperative operation of the total station and the frame station scanner, the absolute coordinates and relative point cloud data of each station are obtained, and the scanning of the target tunnel section is completed;

[0011] Point cloud data preliminary conversion stage: the relative point cloud data of each station is translated and transformed using the absolute coordinates of the frame station scanner, and is initially transformed into the global coordinate system;

[0012] Two-stage registration stage based on rotational translation constraints: the relative point cloud data of adjacent stations after preliminary transformation is subjected to two-stage registration; first, coarse registration is performed, the optimal initial rotation angle of the frame station scanner around the vertical axis is determined through sector division and inner point maximization search; then, fine registration is performed, taking the optimal initial rotation angle as the initial value, using an improved IPC algorithm, iterating under the constraint of fixing the translation parameters and optimizing only the rotation degree of freedom, to solve the optimal rotation angle correction;

[0013] Global point cloud model generation stage: the relative point cloud data of all stations is unified to the tunnel construction global coordinate system through two-stage registration, and a complete tunnel three-dimensional point cloud model is generated.

[0014] Preferably, the data acquisition stage further comprises the following steps:

[0015] A total station is erected on a known control point in the tunnel, and station setting is completed to establish a global coordinate system for tunnel construction; a cooperative target is fixed on the station-mounted scanner, and the three-dimensional absolute coordinates of the cooperative target are measured by the total station as the station center coordinates of the station-mounted scanner; the station-mounted scanner is started to scan the tunnel profile covered by the current station, and relative point cloud data based on the station-mounted scanner's own coordinate system is obtained; the scanner is repeatedly moved to different positions to obtain the station center coordinates and relative point cloud data of the current position until the scanning of the entire tunnel section is completed.

[0016] Preferably, the transformation formula of the translation transformation is as follows:

[0017] , wherein is the point cloud coordinate after preliminary transformation; is the absolute coordinate of the station-mounted scanner measured by the total station; is the coordinate of the point cloud in the relative coordinate system of the station-mounted scanner.

[0018] Preferably, the coarse registration further comprises the following steps:

[0019] The point cloud data of the two adjacent stations after preliminary conversion is divided into multiple sector regions according to the azimuth angle;

[0020] Multiple rotations are performed at preset angle intervals within a preset angle range until the entire preset angle range is traversed;

[0021] For each rotation, the point cloud of one of the stations is rotated around the vertical axis, and the number of inliers of the point cloud of the two adjacent stations in the overlapping region is calculated, wherein the inlier is defined as a point pair in the overlapping region with a distance less than a preset distance threshold.

[0022] The number of inliers under different rotation angles is compared, and the angle with the most inliers is determined as the optimal initial rotation angle.

[0023] Preferably, the fine registration further comprises the following steps:

[0024] The registration problem is parameterized, and is constrained to solve only in the degree of freedom of rotation around the vertical axis, and the translation parameters provided by the total station are used to reduce the complexity of the solution;

[0025] The nearest neighbor point of each point in the target point cloud and the point to be registered is found to form a point pair;

[0026] An optimal rigid transformation matrix that minimizes the root mean square error between all point pairs is calculated, and the optimal rigid transformation matrix includes a rotation angle correction;

[0027] The position of the point cloud is updated by the optimal rigid transformation matrix, and repeated iteration is performed until the optimal rotation angle correction is obtained.

[0028] Preferably, the cooperation target is a prism.

[0029] Preferably, the division interval of the fan-shaped region is 10°, the preset angle is 1°, and the distance threshold is 5 cm.

[0030] Preferably, the singular value decomposition algorithm is used to calculate the optimal rigid transformation matrix, and the iteration stopping condition is that the rotation angle change is less than 0.001°.

[0031] The beneficial effects of the present application are:

[0032] 1) The unity of high precision and low cost is realized: the absolute coordinate measurement capability of the high-precision total station and the high-efficiency scanning capability of the low-cost station scanner are combined creatively. The millimeter-level translation parameters provided by the total station are used as strict constraints, and the complex registration problem is reduced to the rotation freedom around the vertical axis, effectively suppressing error accumulation. The final point cloud splicing precision can reach millimeter level, which is much better than the traditional SLAM scheme (centimeter level) and the traditional station splicing scheme (decimeter level), and the equipment cost can be reduced by more than 50% compared with the high-end total station scanner.

