Method for scanning three-dimensional topography of a narrow space with motion compensation

Abstract

The application discloses a kind of three-dimensional topography scanning methods with motion compensation in narrow space, it is related to three-dimensional topography reconstruction and detail detection technical field.The three-dimensional topography scanning method includes: using coplanar binocular linear array camera and mirror, carries out dynamic line scanning topography measurement based on structured light in narrow space, continuously obtains point cloud profile;Through additional linear array camera, carry out redundant observation;Through image registration, make the two-dimensional image after additional linear array camera and coplanar binocular linear array camera processing form association;Under the assistance of additional linear array camera redundant observation, estimate the relative motion information of line scanning process, realize motion compensation, make line scanning point cloud profile accurate splicing into high-precision three-dimensional point cloud topography.The three-dimensional topography scanning method with motion compensation in narrow space of the above application realizes the efficient scanning of large field of view in narrow space, and solves the motion compensation problem of linear array camera dynamic line scanning, and can reconstruct three-dimensional topography features with high precision.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional topography reconstruction and detail detection technology, and in particular to a three-dimensional topography scanning method with motion compensation in a narrow space. Background Technology

[0002] The reconstruction results of 3D topography measurement are intuitive and rich in feature information. Vision-based 3D topography reconstruction plays an important role in the detailed analysis and detection of traffic and motion, greatly ensuring the safety of people and property, thanks to its advantages of high precision, high resolution, speed and non-contact measurement.

[0003] In practical measurement applications, such as undercarriage scanning and flaw detection for automobiles and 360° circumferential inspection of trains, 3D topography measurements within confined spaces are unavoidable to achieve more comprehensive inspections. When measuring in limited spaces, the measuring instrument is close to the surface being measured. The strict spatial constraints on the working distance make efficient scanning and high-precision 3D reconstruction with a large field of view challenging. Furthermore, the presence of movement within the scene often places higher demands on the dynamics of the measurement.

[0004] In existing technologies, a common approach is to assemble multiple small-field-of-view area scan cameras into a camera array for measurement in confined spaces. This requires stitching together numerous local measurement results, making the entire system complex and cumbersome. To ensure the camera's working distance and field of view requirements, ground pits are typically dug, but this method has limited application scenarios. Furthermore, area scan camera 3D topography measurement is more suited to static environments and is therefore unsuitable for dynamic scanning measurements involving relative motion.

[0005] Line scan cameras, with their one-dimensional imaging capabilities, can achieve extremely high-frequency acquisition, making them ideal for dynamic scanning and measurement of moving objects. However, line scan camera measurements are easily affected by changes in motion, causing significant distortion in the scanned images and point clouds during stitching. Therefore, line scan systems urgently need to address the issues of high-frequency, continuous motion compensation and achieving efficient scanning with a large field of view in confined spaces. Summary of the Invention

[0006] The purpose of this invention is to provide a three-dimensional topography scanning method with motion compensation in a narrow space. It is easy to implement, enables the system to be built and the method to be implemented on-site, and the object to be tested does not need to have a dense texture. It is applicable to the surfaces of various types of industrial products such as the underside of automobiles and trains.

[0007] To achieve the above objectives, the present invention provides a method for three-dimensional topographic scanning in narrow spaces with motion compensation, the specific steps of which are as follows:

[0008] S1. With the assistance of a reflector, the coplanar binocular line array camera achieves the reversal of the imaging optical path, enabling the viewing planes of the two cameras to easily enter narrow spaces and acquire the surface to be measured. With the assistance of the structured light of the projector, dynamic line scanning topography measurement is performed and point cloud contours are continuously acquired.

[0009] S2. An additional line array camera is configured on one side of either line array camera one or line array camera two, and the viewing plane of the additional line array camera does not coincide with the viewing plane of the coplanar binocular line array camera. The surface to be measured is photographed through a reflector to perform redundant observation.

[0010] S3. The line scan images from the additional line scan camera and the line scan images from the coplanar binocular line scan camera are stitched together and preliminarily processed, and then image registration is performed to form a dense association.

[0011] S4. With the assistance of an additional line array camera, the relative motion is reliably estimated by using the imaging equation of the additional line array camera, the point cloud contour coordinates measured by the coplanar binocular line array camera, and the observation time difference of the matching point between different cameras. This provides accurate motion compensation information, enabling the line scan point cloud contours to be stitched together into a high-precision three-dimensional point cloud morphology.

