Method and system for establishing mapping relationship between rotating linear array camera and ground-based laser radar

Abstract

This invention discloses a method and system for establishing a mapping relationship between a rotating linear array camera and a ground-based lidar. The establishment process includes two steps: First, it proposes a geometric mapping relationship between the rotating linear array camera and the 3D lidar, analyzing the attitude and position errors between the camera and the lidar during the camera's rotation, while also considering the impact of camera distortion on the mapping results. Second, a hybrid optimization algorithm further optimizes the data mapping relationship. Utilizing the texture similarity between the point cloud intensity map and the grayscale image of the initially mapped point cloud, the hybrid optimization algorithm evaluates and optimizes their similarity, improving the fusion accuracy of the point cloud and the image. This invention establishes a data mapping relationship between a rotating linear array camera and a 3D laser scanner in two steps, writes the mapping relationship into the equipment development configuration file, achieves single-station data fusion, and obtains a color point cloud.

Description

Technical Field

[0001] This invention relates to the field of geometric mapping relationship technology between 3D laser scanners and cameras, specifically to a method and system for establishing the mapping relationship between a rotating linear array camera and a ground-based 3D laser scanner. Background Technology

[0002] Terrestrial 3D laser scanners are rapid, high-precision real-scene 3D data acquisition devices, widely used in tasks such as digital preservation of cultural relics, disaster early warning and monitoring, 3D reconstruction of buildings, and battlefield environmental reconnaissance. Existing terrestrial 3D laser scanners often combine optical cameras with laser scanning equipment. The working principle of an optical camera is as follows: light emitted from the surface of the object is focused by a lens onto a CCD / CMOS device. A photodiode converts the light into electrical charge, which, after a series of processing steps, forms a video signal output, which is then transmitted to an image processor to obtain an image. Cameras have advantages such as high resolution, rich texture information, and high frame rate, but they are easily affected by lighting conditions and it is difficult to obtain accurate geometric information from 2D images. Ordinary cameras have limited field of view, so 360° panoramic cameras are often used to acquire panoramic images of the scene. The working principle of lidar is as follows: the distance between the scanner and the object is obtained based on the phase difference between the emitted and returned signals or based on the round-trip time of light. Then, by using an angle encoder to measure the vertical and horizontal rotation angles, the 3D coordinates of each point cloud are obtained. The combined use of optical cameras and lidar can achieve complementary data information. The fusion of the two can produce a color point cloud, which can improve the readability and processing efficiency of the data. It has the characteristics of strong realism and easy immersion, and can reach the level of surveying and mapping applications. It is often used in technical fields such as infrastructure detection, battlefield environment mapping, unmanned driving, and new street view maps.

[0003] However, the promotion and application of 3D laser scanners still require technological innovation. Existing 3D laser scanners mostly employ built-in or external single or multiple area scan cameras. Both the camera and laser use asynchronous measurement methods. Their working principle is as follows: the laser first scans the environmental point cloud, and then the camera, driven by a rotating motor, acquires multi-angle environmental images. These images are then stitched together to obtain a panoramic image, and the point cloud is then colored using data mapping relationships. This working mode has some problems. On the one hand, the camera needs to acquire environmental images in multiple poses, which easily accumulates image stitching errors and image redundancy. Data mapping errors are affected by the camera's position and pose, and these errors accumulate as the camera's pose changes. On the other hand, there is a time difference between the data acquisition by the laser and the area scan camera, which may cause different measurement results from the camera and laser at the same angle, leading to incorrect judgments and interpretations. Summary of the Invention

[0004] In view of the above problems, this invention proposes a geometric mapping method and system for a rotating linear scan camera and a ground-based 3D laser scanner. The aim is to establish a mapping relationship between the panoramic image measured by the rotating linear scan camera and the 3D point cloud, thereby enabling the device to measure a color point cloud.

