Image-guided laser point cloud enhanced reconstruction method

Abstract

The present invention belongs to the field of image processing and computer vision technology, and specifically relates to a laser point cloud enhancement and reconstruction method based on image guidance, which first corrects the color image, transforms the original point cloud data based on quaternions, and projects the three-dimensional point cloud data from the three-dimensional space to the two-dimensional image space; then calculates the affine transformation between the color image and the projected depth image, and performs depth enhancement on the two-dimensional image; finally, the two-dimensional image is projected back to the three-dimensional space, and the RGB image information is integrated into the point cloud information to form enhanced six-dimensional data. The present invention has a better effect in maintaining the details, smoothness and integrity of the edges. Through experiments, it is proved that the method proposed by the present invention improves the point cloud resolution and reduces the existence of holes in point cloud enhancement, thereby realizing high-resolution three-dimensional reconstruction of laser point cloud images.

Description

Technical Field

[0001] The invention belongs to the technical field of image processing and computer vision, and specifically relates to a laser point cloud enhanced reconstruction method based on image guidance. Background Art

[0002] Computer vision perception is inseparable from various sensors, each of which has its own advantages and disadvantages. The camera is a passive sensor that can obtain rich visual information, with the advantages of rich imaging details and low cost; but the disadvantage is that it lacks the distance information between the scene and the target, making it difficult to accurately locate the target, and the image quality is easily disturbed by light. LiDAR is an active sensor. The advantage is that it can accurately provide the distance information between the scene and the target, and is not affected by light and bad weather; but the disadvantage is that it lacks visual information such as color texture, and the resolution is low. Each single sensing method has its own advantages and disadvantages, and cannot meet the requirements of the integrity of the perception information required by the unmanned system. At the same time, when one type of sensor fails, the entire unmanned system will be out of control. The multi-sensor fusion method can integrate the information obtained by different sensors, make up for the shortcomings of a single sensor, and still ensure the normal operation of the unmanned system in extreme cases, so that the safety of the unmanned system is greatly improved. Therefore, the image-guided laser point cloud enhanced reconstruction perception method has important research value and application prospects.

[0003] In recent years, many scholars have conducted research on laser point cloud enhancement methods. At present, the main focus is on enhancing the imaging characteristics of the laser radar sensor's point cloud by using interpolation or stretching methods. The Chinese journal "Computers and Modernization" 2021, Vol. 307, No. 3, pp. 101-107 published a paper entitled "Gridding Optimization Algorithm Based on Point Cloud Enhancement". The author Yang Yuhang et al. used bicubic spline interpolation to repair point cloud holes while performing appropriate point cloud downsampling through voxel filtering, and then introduced the moving least squares error function into the point cloud normal calculation to optimize the quality of the point cloud normal vector. The optimized gridding algorithm improves the accuracy of the reconstructed model to a certain extent. However, it still cannot meet the requirements of the integrity of the perceived information required by unmanned systems. Therefore, to achieve high-precision laser point cloud enhancement, it is necessary to study and find more effective, accurate, and real-time technical approaches.

[0004] There are many laser point cloud enhancement algorithms, but they all have various problems, which are mainly reflected in two aspects:

[0005] 1) Algorithm accuracy: When the target is missing or blocked by other objects, the traditional laser point cloud enhancement algorithm will have holes.

[0006] 2) Information integrity: Since the information of a single-band sensor is limited and the contribution of common features is not taken into account, the perception effect is affected, making it more difficult to identify the target object. Summary of the invention

[0007] 1. Technical issues to be resolved

[0008] The technical problem to be solved by the present invention is that the existing point cloud enhancement algorithm cannot enhance targets with various changes in complex scenes. There are many uncertainties in the targets and backgrounds in the actual environment, including the scale, perspective and illumination gradient of the target itself, as well as sudden target occlusion. These situations make it difficult for traditional methods to solve the point cloud enhancement problem under these complex backgrounds.

[0009] (II) Technical solution

[0010] In order to solve the above technical problems, the present invention provides a laser point cloud enhanced reconstruction method based on image guidance, the method comprising:

[0011] Step S1: Initialization, obtaining the original color image and original point cloud data;

[0012] Step S2: Correcting the color image to obtain a corrected color image;

[0013] Step S3: point cloud data processing and projection transformation to obtain a projected depth image;

[0014] Step S4: registering the corrected color image with the projected depth image to obtain a transformed color image;

[0015] Step S5: Deep image enhancement to form enhanced and reconstructed six-dimensional data values.

