A point cloud calibration method and system for plant phenotype extraction

Abstract

The application discloses a kind of point cloud calibration methods and systems for plant phenotype extraction, the method includes the following steps: in plant point cloud acquisition scene, 3D printing cylinder disc and standard color card with known size and explicit plane normal feature are laid out;Extract the point cloud data of cylinder disc area, carry out point cloud scale and direction calibration based on cylinder disc area plane fitting and normal vector calculation;The point cloud data after calibration is cropped to obtain the point cloud of color card area, and a point cloud overhead view is generated;Color block detection is carried out on the point cloud overhead view, and effective color blocks are extracted, the matching relationship between the center points of the reference color blocks of the standard color card and the effective color blocks of the point cloud is constructed, the rotation matrix and the translation vector between the reference color blocks and the point cloud color blocks are calculated based on the two-dimensional rigid transformation model, the reference color blocks are mapped into the point cloud overhead view, and missing color block prediction is carried out;A color correction matrix is constructed to achieve global color calibration.The application can improve the scale, direction and color consistency of point cloud.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and particularly relates to a point cloud calibration method and system for plant phenotype extraction. BACKGROUND

[0002] In plant phenotype analysis, multi-view reconstructed point clouds exist in arbitrary coordinate systems, lacking real-world scale and orientation information, which makes it difficult to quantify key growth parameters such as plant height, crown width, and leaf area. For example, the height of the reconstructed plant model cannot be converted into actual physical units such as centimeters or millimeters; if the spatial orientation of the point cloud is not aligned with the direction of gravity, it will further introduce systematic errors in the calculation of projected leaf area and the analysis of plant erectness. Multi-plant phenotype research relies heavily on color information to identify lesions, evaluate chlorophyll content, and diagnose nutrient status. However, the color of the multi-view reconstructed point cloud is easily disturbed by factors such as natural light changes, differences in shooting angles, and inconsistencies in camera parameters, resulting in color blocks, color differences, and splicing marks on the surface of the reconstructed plant, which cannot reflect the true reflection characteristics of the leaves. This color distortion can directly affect the judgment of leaf health status and even mask the visual features of early diseases, posing a risk of misjudgment for phenotype analysis.

[0003] In terms of geometric scale and orientation recovery, existing methods mainly rely on the following approaches: First, by setting up physical calibration objects such as checkerboards on-site and using their known dimensions to achieve absolute scale recovery. Although this method has high accuracy, it requires manual intervention in the extraction and trimming of calibration objects, resulting in low automation and the inability to identify orientation. Second, by using expensive external equipment such as GNSS / IMU for fusion measurement, which is costly and limited by the accumulation of sensor errors, making it difficult to popularize. Third, by using calibration-free methods, inferring from inherent geometric constraints in the scene (such as planar structures and orthogonal lines). However, such methods have poor robustness and unstable accuracy in unstructured natural scenes such as plants, making it difficult to meet the stringent requirements of scale and orientation consistency in phenotypic measurements. In terms of color correction, mainstream methods also have significant limitations. Global adjustment strategies such as white balance and histogram matching are simple and easy to use, but they cannot adapt to color changes on the plant surface caused by local occlusion and differences in lighting, resulting in harsh correction effects. Methods based on color mapping of overlapping regions can handle local differences, but their effectiveness heavily depends on high-precision point cloud registration, and registration errors are easily amplified into color artifacts. Methods based on physical lighting models or deep learning have shown some potential, but the former is difficult to implement due to model complexity and parameter estimation difficulties, while the latter is limited by the scarcity of training data for plant scenes and insufficient model generalization ability, and has not yet formed a reliable solution. Although existing technologies also use color charts for point cloud color correction, due to the low reflectivity and high color difference characteristics of the color chart material, its performance in point cloud rendering images is more blurred than real images, with lower edge contrast. Some color blocks may even be missing due to occlusion or insufficient point cloud density, resulting in poor or even invalid performance of traditional corner detection methods. Summary of the Invention

[0004] Purpose of the invention: This invention aims to provide a point cloud calibration method and system for plant phenotypic extraction that can improve the consistency of point cloud scale, orientation and color, providing a reliable technical foundation for high-throughput and traceable plant phenotypic analysis.