[0005] The method for establishing the geometric mapping relationship between a rotating linear array camera and a ground-based three-dimensional laser scanner designed in this invention includes the following steps: establishing an initial geometric mapping relationship between the rotating linear array camera and the three-dimensional laser scanner, wherein the initial geometric mapping relationship is calculated based on the corresponding point pairs extracted from the target deployed in the measurement scene, and includes at least correction parameters for camera eccentricity error, image plane tilt error and optical distortion; Based on the initial geometric mapping relationship, the image information acquired by the rotating linear array camera is mapped to the point cloud acquired by the ground 3D laser scanner to generate an initial colorized image; Using the intensity image generated from the inherent reflection intensity information in the point cloud as a spatial reference, a non-rigid image registration algorithm is used to solve the spatial deformation field from the initial color image to the intensity image, wherein the non-rigid image registration is used to correct the nonlinear residual error that is difficult to eliminate in the initial geometric mapping relationship; The system error of the initial geometric mapping relationship is inverted based on the spatial deformation field, and the initial geometric mapping relationship is nonlinearly corrected to obtain the optimized final geometric mapping relationship.

[0006] Furthermore, the initial geometric mapping relationship is obtained through the following steps: Based on multiple targets deployed in the measurement scene, multiple sets of corresponding point pairs are extracted from the point cloud data and the image data acquired by the rotating linear array camera; a geometric mapping model is calculated using the multiple sets of corresponding point pairs to obtain the initial geometric mapping relationship.

[0007] Furthermore, the geometric mapping model is used to describe the geometric relationship between the projection center, rotation center, and image plane of the rotating linear array camera in the object-side point cloud coordinate system, and includes at least correction parameters for eccentricity error, image plane tilt error, and camera optical distortion.

[0008] Furthermore, the non-rigid image registration algorithm is a free deformation model based on a control grid; the calculation of the deformation field includes: setting up a control grid on the initial colorized image, using the displacement vector of the grid nodes as the optimization variable, and maximizing the similarity between the initial colorized image and the intensity image through iterative optimization, thereby determining the deformation field.

[0009] Furthermore, the similarity is measured using a normalized cross-correlation coefficient.

[0010] Furthermore, the iterative optimization employs the Levenberg-Marquardt algorithm, and the iterative process of the algorithm includes: The control step size parameter of the algorithm is dynamically adjusted based on the descent rate of the objective function in the current iteration.

[0011] Based on the same inventive concept, this invention also designs a ground-based three-dimensional laser scanning system, comprising: The system comprises a 3D laser scanning module, a rotating linear array camera module, and a control and processing module. The rotating linear array camera module is rigidly connected to the 3D laser scanning module and configured to synchronously acquire image data and point cloud data of the scene. The control and processing module stores executable instructions. When the executable instructions are executed by the processor, they implement a method for optimizing the geometric mapping relationship between the rotating linear array camera and the ground 3D laser scanner, and generate a color point cloud based on the final geometric mapping relationship.

[0012] Furthermore, the control and processing module is configured to: after optimizing the final geometric mapping relationship, write it into a configuration file for real-time generation of color point clouds in subsequent measurement tasks.

[0013] Based on the same inventive concept, the present invention also designs a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of a method for optimizing the geometric mapping relationship between a rotating linear array camera and a ground-based three-dimensional laser scanner.

[0014] Based on the same inventive concept, this invention also designs a three-dimensional scene color modeling method, including: The target scene is measured at multiple stations using a ground-based 3D laser scanning system. The final geometric mapping relationship of each station is obtained by optimizing the geometric mapping relationship between the rotating linear array camera and the ground-based 3D laser scanner. Using the final geometric mapping relationship of each station, the color information of the rotating linear array camera image of the corresponding station is mapped to the cloud of each station; The point clouds that have been colored at multiple stations are unified under the global coordinate system and fused to generate a complete 3D color scene model.

[0015] Compared with the prior art, the present invention has the following specific beneficial effects: 1. Establish a ground-based 3D laser scanner that combines a rotating linear array camera with laser scanning to achieve the goal of synchronously acquiring environmental data via laser and camera.