[0012] Preferably, in S1, the dynamic line scan topography measurement using a coplanar binocular line array camera includes the following steps:

[0013] S11. The viewing planes of the coplanar binocular line array camera 1 and the line array camera 2 overlap. They detect narrow spaces through a common viewing plane and use narrow reflectors to turn the end optical path so that the line fields of view of the two cameras are aligned with the surface to be measured. Dynamic line scanning is achieved by relying on the relative motion between the coplanar binocular line array camera and the surface to be measured.

[0014] S12. The projector is fixedly connected to the coplanar binocular linear array camera, and the structured light is projected along the camera's observation direction; the structured light consists of a high-speed switching sine fringe pattern and a white pattern.

[0015] S13. The coplanar binocular linear array camera and the projector are precisely synchronized. When projecting the sinusoidal fringe pattern and the white image, one-dimensional images of the surface to be measured are continuously acquired. The projected sinusoidal fringe pattern is used for phase calculation, and the projected white image is used for background normalization and acquisition of object surface texture.

[0016] S14. Based on the phase information, the coplanar binocular linear array camera is matched, and the coordinates are calculated using the principle of binocular stereo vision to obtain a point cloud contour in the viewing plane; the object under test moves continuously relative to the coplanar binocular linear array camera, and as the camera shoots, a continuous point cloud contour is continuously generated.

[0017] Preferably, in S2, the step of using an additional linear array camera for redundant observation includes:

[0018] S21. The additional line array camera is mounted close to any one of the coplanar binocular line array cameras, but the viewing plane of the additional line array camera is separate from the common viewing plane of the coplanar binocular line array cameras.

[0019] S22, the additional line array camera performs dynamic line scanning of the surface under test through a narrow reflector, acquires images, and provides redundant observation for the coplanar binocular line array camera.

[0020] Preferably, in S3, the image registration step includes:

[0021] S31. Acquire one-dimensional images captured by the additional line scan camera and the coplanar binocular line scan camera when projecting the white image, stitch them together in time sequence to obtain the stitched two-dimensional image, perform noise reduction and grayscale histogram adjustment on the two-dimensional image, and construct an image pyramid.

[0022] S32. Perform translation transformation on the low-resolution image at the top of the image pyramid for initial registration, perform affine transformation on the medium-resolution image in the middle of the image pyramid for further registration, and perform optical flow analysis on the high-resolution image at the bottom of the image pyramid for further precise registration, finally obtaining the dense correlation result of the additional line scan image from the linear array camera and the coplanar binocular line scan image.

[0023] Preferably, in S32, the process of registering images of different resolutions in the image pyramid is as follows:

[0024] In the low-resolution image at the top of the image pyramid, initial registration is performed through image translation transformation; the transformation matrix for the translation transformation is:

[0025]

[0026] Where tr1 and tr2 are the translation amounts of the stitched 2D image from the additional line scan camera in the x-axis and y-axis directions;

[0027] In the medium-resolution image at the middle layer of the image pyramid, registration is further performed through affine transformation of the image. The affine transformation matrix is ​​as follows:

[0028]

[0029] Where sc1 and sc2 are the scaling factors of the stitched 2D image from the additional line scan camera in the x-axis and y-axis directions, ro is the rotation angle of the image, and sh1 and sh2 are the shearing of the stitched 2D image from the additional line scan camera in the x-axis and y-axis directions.

[0030] In the high-resolution images at the lower level of the image pyramid, further precise registration is achieved through optical flow analysis, as shown in the following formula:

[0031]

[0032] Where v is the deformation field obtained from optical flow analysis, which includes the displacement of each pixel in the stitched 2D image from the additional linear scan camera in the x-axis and y-axis directions. I1 represents the gradient of the image, I2 represents the two-dimensional image acquired by either line scan camera 1 or line scan camera 2 in a coplanar binocular line scan camera, and I3 represents the two-dimensional image acquired by an additional line scan camera.

[0033] Preferably, in S4, motion compensation and point cloud stitching include the following steps:

[0034] S41. By reliably estimating the relative motion of the surface under test during each two point cloud contour measurements using the imaging equation of the additional linear array camera, the point cloud contour coordinates measured by the coplanar binocular linear array camera, and the observation time difference of the matching points, the amount of forward and backward movement is obtained. Left and right movement amount and turning rotation D ω ;

[0035] S42. Based on the obtained motion parameters and D ω Motion compensation is performed using a recursive approach. By optimizing motion parameters nonlinearly, the point cloud contours are precisely stitched together to obtain a high-precision 3D point cloud topography.