[0016] Wherein, the method further comprises:

[0017] Step S6: Determine whether there is a shutdown command, if yes, then end; otherwise, enter the next cycle.

[0018] The initialization process of step S1 includes:

[0019] Step S1.1: Collect the original color image g through the camera S ;

[0020] Step S1.2: Collecting original point cloud data L through LiDAR B .

[0021] The process of correcting the color image to obtain the corrected color image in step S2 includes:

[0022] Step S2.1: Through camera calibration, the imaging parameters of the camera can be obtained, including the camera center (u0, v0), focal length (f x ,f y ), scale factor (N x ,N y ), lens distortion coefficient k, rotation matrix R, translation matrix T;

[0023] Step S2.2: The original color image g captured by the camera S Correction is performed to obtain the corrected color image g C ;

[0024] The original color image g in the actual computer frame storage coordinate system S (u s ,v s ) is transformed as follows to obtain the corrected color image point g in the ideal computer frame storage coordinate system C (u d ,v d ), expressed in the following odd-order coordinate system:

[0025]

[0026] in:

[0027]

[0028] Store the actual computer frame in image g S Midpoint (u s ,v s )’s pixel gray value g S (u s ,v s ) gives the corrected color image g C The corresponding point (u d ,v d ) to obtain the corrected color image g C .

[0029] The process of performing point cloud data processing and projection transformation in step S3 to obtain a projected depth image includes:

[0030] Step S3.1: transforming the original point cloud data based on quaternion;

[0031] For each image captured by the camera, the position of the drone (p x ,p y ,p z ) and quaternion parameters (o x ,o y ,o z ,o w):

[0032]

[0033] Since the camera's intrinsic parameter matrix K is:

[0034]

[0035] Then the projection transformation matrix M of the integrated drone attitude and camera parameters is:

[0036] M=KQ (5)

[0037] Step S3.2: The three-dimensional original point cloud data L B Project from the three-dimensional space to the two-dimensional image space to obtain the depth image g B ;

[0038] The computer frame image coordinate system is (u, v), and the world coordinate system is (x w ,y w , z w ), the projection transformation matrix M is a 3*4 matrix, and λ is the scale factor; according to the perspective projection transformation relationship:

[0039]

[0040] Therefore, the depth image g is projected from the three-dimensional space to the two-dimensional image space B The corresponding image coordinates (u,v):

[0041]

[0042] The original three-dimensional point cloud data L B After all the point clouds in the image are transformed, the corresponding (u, v) is obtained, and the depth value of the (u, v) point is assigned to the corresponding depth image point g. B (u,v), we get the projected depth image g B .

[0043] Wherein, the step S4 corrects the color image g C With the projected depth image g B The registration process includes:

[0044] Step S4.1: For the color image g C and the projected depth image g B To perform registration, in the color image g C and the projected depth image g B Select the corresponding N pairs of registration points; calculate the corrected color image g C With the projected depth image g B Affine transformation parameters between ;

[0045] When N ≥ 3, the projected depth image g B and the corrected color image g C The relative parameter transformation matrix P between C←B The calculation is done using the following least squares formula:

[0046] P C←B =C*B T *(B*B T ) -1 (8)

[0047] in,

[0048] Step S4.2: Calculate the corrected color image g C With the projected depth image g B The affine transformation between them and the corrected color image g C Transform the corrected color image g C Transform to the projected depth image coordinate system to obtain the transformed color image g CT ;

[0049] According to the projected depth image g B Midpoint B 0 The odd coordinates of the corrected color image g are calculated C The odd coordinates C of the corresponding point in 0 , and its calculation formula is:

[0050] C 0 =P C←B *B 0 (9)

[0051] in, P C←B is the projected depth image g B and the corrected color image g C The relative parameter transformation matrix between 0 The horizontal coordinate of Its value range is 1, 2, ..., W; the vertical coordinate is Its value range is 1, 2, ..., H, C 0 The horizontal coordinate of Its value range is also 1, 2, ..., W; the vertical coordinate is Its value range is also 1, 2, …, H; W is the width of the image, and H is the height of the image;

[0052] The corrected color image g C Midpoint C 0 The pixel gray value g c (i,j) is assigned the projected depth image g BThe corresponding point B 0 After that, the transformed color image g is obtained. CT .