[0033] 2) The efficiency and robustness of registration are improved: a two-stage strategy of "coarse registration + fine registration" is adopted, the optimal initial rotation angle is quickly determined through coarse registration based on fan-shaped region division and inlier maximization search, and the parameter search space is greatly reduced. The improved ICP fine registration is carried out based on the initial value, which greatly improves the convergence speed and avoids the problem that the traditional ICP algorithm easily falls into local optimal solution due to poor initial value, ensuring the success rate and stability of registration, especially suitable for long-distance and large-scene tunnel scanning.

[0034] 3) The anti-interference ability and applicability are enhanced: the registration core of the method depends on the geometric point cloud data itself, not on environmental features or magnetic field signals, so it is completely unaffected by the common GPS signal loss, strong magnetic interference, and dim light in the tunnel, and has stronger environmental adaptability and robustness.

[0035] 4) It has good practicability and generalizability: the operation process of the method is clear, and it is highly compatible with the existing control measurement and setting-out process on the construction site, without changing the conventional operation habit, and the measurement personnel are easy to learn and master. It fully utilizes the value of existing common equipment (total station), and realizes high-precision digitization of the tunnel by introducing a low-cost scanner, providing efficient and reliable data support for intelligent and information management of tunnel construction, which has great market promotion value. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 The figure is a flowchart of the method of the present application. DETAILED DESCRIPTION

[0037] The technical solutions of the present application will be described clearly and completely in combination with the embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0038] Referring to Figure 1 The present application provides a technical solution: a tunnel point cloud high-precision splicing method based on rotation and translation constraints, comprising the following steps:

[0039] Data acquisition stage: through the cooperative operation of the total station and the station-mounted scanner, the absolute coordinates and relative point cloud data of each station are obtained, and the scanning of the target tunnel section is completed;

[0040] Point cloud data preliminary conversion stage: the relative point cloud data of each station is translated and transformed by using the absolute coordinates of the station-mounted scanner, and is initially transformed into the global coordinate system;

[0041] Two-stage registration stage based on rotation and translation constraints: the relative point cloud data of adjacent stations after preliminary transformation is subjected to two-stage registration; first, coarse registration is performed, the optimal initial rotation angle of the station-mounted scanner around the vertical axis is determined through sector region division and inner point maximum search; then, fine registration is performed, taking the optimal initial rotation angle as the initial value, and using the improved IPC algorithm, the optimal rotation angle correction is solved under the constraint of fixing the translation parameter and optimizing only the rotation degree of freedom through iteration;

[0042] Global point cloud model generation stage: the relative point cloud data of all stations is unified to the tunnel construction global coordinate system through two-stage registration, and a complete tunnel three-dimensional point cloud model is generated.

[0043] In this embodiment, first, a plurality of measurement control points are laid along the line in the tunnel, and the three-dimensional coordinates of these control points are accurately determined by using a high-precision total station through a closed traverse measurement method, and a stable tunnel construction global coordinate system is established. A station-mounted three-dimensional laser scanner is prepared, and a special prism is fixed on the top of the scanner as a cooperative target.

[0044] In some embodiments, the data acquisition stage further comprises the following steps:

[0045] A total station is set up on a known control point in the tunnel, and station setting is completed to establish a global coordinate system for tunnel construction. A cooperative target is fixed on the station-mounted scanner, and the three-dimensional absolute coordinates of the cooperative target are measured by the total station as the station center coordinates of the station-mounted scanner. The station-mounted scanner is started to scan the tunnel profile covered by the current station, and relative point cloud data based on the coordinate system of the scanner itself is obtained. The scanner is moved to different positions, and the station center coordinates and relative point cloud data of the current position are obtained, until the scanning of the entire tunnel section is completed.

[0046] In this embodiment, a total station is set up on a known control point in the tunnel, and station setting is completed to establish a global coordinate system for tunnel construction. A cooperative target (such as a prism) is fixed on the station-mounted scanner, and the three-dimensional absolute coordinates of the cooperative target are accurately measured by the total station as the station center coordinates of the scanner. The station-mounted scanner is started to scan the tunnel profile covered by the current station, and relative point cloud data based on the coordinate system of the scanner itself is obtained. The scanner is moved to different positions, and the station center coordinates and relative point cloud data of the current position are obtained, until the scanning of the entire tunnel section is completed.

[0047] In some embodiments, the transformation formula of the translation transformation is as follows:

[0048] wherein is the point cloud coordinate after preliminary transformation; is the absolute coordinate of the station-mounted scanner measured by the total station; is the coordinate of the point cloud in the relative coordinate system of the station-mounted scanner.