[0036] Preferably, in S41, the method for estimating relative motion is as follows:

[0037] By utilizing the imaging constraint equations of an additional linear scan camera, constraints are applied to n points of the point cloud contour. Based on a nonlinear optimization method, the motion parameters of the point cloud contour are reliably estimated. and D ω The constraint equations are:

[0038]

[0039] in,

[0040]

[0041]

[0042]

[0043] l i =v i -v 2_i

[0044]

[0045]

[0046] In the formula, i is the i-th point on the cloud outline (1≤i≤n), [X i ,Y i Z i ] T Let u be the coordinates of the i-th point. i ,v i ) and (u 2_i ,v 2_i ) are the pixel coordinates of the i-th point in two-dimensional image I3 and two-dimensional image I2, respectively. i Related to the acquisition time difference between the additional linear array camera and the coplanar binocular linear array camera, [X' i ,Y' i ,Z' i ] T Let R be the camera coordinates of point i when it is captured by the additional linear scan camera 3. i and [T] 1i ,T 2i ,0] T Represent the rotational and translational motions of point i, respectively; (u c ,0) and F c These are the principal point coordinates and focal length of the additional line scan camera, R. c and T c Let represent the rotation matrix and translation vector of the additional linear array camera, respectively.

[0047] Preferably, in S42, the precise stitching of the point cloud contour is as follows:

[0048] Based on the estimated motion parameters and D ω Motion compensation is performed, and the splicing is done recursively, using the following formula:

[0049]

[0050]

[0051] Among them, R j and T j Let R be the rotation and translation transformation matrix when stitching the j-th point cloud contour. j-1 and T j-1 These are the rotation and translation transformation matrices for stitching the (j-1)th point cloud contour, respectively. and D ω_j The motion parameters of the j-th point cloud contour are estimated through nonlinear optimization.

[0052] This invention utilizes the one-dimensional imaging capabilities of line-scan cameras. By designing a structure combining a coplanar dual-line-scan camera with a reflector, it enables dynamic, wide-field-of-view, and efficient three-dimensional topography measurement in narrow spaces, achieving high-resolution point cloud reconstruction. Addressing the impact of motion changes on line-scan camera scanning measurements, this invention proposes a coplanar dual-line-scan measurement system combined with an additional line-scan camera. Employing a continuous, high-frequency motion estimation strategy, and through effective motion compensation and precise point cloud contour stitching, it acquires high-precision three-dimensional point cloud topography. This provides insights and value for high-precision detail inspection in narrow spaces such as the undercarriages of automobiles and trains in the transportation sector.

[0053] The present invention employs the above-mentioned three-dimensional topographic scanning method with motion compensation in narrow spaces, and its advantages are as follows:

[0054] (1) This invention achieves three-dimensional topography measurement and high-resolution point cloud reconstruction in a narrow space by designing a coplanar binocular linear array camera combined with a reflector, so that the working distance and field of view of the system are not limited by space.

[0055] (2) This invention uses an additional linear array camera to perform redundant observations, thereby achieving high-frequency and continuous estimation of motion state, providing motion compensation for accurate stitching of point cloud contours, and obtaining high-precision three-dimensional point cloud morphology.

[0056] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0057] Figure 1 A flowchart illustrating the workflow of a three-dimensional topography scanning method for narrow spaces with motion compensation;

[0058] Figure 2 Here is a system architecture diagram of the proposed method;

[0059] Figure 3 A flowchart for matching and associating line scan images;

[0060] Figure 4 A flowchart for motion compensation and precise stitching of point cloud contours.

[0061] Figure Labels

[0062] 1. Line scan camera one; 2. Line scan camera two; 3. Additional line scan camera; 4. Projector; 5. Mirror; 6. Surface to be measured. Detailed Implementation

[0063] An embodiment will now be described with reference to the accompanying drawings.

[0064] A flowchart illustrating the workflow of a three-dimensional topography scanning method with motion compensation in a narrow space, as shown below. Figure 1 As shown.