[0016] 2. This invention constructs a rigorous geometric mapping model that includes eccentricity error, image plane tilt error, and optical distortion parameters. Compared with existing methods that rely on precision mechanical adjustment (such as rigid parallel binding) or simple coordinate transformation, this invention transforms the installation error between sensors from 'physical elimination' to 'mathematical model compensation', significantly reducing the dependence on hardware processing and assembly precision, while improving the correctability and adaptability of the mapping relationship. 3. Based on the above rigorous geometric model, this invention can accurately describe the complex spatial pose relationship between the rotating linear array camera and the laser scanner without changing the hardware structure, providing accurate initial values ​​for subsequent nonlinear optimization and laying the theoretical foundation for high-precision fusion. 4. This invention creatively utilizes the inherent reflection intensity information of point clouds to generate intensity images, and uses these images as a spatial reference to correct the initial geometric mapping relationship through a non-rigid image registration algorithm. Compared with existing methods that use optical image features or depth images for registration, the advantages of this invention are: the reflection intensity information is not affected by changes in ambient lighting, and it has a physical texture similarity to the grayscale image captured by the camera, thus providing a more stable and reliable optimization reference and effectively avoiding the interference of lighting changes on registration accuracy.

[0017] 5. This invention employs a non-rigid registration algorithm based on a free-deformation model (FFD), specifically designed to correct nonlinear residual errors in the initial geometric mapping relationship that are difficult to parametrically model, including image stitching errors and edge distortion unique to rotating linear array cameras. Experimental data shows that, through the optimization steps of this invention, the mapping error is reduced by up to 53.6%, achieving a final mapping accuracy of 1.07 pixels. This specialized correction of "geometric model residual errors" is a technique not previously revealed in existing technologies, resulting in a significant improvement in accuracy.

[0018] This invention establishes a rigorous mapping relationship between rotating linear array images and three-dimensional point clouds, and supports the establishment of mapping relationships between multiple cameras and lidar, thereby achieving high-precision fusion of linear array panoramic images and lidar point clouds. Attached Figure Description

[0019] Figure 1 This represents the geometric relationship between the linear scan camera and the laser scanner under ideal conditions.

[0020] Figure 2 This is a schematic diagram of the geometric system error of a rotating linear array camera.

[0021] Figure 3 This is a schematic diagram of the experimental equipment.

[0022] Figure 4 This is a schematic diagram of the experimental data.

[0023] Figure 5 This is a map showing the distribution of indoor targets.

[0024] Figure 6 The mapping accuracy of 15 flat targets at different heights before and after iLM-FFD processing. Detailed Implementation The technical solution of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0025] Example 1 This embodiment discloses a geometric mapping method between a rotating linear array camera and a ground-based 3D laser scanner, including the following steps: First, the geometric relationship between the linear scan camera and the laser scanner under ideal conditions is established. Then, considering the position and orientation errors between the sensors, a rigorous geometric mapping relationship is constructed. Finally, to mitigate mapping errors caused by image stitching errors, edge distortion, and other factors, a hybrid algorithm combining LM and FFD is used to optimize the mapping relationship. The specific establishment process is as follows: I) Ideal geometric model.

[0026] Establishing an ideal geometric model is the process of constructing a geometric model from object space to image space. A single linear array panoramic camera rotates around a vertical axis and images continuously, acquiring continuous image segments around the station. During the imaging process, strict central projection relationships are satisfied along the pixel arrangement direction, and the points, image points, and projection centers satisfy the collinearity condition equation.

[0027] To provide a simple description of the geometric model of the linear scan camera, four coordinate systems are defined: pixel coordinate system. Linear array image space coordinate system Rotational auxiliary coordinate system and object coordinate system (Point cloud coordinate system). Figure 1 This illustrates the relationship between four coordinate systems, where the linear array image space coordinate system is defined by the rotation direction. Direction (counterclockwise is positive) The direction is the arrangement direction of the linear array pixels. In the ideal geometric model, the origin of the image space coordinate system coincides with the origin of the rotation auxiliary coordinate system. The point, whose coordinates in the object coordinate system are... Rotating auxiliary coordinate system The axis passes through the center of the projection, and Axis coincidence; before rotation begins The axis passes through the starting point of the linear cell array.