[0049] In this embodiment, the preliminary conversion (absolute coordinate translation) of the point cloud data preliminarily places the point cloud in the global coordinate system, but does not consider the attitude rotation of the scanner and has not been accurately corrected.

[0050] In some embodiments, the coarse registration further includes the following steps:

[0051] The point cloud data of the two adjacent stations after preliminary conversion is divided into multiple sector regions according to the azimuth angle;

[0052] Multiple rotations are performed at preset angle intervals within a preset angle range until the entire preset angle range is traversed;

[0053] For each rotation, the point cloud of one of the stations is rotated around the vertical axis, and the number of inliers of the point cloud of the two adjacent stations in the overlapping region is calculated, wherein the inlier is defined as a point pair in the overlapping region with a distance less than a preset distance threshold.

[0054] The number of inliers under different rotation angles is compared, and the angle with the most inliers is determined as the optimal initial rotation angle.

[0055] In this embodiment, the point cloud of the adjacent station 1 and station 2 (denoted as P1 and P2) is taken as an example for registration. Coarse registration - based on sector division and inlier maximization search:

[0056] a. Sector division: The point cloud data of P1 and P2 after preliminary translation is divided into 36 sectors with the global coordinate system origin as the vertex (or with the center of the point cloud as the reference) according to the azimuth angle. For example, the range of 0° to 360° is divided into 36 sectors with an interval of 10°.

[0057] b. Rotational space search: A series of candidate rotation angles θ (θ = 0°, 1°, 2°,..., 359°) are generated within the range of 0° to 360° with a step size of 1°.

[0058] c. Inlier collision detection: For each candidate angle θ, the point cloud P2 is rotated by θ degrees around the Z-axis. Then, the number of corresponding point pairs in the overlapping sector regions of P1 and the rotated P2 is calculated. A distance threshold d_threshold (preferably 5 cm in this example) is set, and when the Euclidean distance between two points in a point pair is less than d_threshold, the point pair is considered as an "inlier".

[0059] d. Optimal angle selection: Compare the number of inliers corresponding to all candidate rotation angles θ, and determine the optimal initial rotation angle θ_max when the number of inliers reaches the maximum value.

[0060] In some embodiments, the fine registration further includes the following steps:

[0061] Parameterize the registration problem and constrain it to only solve in the vertical axis rotation degree of freedom, use the translation parameters provided by the total station to reduce the complexity of the solution;

[0062] Find the nearest neighbor of each point in the target point cloud to the point to be registered, forming a point pair;

[0063] Calculate the optimal rigid transformation matrix that minimizes the root mean square error between all point pairs, which includes the rotation angle correction;

[0064] Update the point cloud position through the optimal rigid transformation matrix and repeat the iteration until the optimal rotation angle correction is obtained.

[0065] In this embodiment, the fine registration is based on the improved Iterative Closest Point (ICP) algorithm:

[0066] a. Initialization and constraint setting: the transformation parameters obtained in coarse registration are used as the initial values for fine registration. The registration problem is parameterized and constrained to solve only one degree of freedom of transformation, i.e. a small rotation correction amount Δθ around the Z axis, while the translation parameters are fixed.

[0067] b. Nearest point iteration: find the nearest neighbor of each point in P2 in P1 after rotation, forming a temporary point pair set.

[0068] c. Error minimization solution: after eliminating invalid point pairs with too large distance, use singular value decomposition (SVD) algorithm to calculate an optimal rigid transformation matrix that minimizes the root mean square error (RMSE) between all valid point pairs. Since the translation is constrained, this transformation matrix mainly contains a small rotation angle correction amount Δθ around the Z axis.

[0069] d. Iterative update: update the position of the point cloud using the calculated optimal rigid transformation matrix. Repeat the iteration until the absolute value of the calculated rotation angle correction amount Δθ is less than a preset convergence threshold (e.g. 0.001°), then the algorithm is considered to have converged. At this time, the optimal rotation angle correction amount is obtained.

[0070] In some embodiments, the cooperative target is a prism.

[0071] In some embodiments, the division interval of the fan-shaped region is 10°, the preset angle is 1°, and the distance threshold is 5 cm.

[0072] In some embodiments, the singular value decomposition algorithm is used to calculate the optimal rigid transformation matrix, and the iteration stopping condition is that the rotation angle change is less than 0.001°.