[0065] S1, the viewing planes of line scan cameras 1 and 2 can enter a narrow space. With the help of mirror 5, the imaging optical path is redirected and aligned with the surface to be measured. With the assistance of structured light from projector 4, dynamic line scan topography measurement is performed and point cloud contours are continuously acquired. The surface to be measured 6 moves continuously relative to the camera's field of view, confining the measurement system to a narrow space, such as... Figure 2 As shown.

[0066] S11, the viewing planes of line scan camera 1 and line scan camera 2 coincide with each other. They detect the narrow space through the common viewing plane and use the narrow reflector to turn the end optical path so that the line field of view of the two cameras is aligned with the surface to be measured. Dynamic line scanning is achieved by relying on the relative movement of the two cameras and the surface to be measured.

[0067] S12, the projector 4 is fixedly connected to the line array camera 1 and the line array camera 2 in the coplanar binocular camera, and projects structured light along the observation direction of the camera, wherein the structured light is a high-speed switching sine fringe pattern and white pattern.

[0068] The reasons for using structured light-assisted measurement with a projector are as follows:

[0069] For the surfaces of most industrial products, there are often no very dense and rich texture features. Structured light can help achieve high-resolution point cloud measurement without relying on the texture of the object's surface, thus improving the practicality of the method.

[0070] S13, Line Array Camera 1 and Line Array Camera 2 are precisely synchronized with Projector 4. While projecting sinusoidal fringe patterns and white images, they continuously acquire one-dimensional images of the surface of the object under test. The projected sinusoidal fringe patterns are used for phase calculation, and the projected white images are used for background normalization and acquisition of object surface texture.

[0071] In this process, the projected sinusoidal fringe pattern is modulated by the surface under test and acquired by a coplanar binocular line scan camera. Phase information is calculated from the acquired fringe pattern, and binocular matching is performed between line scan camera 1 and line scan camera 2 based on the phase information. When the white image is projected by projector 4, the line scan camera acquires grayscale information such as texture of the surface under test, realizing a background normalization process. This removes background interference for the phase calculation based on the sinusoidal fringe pattern, improving the accuracy of the phase calculation. Simultaneously, the texture information acquired by line scan camera 1 or line scan camera 2 during the projection of the white image plays a crucial role in the image registration process in step S3.

[0072] S14. Based on the phase information, phase matching is performed on line scan camera 1 and line scan camera 2, and coordinate calculation is performed using the principle of binocular stereo vision to obtain a point cloud contour in the viewing plane. The object under test moves continuously relative to line scan camera 1 and line scan camera 2, and a continuous point cloud contour is generated as the cameras continuously capture images. Due to the high-frequency and high-resolution acquisition characteristics of the line scan cameras, high-precision and high-density point cloud contours can still be generated even with relative motion.

[0073] S2, the additional line array camera 3 is located near line array camera 1 or line array camera 2, and its viewing plane does not coincide with the viewing plane of the first two cameras. It takes pictures of the surface to be measured through a reflector to provide redundant observation.

[0074] The additional line scan camera 3 is placed close to the line scan camera 1 or the line scan camera 2 to keep the viewing angles of the cameras basically consistent, improve the similarity of the line scan images, and help with the image registration operation in step S3.

[0075] In this embodiment, the additional line scan camera 3 is positioned close to the second line scan camera 2, such as... Figure 2 As shown. All the following steps are performed under this premise, especially in step S3, where image registration refers to registering the two-dimensional image acquired by the additional line scan camera 3 with the two-dimensional image acquired by the second line scan camera 2. It should be noted that in other embodiments, the additional line scan camera 3 is located close to the first line scan camera 1, and the final effect is exactly the same.

[0076] S21. The additional line scan camera 3 is mounted close to the second line scan camera 2, but the viewing plane of the additional line scan camera 3 is separate from the common viewing plane of the first line scan camera 1 and the second line scan camera 2.

[0077] S22, the additional line array camera 3 performs dynamic line scanning of the surface under test through the narrow reflector 5, acquires images, and provides redundant observations for line array camera 1 and line array camera 2.

[0078] S3. The line scan images from the additional line scan camera 3 and the line scan images from the second line scan camera 2 are stitched together and preliminarily processed. Then, image registration is performed to finally form a dense association. The specific process is as follows: Figure 3 As shown.

[0079] S31. When the white image is projected on the projector 4, the one-dimensional images acquired by the additional line scan camera 3 and the second line scan camera 2 are obtained and stitched together in time sequence to obtain the stitched two-dimensional image. The two-dimensional image is then subjected to noise reduction and grayscale histogram adjustment processing, and image pyramids are constructed respectively.