[0053] The process of performing depth image enhancement in step S5 includes:

[0054] Step S5.1: Transform the color image g CT Guide the projected depth image g B By extracting the structural features of the color image and the depth image, the common global features are obtained, and the obtained structural features are subjected to joint bilateral filtering. Finally, the depth image is enhanced based on the Markov random field method to obtain the enhanced two-dimensional depth image g BE ;

[0055] Step S5.2: The enhanced two-dimensional depth image g BE Project back into three-dimensional space;

[0056] Because z w It can be obtained from the depth image data. Therefore, for any point cloud (u, v, z w ), can be projected back into three-dimensional space (x w ,y w ,z w ), and obtain its position in three-dimensional space;

[0057] Step S5.3: Integrate the RGB image information into the point cloud information to form the enhanced reconstructed six-dimensional data value L E (R,G,B,x w ,y w ,z w ).

[0058] The method combines the advantages of RGB images and laser radar, uses high-precision RGB images to guide low-resolution laser point cloud images, and realizes high-resolution three-dimensional reconstruction of laser point cloud images.

[0059] The laser point cloud enhancement reconstruction method is divided into multiple steps: color image correction, point cloud projection transformation, image enhancement, and projection inverse transformation;

[0060] First, the color image is corrected, the original point cloud data is transformed based on quaternions, and the three-dimensional point cloud data is projected from the three-dimensional space to the two-dimensional image space; then the affine transformation between the color image and the projected depth image is calculated, and the two-dimensional image is depth enhanced; finally, the two-dimensional image is projected back to the three-dimensional space, and the RGB image information is integrated into the point cloud information to form enhanced six-dimensional data.

[0061] Wherein, the camera is arranged on a drone.

[0062] (III) Beneficial effects

[0063] In order to solve the problems of image blur, holes and noise in traditional point clouds, the present invention achieves the purpose of deep image enhancement and hole repair based on structural features and guided by color images without changing the physical structure of the sensor imaging system, thereby improving the quality of the deep image. By extracting the structural features of the color image and the depth image, the common global features are obtained, and the obtained structural features are subjected to joint bilateral filtering. Finally, the deep image is enhanced based on the Markov random field method, so as to achieve low-cost acquisition of deep enhanced images. The present invention has better effects in maintaining the details, smoothness and integrity of the edges. Through experiments, it is proved that the method proposed in the present invention improves the point cloud resolution, reduces the existence of holes, and realizes high-resolution three-dimensional reconstruction of laser point cloud images in terms of point cloud enhancement. After enhancing the laser point cloud shown in FIG2(c) and the point cloud image shown in FIG2(d), the holes can be removed and the resolution of the point cloud can be improved, as shown in FIG2(f) for the enhanced point cloud projection image and FIG2(g) for the color image guided enhanced three-dimensional point cloud. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is a flow chart of the present invention;

[0065] Figure 2(a) to Figure 2(g) It is the experimental result of the point cloud enhancement algorithm of the present invention.

[0066] Among them, Figure 2(a) is the original color image frame; Figure 2(b) is the corrected color image frame; Figure 2(c) is the original laser point cloud data; Figure 2(d) is the projected point cloud image; Figure 2(e) is the corrected color image frame after affine projection transformation; Figure 2(f) is the enhanced point cloud projection image; Figure 2(g) is the three-dimensional point cloud after color image guidance enhancement. DETAILED DESCRIPTION

[0067] In order to make the purpose, content, and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the accompanying drawings and examples.

[0068] In view of the defects of the traditional point cloud enhancement algorithm, the present invention has carried out the following work:

[0069] 1) The advantages of RGB images are high resolution and rich texture details, but the disadvantage is that they do not have depth and distance information; the advantage of laser point clouds is that they have high-precision distance and depth information, but the disadvantage is that they have low resolution and are relatively sparse point clouds. In order to solve the problem of low resolution of point cloud images, this patent combines the advantages of RGB images and laser radar, uses high-precision RGB images to guide low-resolution laser point cloud images, and realizes high-resolution 3D reconstruction of laser point cloud images.

[0070] 2) In order to solve the quality problems such as image blur, holes and noise in the depth map, the present invention starts from the perspective of software. Without changing the physical structure of the sensor imaging system, it is based on structural features and uses color images as a guide to achieve the purpose of depth image enhancement and hole repair, thereby improving the quality of the depth image. By extracting the structural features of the color image and the depth image, the common global features are obtained, and the obtained structural features are jointly bilaterally filtered. Finally, the depth image is enhanced based on the Markov random field method, so that the depth enhanced image can be obtained at a low cost. The present invention has a better effect in maintaining the details, smoothness and integrity of the edges.