[0073] The above description is only the preferred embodiments of the present application, and it should be understood that the present application is not limited to the forms disclosed herein, and should not be considered as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concepts described herein, by the above teachings or related art or knowledge. Any modification and change made by those skilled in the art without departing from the spirit and scope of the present application shall be within the protection scope of the appended claims of the present application.

Claims

1. A high-precision stitching method for tunnel point clouds based on rotation and translation constraints, characterized in that: Includes the following steps: Data acquisition phase: Through the collaborative operation of total station and station scanner, the absolute coordinates and relative point cloud data of each station are obtained to complete the scanning of the target tunnel section; Preliminary point cloud data conversion stage: The relative point cloud data of each station is translated and transformed using the absolute coordinates of the station scanner, and initially transformed into the global coordinate system; The two-stage registration process based on rotation and translation constraints involves two-stage registration of the relative point cloud data of adjacent stations after preliminary transformation. First, coarse registration is performed by dividing the area into fan-shaped regions and maximizing the search for interior points to determine the optimal initial rotation angle of the station scanner around the vertical axis. Then, fine registration is performed by using the optimal initial rotation angle as the initial value and employing an improved IPC algorithm to iterate under the constraint of fixed translation parameters and optimizing only the rotational degrees of freedom to solve for the optimal rotation angle correction. Global point cloud model generation stage: The relative point cloud data of all stations are unified into the global coordinate system of tunnel construction through two-level registration to generate a complete three-dimensional point cloud model of the tunnel. The coarse registration also includes the following steps: The point cloud data of two adjacent stations, after preliminary transformation, are divided into multiple sector regions according to their azimuth angles. Rotate multiple times within a preset angle range at preset angle intervals until the entire preset angle range has been traversed. For each rotation, after rotating the point cloud of one station around the vertical axis, the number of inner points in the overlapping area of ​​the point clouds of two adjacent stations is calculated. The inner point is defined as a pair of points in the overlapping area whose distance is less than a preset distance threshold. Compare the number of interior points under different rotation angles, and determine the angle with the most interior points as the optimal initial rotation angle; The fine registration also includes the following steps: The registration problem is parameterized, and it is constrained to be solved only on the degree of freedom of rotation along the vertical axis. The translation parameters provided by the total station are used to reduce the complexity of the solution. Find the nearest neighbor of each point in the target point cloud and the point cloud to be registered, and form point pairs; Calculate the optimal rigid body transformation matrix that minimizes the root mean square error between all pairs of points, wherein the optimal rigid body transformation matrix includes rotation angle correction. The point cloud position is updated by updating the optimal rigid body transformation matrix, and the process is repeated until the optimal rotation angle correction is obtained.

2. The high-precision stitching method for tunnel point clouds based on rotation and translation constraints according to claim 1, characterized in that: The data acquisition phase also includes the following steps: A total station is set up at known control points inside the tunnel to establish a global coordinate system for tunnel construction. The cooperative target is fixed on the station-mounted scanner, and the three-dimensional absolute coordinates of the cooperative target are measured using the total station, which serve as the station center coordinates of the station-mounted scanner. The station-mounted scanner is started to scan the tunnel outline covered by the current station, and relative point cloud data based on the station-mounted scanner's own coordinate system is obtained. The scanner is repeatedly moved to different positions to obtain the station center coordinates and relative point cloud data of the current position until the scanning of the entire tunnel section is completed.

3. The high-precision stitching method for tunnel point clouds based on rotation and translation constraints according to claim 1, characterized in that: The transformation formula for the translation transformation is as follows: ,in These are the coordinates of the point cloud after the initial transformation; These are the absolute coordinates of the station-mounted scanner measured by the total station; The coordinates of the point cloud in the relative coordinate system of the station scanner.

4. The high-precision stitching method for tunnel point clouds based on rotation and translation constraints according to claim 2, characterized in that: The objective of the cooperation is a prism.

5. The high-precision stitching method for tunnel point clouds based on rotation and translation constraints according to claim 1, characterized in that: The fan-shaped region is divided at intervals of 10°, the preset angle is 1°, and the distance threshold is 5cm.

6. The high-precision stitching method for tunnel point clouds based on rotation and translation constraints according to claim 1, characterized in that: The optimal rigid body transformation matrix is ​​calculated using the singular value decomposition algorithm, with the iteration stopping condition being that the change in rotation angle is less than 0.001°.

Images