[0080] The image pyramid is obtained by continuously downsampling the original image, with lower resolution at higher levels and higher resolution at lower levels. Using an image pyramid can improve the speed of image registration while ensuring sufficiently high accuracy and reliability.

[0081] S32. Perform translation transformation on the low-resolution image at the top of the image pyramid for initial registration, perform affine transformation on the medium-resolution image in the middle of the image pyramid for further registration, and perform optical flow analysis on the high-resolution image at the bottom of the image pyramid for further precise registration. Finally, obtain the dense correlation result between the two-dimensional image acquired by the additional line scan camera 3 and the two-dimensional image acquired by the second line scan camera 2.

[0082] Because the viewing plane of the additional line scan camera 3 does not coincide with the viewing plane of the second line scan camera 2, there is a time difference in the scanning of the same point by the two cameras; furthermore, the positions of the additional line scan camera 3 and the second line scan camera 2 are different, resulting in a deviation in the imaging position of the same point on the camera sensor. Therefore, the two-dimensional image I3 obtained by the additional line scan camera 3 is inconsistent with the two-dimensional image I2 obtained by the second line scan camera 2, most notably in the translational deviation between the two images.

[0083] Therefore, applying the image translation transformation to the lowest resolution image for initial registration can quickly eliminate the effects of the most obvious deviations, improve speed and efficiency, and enhance the reliability of subsequent registration.

[0084] The matrix for the translation transformation is:

[0085]

[0086] In the formula, tr1 and tr2 are the translation amounts of the stitched two-dimensional image I3 of the additional linear scan camera 3 in the x-axis and y-axis directions;

[0087] Furthermore, due to the uncertainty of motion and subtle differences in the camera's observation angle, the registration of I3 and I2 requires consideration of more complex image transformations. Therefore, an affine transformation is introduced to register medium-resolution images, further improving the registration accuracy.

[0088] The matrix incorporating the affine transformation is as follows:

[0089]

[0090] In the formula, sc1 and sc2 are the scaling factors of the two-dimensional image I3 in the x-axis and y-axis directions, ro is the rotation angle of the image, and sh1 and sh2 are the shearing of the two-dimensional image I3 in the x-axis and y-axis directions.

[0091] In addition, when more complex perturbations occur during motion, image deformation becomes more complex, mainly consisting of small-scale, irregular, non-rigid deformations. Optical flow analysis can effectively address these small-scale, irregular, non-rigid deformations, achieving more precise registration.

[0092] Therefore, optical flow analysis is used to register high-resolution images, and the specific formula is as follows:

[0093]

[0094] In the formula, v represents the deformation field obtained from optical flow analysis, which includes the displacement of each pixel in the stitched two-dimensional image from the additional linear scan camera 3 in the x-axis and y-axis directions. This indicates that the gradient of the image is being calculated.

[0095] Through the above transformation process, the pixel coordinates (u,v) of the pixels in the two-dimensional image I3 are converted to (u',v'), which are the pixel coordinates of the matching points in the two-dimensional image I2. This transformation is then applied to all pixels in the two-dimensional image I3, ultimately yielding the dense association result between the two-dimensional images I2 and I3.

[0096] S4. With the assistance of redundant observations from the additional line-scan camera 3, the relative motion is reliably estimated using the imaging equation of the additional line-scan camera 3, the point cloud contour coordinates measured by line-scan camera 1 and line-scan camera 2, and the observation time difference of matching points on different cameras. This provides accurate motion compensation information, enabling the line-scan point cloud contours to be stitched together into a high-precision three-dimensional point cloud morphology. The overall process is as follows: Figure 4 As shown.

[0097] S41. Using the imaging equation of the additional linear scan camera 3, constraints are formed on n points of the point cloud contour. Based on the nonlinear optimization method, the motion parameters of the point cloud contour are reliably estimated, that is, the relative motion of the measured surface during each two point cloud contour measurements, including the forward and backward movement. Left and right movement amount and turning rotation D ω .