[0028] The rotation matrix between the object coordinate system and the auxiliary rotation coordinate system is: The conversion relationship between the two is as follows:

[0029] The transformation relationship between the rotation auxiliary coordinate system and the linear array image space coordinate system is as follows:

[0030] in, The coordinate scale factor; For linear array image space coordinate system to rotate coordinate system Axis rotation matrix; This is the transformation matrix from linear pixel coordinates to a rotated coordinate system; The focal length of the camera; This refers to the number of pixels in a linear scan camera. , For rotational resolution; This refers to the size of a single pixel in a linear scan camera. The pixel coordinates of the point The coordinates of the image point in the image space coordinate system of the online array.

[0031] Based on the above process, and combining formulas (1) and (2), the ideal transformation relationship between the linear array image space coordinate system and the object point cloud coordinate system is constructed as shown in formula (3).

[0032]

[0033] Ⅱ) A rigorous geometric model that takes into account position and attitude errors.

[0034] There are discrepancies between equipment design and actual measurements; installation errors are unavoidable; and the actual sensor geometric model is affected by certain systematic errors, such as... Figure 2 The systematic errors include the following aspects: (1) Actual projection center of the line scan camera The axis of rotation does not coincide with the axis of rotation, that is, the projection center and the center of rotation are not coincident. There is an eccentricity error When the projection center On the plane of revolution When, ,like Figure 2 As shown in (a).

[0035] (2) The imaging stripe of a line scan camera is not perfectly parallel to the rotation axis, that is, there are two angles between the image plane and the auxiliary coordinate system of the rotation axis: roll angle. and pitch angle , respectively Figure 2 (b) and Figure 2 As shown in (c).

[0036] (3) Optical distortion of line scan camera and the offset of the principal point on the image plane The distortion correction model for a linear scan camera is as follows: .

[0037] In summary, by adding eccentricity error correction, image plane tilt correction, and camera intrinsic parameter correction to the ideal geometric model, a rigorous geometric mapping relationship between the point cloud and the rotating linear array camera is obtained.

[0038]

[0039] , , There are three rotation matrices.

[0040] Ⅲ) Optimization of mapping relationships.

[0041] Due to image stitching errors, edge distortion, and other factors, the initial mapping accuracy of some image regions only reaches the pixel level, requiring further optimization. Free deformation is a local feature transformation model that transforms the deformed image (…). ) Deployment The grid is defined with grid nodes as control points, and each control point corresponds to... direction and Two control parameters for direction The positional offset is calculated using control parameters of 4×4 control points surrounding each pixel in the deformed image. Based on the initial mapping relationship, a hybrid optimization algorithm (iLM-FFD) combining improved Levenberg-Marquardt (LM) and Free-Form Deformation (FFD) is used to transform the problem of optimizing the mapping relationship into the original point cloud. Intensity map and point cloud after coloring The issue of calibration between grayscale images.

[0042] A point cloud intensity map is an image generated by the reflection intensity of a laser scanner. The generation process is as follows: a 3D point cloud is projected onto a unit sphere. The sphere is rasterized into multiple square grids at equal angles (0.05° in this study). The maximum point cloud intensity in each grid is taken as the grid's intensity. The unit sphere is then unfolded into a 2D planar image at a 0° horizontal angle. Based on the grid intensity range, an interpolation method is used to convert the values ​​to grayscale values ​​between 0 and 255, which are then assigned to the grid, thus obtaining the intensity map of the 3D point cloud.

[0043] The initial colorized grayscale image of the point cloud refers to the image generated after assigning camera grayscale information to the point cloud according to the initial mapping relationship. The generation process is as follows: The 3D point cloud is projected onto a unit sphere, which is then rasterized into multiple square grids at equal angles. The average RGB value of the point cloud in each grid is calculated and converted into a grayscale value, which is then assigned to the grid. The unit sphere is then unfolded into a 2D planar image at a 0° horizontal angle, thus obtaining the initial colorized grayscale image of the point cloud.