[0071] In order to solve the above technical problems, the present invention provides a laser point cloud enhanced reconstruction method based on image guidance, the method comprising:

[0072] Step S1: Initialization, obtaining the original color image and original point cloud data;

[0073] Step S2: Correcting the color image to obtain a corrected color image;

[0074] Step S3: point cloud data processing and projection transformation to obtain a projected depth image;

[0075] Step S4: registering the corrected color image with the projected depth image to obtain a transformed color image;

[0076] Step S5: Deep image enhancement to form enhanced and reconstructed six-dimensional data values.

[0077] Wherein, the method further comprises:

[0078] Step S6: Determine whether there is a shutdown command, if yes, then end; otherwise, enter the next cycle.

[0079] The initialization process of step S1 includes:

[0080] Step S1.1: Collect the original color image g through the camera S ;

[0081] Step S1.2: Collecting original point cloud data L through LiDAR B .

[0082] The process of correcting the color image to obtain the corrected color image in step S2 includes:

[0083] Step S2.1: Through camera calibration, the imaging parameters of the camera can be obtained, including the camera center (u0, v0), focal length (f x ,f y ), scale factor (N x ,N y ), lens distortion coefficient k, rotation matrix R, translation matrix T;

[0084] Step S2.2: The original color image g captured by the camera S Correction is performed to obtain the corrected color image g C ;

[0085] The original color image g in the actual computer frame storage coordinate system S (u s ,v s ) is transformed as follows to obtain the corrected color image point g in the ideal computer frame storage coordinate system C (u d ,v d ), expressed in the following odd-order coordinate system:

[0086]

[0087] in:

[0088]

[0089] Store the actual computer frame in image g S Midpoint (u s ,v s )’s pixel gray value g S (u s ,v s ) gives the corrected color image g C The corresponding point (u d ,v d ) to obtain the corrected color image g C .

[0090] The process of performing point cloud data processing and projection transformation in step S3 to obtain a projected depth image includes:

[0091] Step S3.1: transforming the original point cloud data based on quaternion;

[0092] For each image captured by the camera, the position of the drone (p x ,py ,p z ) and quaternion parameters (o x ,o y ,o z ,o w ):

[0093]

[0094] Since the camera's intrinsic parameter matrix K is:

[0095]

[0096] Then the projection transformation matrix M of the integrated drone attitude and camera parameters is:

[0097] M=KQ (5)

[0098] Step S3.2: The three-dimensional original point cloud data L B Project from the three-dimensional space to the two-dimensional image space to obtain the depth image g B ;

[0099] The computer frame image coordinate system is (u, v), and the world coordinate system is (x w ,y w , z w ), the projection transformation matrix M is a 3*4 matrix, and λ is the scale factor; according to the perspective projection transformation relationship:

[0100]

[0101] Therefore, the depth image g is projected from the three-dimensional space to the two-dimensional image space B The corresponding image coordinates (u,v):

[0102]

[0103] The original three-dimensional point cloud data L B After all the point clouds in the image are transformed, the corresponding (u, v) is obtained, and the depth value of the (u, v) point is assigned to the corresponding depth image point g. B (u,v), we get the projected depth image g B .

[0104] Wherein, the step S4 corrects the color image g C With the projected depth image g B The registration process includes:

[0105] Step S4.1: For the color image g C and the projected depth image g B To perform registration, in the color image g C and the projected depth image gB Select the corresponding N pairs of registration points; calculate the corrected color image g C With the projected depth image g B Affine transformation parameters between ;

[0106] When N ≥ 3, the projected depth image g B and the corrected color image g C The relative parameter transformation matrix P between C←B The calculation is done using the following least squares formula:

[0107] P C←B =C*B T *(B*B T ) -1 (8)

[0108] in,

[0109] Step S4.2: Calculate the corrected color image g C With the projected depth image g B The affine transformation between them and the corrected color image g C Transform the corrected color image g C Transform to the projected depth image coordinate system to obtain the transformed color image g CT ;

[0110] According to the projected depth image g B Midpoint B 0 The odd coordinates of the corrected color image g are calculated C The odd coordinates C of the corresponding point in 0 , and its calculation formula is:

[0111] C 0 =P C←B *B 0 (9)

[0112] in, P C←B is the projected depth image g B and the corrected color image g C The relative parameter transformation matrix between 0 The horizontal coordinate of Its value range is 1, 2, ..., W; the vertical coordinate is Its value range is 1, 2, ..., H, C 0 The horizontal coordinate of Its value range is also 1, 2, ..., W; the vertical coordinate is Its value range is also 1, 2, …, H; W is the width of the image, and H is the height of the image;

[0113] The corrected color image g C Midpoint C 0 The pixel gray value g c (i,j) is assigned the projected depth image g B The corresponding point B 0 After that, the transformed color image g is obtained. CT .

[0114] The process of performing depth image enhancement in step S5 includes:

[0115] Step S5.1: Transform the color image g CT Guide the projected depth image g B By extracting the structural features of the color image and the depth image, the common global features are obtained, and the obtained structural features are subjected to joint bilateral filtering. Finally, the depth image is enhanced based on the Markov random field method to obtain the enhanced two-dimensional depth image g BE ;

[0116] Step S5.2: The enhanced two-dimensional depth image g BE Project back into three-dimensional space;

[0117] Because z w It can be obtained from the depth image data. Therefore, for any point cloud (u, v, z w ), can be projected back into three-dimensional space (x w ,y w ,z w ), and obtain its position in three-dimensional space;

[0118] Step S5.3: Integrate the RGB image information into the point cloud information to form the enhanced reconstructed six-dimensional data value L E (R,G,B,x w ,y w ,z w ).

[0119] The method combines the advantages of RGB images and laser radar, uses high-precision RGB images to guide low-resolution laser point cloud images, and realizes high-resolution three-dimensional reconstruction of laser point cloud images.

[0120] The laser point cloud enhancement reconstruction method is divided into multiple steps: color image correction, point cloud projection transformation, image enhancement, and projection inverse transformation;

[0121] First, the color image is corrected, the original point cloud data is transformed based on quaternions, and the three-dimensional point cloud data is projected from the three-dimensional space to the two-dimensional image space; then the affine transformation between the color image and the projected depth image is calculated, and the two-dimensional image is depth enhanced; finally, the two-dimensional image is projected back to the three-dimensional space, and the RGB image information is integrated into the point cloud information to form enhanced six-dimensional data.

[0122] Wherein, the camera is arranged on a drone.