[0098] The imaging equation for the additional line scan camera 3 is as follows:

[0099]

[0100] in,

[0101]

[0102]

[0103]

[0104] l i =vi -v 2_i

[0105]

[0106]

[0107] In the formula, i is the i-th point on the cloud outline (1≤i≤n), [X i ,Y i Z i ] T Let u be the coordinates of the i-th point. i ,v i ) and (u 2_i ,v 2_i ) are the pixel coordinates of the i-th point in two-dimensional image I3 and two-dimensional image I2, respectively. i Related to the acquisition time difference between the additional line scan camera 3 and the second line scan camera 2, [X' i ,Y' i ,z' i ] T Let R be the camera coordinates when point i is captured by the additional linear scan camera 3. i and [T] 1i ,T 2i ,0] T Represent the rotational and translational motions of point i, respectively; (u c ,0) and F c These are the principal point coordinates and focal length of the additional line scan camera 3, R. c and T c These represent the rotation matrix and translation vector of the additional linear array camera 3, respectively, used to describe the camera's attitude and position in the world coordinate system.

[0108] S42. Based on the obtained motion parameters and D ω Motion compensation is performed using a recursive approach. By optimizing motion parameters nonlinearly, the point cloud contours are precisely stitched together to obtain a high-precision 3D point cloud topography.

[0109] The splicing is performed using a recursive method, as shown in the following formula:

[0110]

[0111]

[0112] Among them, R j and T j Let R be the rotation and translation transformation matrix when stitching the j-th point cloud contour. j-1 and T j-1These are the rotation and translation transformation matrices for stitching the (j-1)th point cloud contour, respectively. and D ω_j The motion parameters of the j-th point cloud contour are estimated through nonlinear optimization.

[0113] Therefore, the present invention adopts the above-mentioned three-dimensional topography scanning method with motion compensation in narrow spaces. By using a coplanar dual-line array camera combined with a reflector structure, efficient scanning measurement in narrow spaces is achieved. Based on the motion estimation strategy assisted by the additional line array camera, effective motion compensation is achieved, and high-precision three-dimensional topography is obtained. This provides ideas and value for high-precision detail detection in narrow spaces such as the undercarriages of automobiles and trains in the transportation field.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for scanning the three-dimensional topography of a narrow space with motion compensation, characterized in that, Includes the following steps: S1. The surface to be measured is acquired by using a coplanar binocular line array camera and a reflector. With the assistance of structured light from a projector, dynamic line scanning morphology measurement is performed to continuously acquire point cloud contours. S2. An additional linear array camera is placed on one side of the coplanar binocular linear array camera, and the viewing plane of the additional linear array camera does not coincide with the viewing plane of the coplanar binocular linear array camera. The surface to be measured is photographed through a reflector to perform redundant observation. S3. The line scan images from the additional line scan camera and the line scan images from the coplanar binocular line scan camera are stitched together and preliminarily processed, and then image registration is performed to form a dense association. S4. With the assistance of an additional line scan camera, the relative motion is reliably estimated, providing accurate motion compensation information, so that the line scan point cloud contours can be stitched together into a high-precision three-dimensional point cloud morphology. S4 is as follows: S41. By using the imaging equation of the additional linear array camera, the point cloud contour coordinates measured by the coplanar binocular linear array camera, and the observation time difference of the matching points, a reliable estimate of the relative motion is made, obtaining the relative motion of the surface to be measured during each two point cloud contour measurements, i.e., the forward and backward movement. , left and right movement amount and turning rotation ; S42. Based on the obtained motion parameters , and Motion compensation is performed using a recursive approach. Motion parameters are optimized nonlinearly to accurately stitch together point cloud contours and obtain high-precision 3D point cloud topography. In S41, the method for estimating relative motion is as follows: Using the imaging equation of an additional linear scan camera, the point cloud contour is... By establishing constraints at each point, and using a nonlinear optimization method, the motion parameters of the point cloud contour can be reliably estimated. , and The constraint equations are: in, In the formula, The first cloud outline Point, among which, , For the first The coordinates of the point and The first Points in a two-dimensional image and two-dimensional images pixel coordinates on Related to the acquisition time difference between the additional linear array camera and the coplanar binocular linear array camera, For the first The camera coordinates of the point when it is captured by the attached line scan camera. and They represent Rotational and translational motion of a point; and These are the principal point coordinates and focal length of the additional line scan camera, respectively. and Let represent the rotation matrix and translation vector of the additional linear array camera, respectively.

2. The method for scanning the three-dimensional topography of a narrow space with motion compensation according to claim 1, characterized in that, The coplanar binocular line scan camera includes a line scan camera 1 and a line scan camera 2 whose viewing planes overlap.