[0044] Based on the intensity map The grayscale image of the point cloud after coloring is used as a floating image. Using FFD as the transformation model for pixel-by-pixel deformation, and with Registration and calculation With the deformation The similarity between the two is calculated, and the optimal control parameters that maximize the similarity are obtained. Simultaneously, the parameter optimization process is accelerated based on an improved LM algorithm, and the optimal control parameters are used to... The mapping relationship is then optimized through deformation. The normalized cross-correlation (NCC) is unaffected by linear changes in grayscale values ​​and is suitable for textured and blurred images. This parameter is used as... and The similarity evaluation metric calculates the normalized cross-correlation coefficient of the images by averaging the results across blocks:

[0045] In the formula, , for The average grayscale value of the block. Total number of pixels Number of image blocks Since the LM algorithm aims to find the minimum value of the objective function, let the objective function of the model be... The process is as follows: 1) Control parameter initialization. Set the mesh size to... OK, Column, generate a column of length. one-dimensional array Assign a random number between [-0.001, 0.001].

[0046] 2) Calculation any pixel Coordinate offset relative to the original image First, calculate the integer grid coordinates of the pixels. and decimals Using B-spline basis functions Calculate the weighting coefficients in the x and y directions. , , , , , , , . As shown in equation (6).

[0047] To obtain the floating-point coordinates of the corresponding reference image, bilinear interpolation is needed to calculate the grayscale value and assign it to... , as in equation (7).

[0048]

[0049] 3) Improve the LM algorithm to optimize control parameters.

[0050] a) The gradient of the control parameters is calculated using the difference method, as shown in equation (8). Let be the objective function. .

[0051]

[0052] The gradients of all control points form a one-dimensional gradient vector. It is also the Jacobian matrix of the LM algorithm. : b) Calculate the objective function for the current control parameters. and correlation matrix , For the unit array; To control the step size.

[0053] c) Determine the optimal control parameters. (The parameters will satisfy...) Solution As the optimal control parameter Take a very small number.

[0054] d) Update parameters. Adjust parameters based on calculation results. First, update the control parameters, calculate the objective function value and step size factor. , as in equation (9).

[0055]

[0056] The step size controlled in this article is:

[0057] To improve the efficiency of algorithm parameter updates, the descent ratio is utilized. As a basis for judgment, the similarity calculated each time is compared with the previous worst similarity. The parameters are updated according to equation (11).

[0058]

[0059] in, In this embodiment , , Proceed to the next iteration of the calculation.

[0060] 4) Establish the optimized mapping relationship. The optimal control parameters are obtained based on the above steps. Establish an optimized and mapping relationship .

[0061] 5) Using the initial mapped point cloud Using this as a medium, establish point clouds. With pixels The pixel-level mapping relationship is used to obtain the final color point cloud. This represents the initial mapping relationship between the point cloud and the image.

[0062]

[0063] Step 1: Using the center of the spherical target as the feature point, obtain the corresponding point in the point cloud and image. Place multiple spherical targets in the environment, manually extract the target point cloud from the point cloud, and fit the center of the spherical target using the least squares method. The target image was manually obtained from the image, and the coordinates of the center of the spherical target in the image were extracted using the Hough circle detection method. This forms m pairs of points with the same name.

[0064] Step 2: Solve the geometric model mapping relationship using formula (4). Substitute all pairs of points with the same name from Step 1 into formula (4), where Construct a system of error equations The least squares method is used to solve for the parameters of each model. The coefficient matrix of the error equation, For the residual vector, This is the error correction value. This is a constant term.

[0065] Step 3: Mapping Optimization. Based on the initial mapping, a hybrid optimization algorithm based on improved LM and FFD is used to transform the problem of optimizing the mapping into the original point cloud. Intensity map and point cloud after coloring The issue of calibration between grayscale images. Using the intensity image as the reference image. The grayscale image of the point cloud after coloring is used as a floating image. Using FFD as the transformation model for pixel-by-pixel deformation, and with Registration and calculation With the deformation The similarity between the two is calculated, and the optimal control parameters that maximize the similarity are obtained. Simultaneously, the parameter optimization process is accelerated based on an improved LM algorithm, and the optimal control parameters are used to... The transformation optimizes the mapping relationship.