[0123] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A laser point cloud enhanced reconstruction method based on image guidance, characterized in that: The method comprises: Step S1: Initialization, obtaining the original color image and original point cloud data; Step S2: Correcting the color image to obtain a corrected color image; Step S3: point cloud data processing and projection transformation to obtain a projected depth image; Step S4: registering the corrected color image with the projected depth image to obtain a transformed color image; Step S5: Deep image enhancement to form enhanced and reconstructed six-dimensional data values; The process of correcting the color image to obtain the corrected color image in step S2 includes: Step S2.1: Through camera calibration, the imaging parameters of the camera can be obtained, including the camera center (u0, v0), focal length (f x ,f y ), scale factor (N x ,N y ), lens distortion coefficient k, rotation matrix R, translation matrix T; Step S2.2: The original color image g captured by the camera S Correction is performed to obtain the corrected color image g C ; The original color image g in the actual computer frame storage coordinate system S (u s ,v s ) is transformed as follows to obtain the corrected color image point g in the ideal computer frame storage coordinate system C (u d ,v d ), expressed in the following odd-order coordinate system: in: The actual computer frame storage image g S Midpoint (u s ,v s )’s pixel gray value g S (u s ,v s ) gives the corrected color image g C The corresponding point (u d ,v d ) to obtain the corrected color image g C； The process of performing point cloud data processing and projection transformation in step S3 to obtain a projected depth image includes: Step S3.1: transforming the original point cloud data based on quaternion; For each image captured by the camera, the position of the drone (p x ,p y ,p z ) and quaternion parameters (o x ,o y ,o z ,o w ): Since the camera's intrinsic parameter matrix K is: Then the projection transformation matrix M of the integrated drone attitude and camera parameters is: M=KQ (5) Step S3.2: The three-dimensional original point cloud data L B Project from the three-dimensional space to the two-dimensional image space to obtain the depth image g B ; The computer frame image coordinate system is (u, v), and the world coordinate system is (x w ,y w , z w ), the projection transformation matrix M is a 3*4 matrix, and λ is the scale factor; according to the perspective projection transformation relationship: Therefore, the depth image g is projected from the three-dimensional space to the two-dimensional image space B The corresponding image coordinates (u,v): The original three-dimensional point cloud data L B After all the point clouds in the image are transformed, the corresponding (u, v) is obtained, and the depth value of the (u, v) point is assigned to the corresponding depth image point g. B (u,v), we get the projected depth image g B ; Wherein, the step S4 corrects the color image g C With the projected depth image g B The registration process includes: Step S4.1: For the color image g C and the projected depth image g B To perform registration, in the color image g C and the projected depth image g B Select the corresponding N pairs of registration points; calculate the corrected color image g C With the projected depth image g B Affine transformation parameters between ; When N ≥ 3, the projected depth image g B and the corrected color image g C The relative parameter transformation matrix P between C←B The calculation is done using the following least squares formula: P C←B =C*B T *(B*B T ) -1 (8) in, Step S4.2: Calculate the corrected color image g C With the projected depth image g B The affine transformation between them and the corrected color image g C Transform the corrected color image g C Transform to the projected depth image coordinate system to obtain the transformed color image g CT ; According to the projected depth image g B Midpoint B 0 The odd coordinates of the corrected color image g are calculated C The odd coordinates C of the corresponding point in 0 , and its calculation formula is: C 0 =P C←B *B 0 (9) in, P C←B is the projected depth image g B and the corrected color image g C The relative parameter transformation matrix between 0 The horizontal coordinate is B x 0 , whose value range is 1,2,…,W; the vertical coordinate is Its value range is 1, 2, ..., H, C 0 The horizontal coordinate of Its value range is also 1, 2, ..., W; the vertical coordinate is Its value range is also 1, 2, …, H; W is the width of the image, and H is the height of the image; The corrected color image g C Midpoint C 0 The pixel gray value g c (i,j) is assigned the projected depth image g B The corresponding point B 0 After that, the transformed color image g is obtained. CT ; The process of performing depth image enhancement in step S5 includes: Step S5.1: Transform the color image g CT Guide the projected depth image g B By extracting the structural features of the color image and the depth image, the common global features are obtained, and the obtained structural features are subjected to joint bilateral filtering. Finally, the depth image is enhanced based on the Markov random field method to obtain the enhanced two-dimensional depth image g BE ; Step S5.2: The enhanced two-dimensional depth image g BE Project back into three-dimensional space; Because z w It can be obtained from the depth image data. Therefore, for any point cloud (u, v, z w ), can be projected back into three-dimensional space (x w ,y w ,z w ), and obtain its position in three-dimensional space; Step S5.3: Integrate the RGB image information into the point cloud information to form the enhanced reconstructed six-dimensional data value L E (R,G,B,x w ,y w ,z w ).

2. The image-guided laser point cloud enhanced reconstruction method according to claim 1, characterized in that: The method further comprises: Step S6: Determine whether there is a shutdown command, if yes, then end; otherwise, enter the next cycle.

3. The image-guided laser point cloud enhanced reconstruction method according to claim 1, characterized in that: The initialization process of step S1 includes: Step S1.1: Collect the original color image g through the camera S ; Step S1.2: Collecting original point cloud data L through LiDAR B .

4. The image-guided laser point cloud enhanced reconstruction method according to claim 1, characterized in that: The method combines the advantages of RGB images and laser radar, uses high-precision RGB images to guide low-resolution laser point cloud images, and realizes high-resolution three-dimensional reconstruction of laser point cloud images.

5. The image-guided laser point cloud enhanced reconstruction method according to claim 1, characterized in that: The laser point cloud enhancement and reconstruction method is divided into multiple steps: color image correction, point cloud projection transformation, image enhancement, and projection inverse transformation; First, the color image is corrected, the original point cloud data is transformed based on quaternions, and the three-dimensional point cloud data is projected from the three-dimensional space to the two-dimensional image space; then the affine transformation between the color image and the projected depth image is calculated, and the two-dimensional image is depth enhanced; finally, the two-dimensional image is projected back to the three-dimensional space, and the RGB image information is integrated into the point cloud information to form enhanced six-dimensional data.

6. The image-guided laser point cloud enhanced reconstruction method according to claim 1, characterized in that: The camera is arranged on the drone.

Images