3. The method for scanning the three-dimensional topography of a narrow space with motion compensation according to claim 2, characterized in that, In S1, the dynamic line scan topography measurement method is as follows: S11. Line scan camera 1 and line scan camera 2, which use coplanar binocular line scan cameras, detect the narrow space through a common viewing plane and use a narrow reflector to turn the end optical path so that the line field of view of line scan camera 1 and line scan camera 2 is aligned with the surface to be measured. Dynamic line scanning is achieved by relying on the relative motion between line scan camera 1, line scan camera 2 and the surface to be measured. S12. The projector is fixedly connected to the coplanar binocular linear array camera, and structured light is projected along the observation direction of the coplanar binocular linear array camera; wherein, the structured light is a high-speed switching sine fringe pattern and white pattern. S13. The coplanar binocular linear array camera and the projector are precisely synchronized. When projecting the sinusoidal fringe pattern and the white image, one-dimensional images of the surface of the object under test are continuously acquired. The projected sinusoidal fringe pattern is used for phase calculation, and the projected white image is used for background normalization and acquisition of object surface texture. S14. Based on the phase information, the coplanar binocular linear array camera is matched, and the coordinates are calculated using the principle of binocular stereo vision to obtain a point cloud profile in the viewing plane; the object under test moves continuously relative to the coplanar binocular linear array camera, and as the camera continuously takes pictures, a continuous point cloud profile is generated.

4. The method for scanning the three-dimensional topography of a narrow space with motion compensation according to claim 3, characterized in that, In S2, the additional linear array camera performs redundant observations as follows: S21. An additional line scan camera is mounted on one side of line scan camera one or line scan camera two, and the viewing plane of the additional line scan camera is separate from the common viewing plane of the coplanar binocular line scan camera. S22, the additional line array camera performs dynamic line scanning of the surface under test through a narrow reflector, acquires images, and provides redundant observation for the coplanar binocular line array camera.

5. A three-dimensional topographic scanning method with motion compensation in a narrow space according to claim 4, characterized in that, In S3, the image registration is as follows: S31. Acquire one-dimensional images captured by the additional line scan camera and the coplanar binocular line scan camera when projecting the white image, stitch them together in time sequence to obtain the stitched two-dimensional image, perform noise reduction and grayscale histogram adjustment on the two-dimensional image, and construct image pyramids respectively. S32. Perform translation transformation on the low-resolution image at the top of the image pyramid for initial registration, perform affine transformation on the medium-resolution image in the middle of the image pyramid for further registration, and perform optical flow analysis on the high-resolution image at the bottom of the image pyramid for further precise registration, finally obtaining the dense correlation result of the additional line scan image from the linear array camera and the coplanar binocular line scan image.

6. The method for scanning the three-dimensional topography of a narrow space with motion compensation according to claim 5, characterized in that, In S32, the images in the image pyramid are registered, as follows: In the low-resolution image at the top of the image pyramid, initial registration is performed through image translation transformation; the transformation matrix for the translation transformation is: in, and To stitch together 2D images from an additional line scan camera shaft and Translation along the axial direction; In the medium-resolution image at the middle layer of the image pyramid, registration is further performed through affine transformation of the image. The affine transformation matrix is ​​as follows: in, and To stitch together 2D images from an additional line scan camera shaft and Scaling factor in the axial direction, The rotation angle of the image. and To stitch together 2D images from an additional line scan camera shaft and axial shear; In the high-resolution images at the lower level of the image pyramid, further precise registration is achieved through optical flow analysis, as shown in the following formula: in, The deformable field is obtained from optical flow analysis, and includes each pixel in the stitched 2D image from the additional linear scan camera. shaft and Displacement in the axial direction, This indicates that the gradient of the image is being calculated. This refers to the two-dimensional image acquired by either line scan camera one or line scan camera two in a coplanar binocular line scan camera system. This represents a two-dimensional image acquired by an additional linear scan camera.

7. The method for three-dimensional topographic scanning of a narrow space with motion compensation according to claim 1, characterized in that, In S42, the precise stitching of the point cloud outline is as follows: Based on the estimated motion parameters , and Motion compensation is performed, and the splicing is done recursively, using the following formula: in, and The first Rotation and translation transformation matrices for stitching together point cloud contours and The first Rotation and translation transformation matrices for stitching together point cloud contours , and It is the first The motion parameters of the point cloud contour are estimated through nonlinear optimization.

Images