[0066] Apply the above steps in real-world scenarios. Figure 3 The experimental setup and equipment included a spherical target as the registration primitive. The equipment acquired panoramic images and 3D point clouds of four environments: high-rise buildings, playgrounds, plazas, and indoor laboratories. Figure 4 .

[0067] This study establishes a mapping relationship between a rotating linear array camera and a laser scanner. To quantitatively analyze the accuracy of the mapping results, the pixel offset values ​​of the corresponding target centers in the point cloud color image and the point cloud intensity image are calculated. The average pixel offset is defined as the mapping error. To further illustrate the improvement effect of the optimization method on the point cloud mapping relationship, 15 targets are deployed at different heights indoors, such as... Figure 5 Compare the target center offset before and after the optimization method, such as Figure 6 The algorithm can correct color shifts caused by regional image distortions, reducing mapping errors by up to 53.6%, while having no significant effect on distortion-free areas.

[0068] After algorithm optimization, a stable and accurate data mapping relationship was obtained, with a mapping accuracy of 1.07 pixels between point clouds and images in 10 sets of data. This result can be used to generate true-color point clouds, facilitating regular verification of the data quality of 3D scanners.

[0069] Example 2 Based on the same inventive concept, this invention also designs a ground-based three-dimensional laser scanning system, comprising: The system comprises a 3D laser scanning module, a rotating linear array camera module, and a control and processing module. The rotating linear array camera module is rigidly connected to the 3D laser scanning module and configured to synchronously acquire image data and point cloud data of the scene. The control and processing module stores executable instructions. When the executable instructions are executed by the processor, they implement a method for optimizing the geometric mapping relationship between the rotating linear array camera and the ground 3D laser scanner, and generate a color point cloud based on the final geometric mapping relationship.

[0070] The control and processing module is configured to write the final geometric mapping relationship into a configuration file after optimization, so as to generate a color point cloud in real time during subsequent measurement tasks.

[0071] Since the system described in Embodiment 2 of this invention is the system used to implement the method for optimizing the geometric mapping relationship between the rotating linear array camera and the ground-based three-dimensional laser scanner in Embodiment 1 of this invention, those skilled in the art can understand the specific structure and variations of this system based on the method described in Embodiment 1 of this invention, and therefore will not be repeated here. All systems used in any method of this invention's embodiments fall within the scope of protection of this invention.

[0072] Example 3 Based on the same inventive concept, the present invention also designs a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of a method for optimizing the geometric mapping relationship between a rotating linear array camera and a ground-based three-dimensional laser scanner.

[0073] Since the system described in Embodiment 3 of this invention uses a computer-readable storage medium to implement the geometric mapping relationship optimization method between the rotating linear array camera and the ground-based three-dimensional laser scanner in Embodiment 1 of this invention, those skilled in the art can understand the specific structure and variations of this system based on the method described in Embodiment 1 of this invention, and therefore will not be repeated here. All computer-readable storage media used in any method of this invention are within the scope of protection of this invention.

[0074] Example 4 Based on the same inventive concept, this invention also designs a three-dimensional scene color modeling method, including: The target scene is measured at multiple stations using a ground-based 3D laser scanning system. The final geometric mapping relationship of each station is obtained by optimizing the geometric mapping relationship between the rotating linear array camera and the ground-based 3D laser scanner. Using the final geometric mapping relationship of each station, the color information of the rotating linear array camera image of the corresponding station is mapped to the cloud of each station; The point clouds that have been colored at multiple stations are unified under the global coordinate system and fused to generate a complete 3D color scene model.

[0075] Since the three-dimensional scene color modeling method described in Embodiment 4 of this invention is the same three-dimensional scene color modeling method used in Embodiments 1 and 2, those skilled in the art can understand the specific structure and variations of the system based on the method described in Embodiment 1 of this invention, and therefore will not be repeated here. All three-dimensional scene color modeling methods used in Embodiments 1 and 2 of this invention fall within the scope of protection of this invention.

[0076] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the invention.

Claims

1. A method for establishing the geometric mapping relationship between a rotating linear array camera and a ground-based 3D laser scanner, characterized in that, Includes the following steps: An initial geometric mapping relationship is established between a rotating linear array camera and a 3D laser scanner. The initial geometric mapping relationship is calculated based on the corresponding point pairs extracted from the targets deployed in the measurement scene, and includes at least correction parameters for camera eccentricity error, image plane tilt error and optical distortion. Based on the initial geometric mapping relationship, the image information acquired by the rotating linear array camera is mapped to the point cloud acquired by the ground 3D laser scanner to generate an initial colorized image; Using the intensity image generated from the inherent reflection intensity information in the point cloud as a spatial reference, a non-rigid image registration algorithm is used to solve the spatial deformation field from the initial color image to the intensity image, wherein the non-rigid image registration is used to correct the nonlinear residual error that is difficult to eliminate in the initial geometric mapping relationship; The system error of the initial geometric mapping relationship is inverted based on the spatial deformation field, and the initial geometric mapping relationship is nonlinearly corrected to obtain the optimized final geometric mapping relationship.

2. The method according to claim 1, characterized in that, The initial geometric mapping relationship is obtained through the following steps: Based on multiple targets deployed in the measurement scene, multiple sets of corresponding point pairs are extracted from the point cloud data and the image data acquired by the rotating linear array camera; a geometric mapping model is calculated using the multiple sets of corresponding point pairs to obtain the initial geometric mapping relationship.

3. The method according to claim 2, characterized in that, The geometric mapping model is used to describe the geometric relationship between the projection center, rotation center, and image plane of the rotating linear array camera in the object point cloud coordinate system, and includes at least correction parameters for eccentricity error, image plane tilt error, and camera optical distortion.

4. The method according to claim 1, characterized in that: The method as described in claim 1, wherein the step of using the intensity image generated from the inherent reflection intensity information in the point cloud as a spatial reference further comprises: projecting the three-dimensional point cloud onto a unit sphere, rasterizing it at equal angles, and converting the maximum value of the point cloud reflection intensity in each raster into a grayscale value to generate the intensity image.

5. The method according to claim 1, characterized in that: The non-rigid image registration algorithm is a free deformation model based on a control grid; the calculation of the deformation field includes: setting up a control grid on the initial colorized image, using the displacement vector of the grid nodes as the optimization variable, and maximizing the similarity between the initial colorized image and the intensity image through iterative optimization, thereby determining the deformation field.

6. The method according to claim 1, characterized in that: The similarity is measured using the normalized cross-correlation coefficient.

7. The method according to claim 4 or 5, characterized in that: The iterative optimization employs the Levenberg-Marquardt algorithm, and the algorithm's iterative process includes: The control step size parameter of the algorithm is dynamically adjusted based on the descent rate of the objective function in the current iteration.

8. A system, characterized in that, include: 3D laser scanning module, rotating linear array camera module, control and processing module; The rotating linear array camera module is rigidly connected to the three-dimensional laser scanning module and configured to synchronously acquire image data and point cloud data of the scene; the control and processing module stores executable instructions, which, when executed by the processor, implement the geometric mapping relationship optimization method as described in any one of claims 1 to 6, and map the color information of the image acquired by the rotating linear array camera module to the point cloud acquired by the three-dimensional laser scanning module according to the final geometric mapping relationship to generate a color point cloud.

9. The system according to claim 8, characterized in that: The control and processing module is configured to write the final geometric mapping relationship into a configuration file for real-time generation of color point clouds in subsequent multi-station measurement tasks, and to fuse the multi-station color point clouds to generate a three-dimensional color scene model.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the geometric mapping relationship optimization method as described in any one of claims 1 to 7.

Images