[0005] Technical solution: The point cloud calibration method for plant phenotypic extraction described in this invention includes the following steps:

[0006] (1) In the plant point cloud acquisition scene, set up a 3D printed cylindrical disk and a standard color card with known size and clear planar normal features, and collect point cloud data;

[0007] (2) Extract point cloud data of the cylindrical disk region, and perform point cloud scale and orientation calibration based on planar fitting and normal vector calculation of the cylindrical disk region;

[0008] (3) The point cloud data after scale and orientation calibration is cropped in the height direction to obtain the point cloud of the color card area and generate a top view of the point cloud;

[0009] (4) Color block detection is performed on the point cloud top view image, effective color blocks are extracted through edge detection and geometric constraints, a matching relationship between the center points of the standard color card reference color blocks and the extracted point cloud effective color blocks is constructed, a rotation matrix and a translation vector between the reference color blocks and the point cloud color blocks are calculated based on a two-dimensional rigid transformation model, the standard color card reference color blocks are mapped into the point cloud top view, and missing color block prediction is performed;

[0010] (5) Color values of all color blocks in the color card after color block prediction are extracted, a color correction matrix is constructed, and the color correction matrix is applied to all color vectors of the plant point cloud to realize global color calibration.

[0011] Preferably, when the point cloud in the cylindrical disc region is extracted, a color filtering range is set according to the color of the cylindrical disc to screen a candidate point set of the cylindrical disc.

[0012] Preferably, in step (2), the point cloud scale calibration based on the plane fitting of the cylindrical disc region comprises projecting the point cloud in the cylindrical disc region onto a two-dimensional plane, fitting a circular shape of the edge of the cylindrical disc, solving the size of the fitted circle, calculating a scale factor based on the actual projection diameter and the fitted diameter of the cylindrical disc, and multiplying all point cloud coordinates by the calculated scale factor to complete the point cloud scale calibration.

[0013] Preferably, in step (2), the point cloud direction calibration based on the plane normal vector calculation of the cylindrical disc region comprises fitting a plane equation of the point cloud in the cylindrical disc region by using a least square method, calculating a unit normal vector of the plane of the cylindrical disc, and then calculating a rotation axis and a rotation angle based on the unit normal vector of the plane of the cylindrical disc and the standard Z-axis direction The rotation axis and the rotation angle are used to construct a rotation matrix, and the rotation matrix is used to realize the point cloud direction calibration.

[0014] Preferably, in step (4), the matching relationship between the standard color card reference color blocks and the center points of the extracted point cloud effective color blocks comprises:

[0015] (a) A standard color card reference coordinate system is established with the center point of the upper left color block of the standard color card as the coordinate origin, the row direction as the horizontal axis, and the column direction as the vertical axis, and the reference center coordinates of the nth row and mth column color block are defined as ; the reference center point coordinates of all color blocks form a reference color block center point coordinate set;

[0016] (b) The geometric center of the four corner points of the effective color block in the point cloud top view is calculated by detecting the four corner points, and the center point coordinates of all effective color blocks form a point cloud color block center point coordinate set;

[0017] (c) Color feature matching is performed, color feature values of the extracted effective color blocks in the point cloud top view are calculated, and the standard color card reference color blocks are matched in a color difference minimization manner to establish a corresponding relationship between the point cloud color blocks and the reference color block center points.

[0018] Preferably, the calculation of the rotation matrix and the translation vector between the reference color block and the point cloud color block based on the two-dimensional rigid transformation model, the mapping of the standard color card reference color block to the point cloud overhead view, and the prediction of the missing color block comprise:

[0019] Based on the matching point pair set, the rotation matrix and the translation vector are solved by a two-dimensional rigid transformation model, so that the error between the center of the point cloud color block and the predicted position of the standard color card reference coordinate system is minimized.

[0020] The fitted rotation matrix and translation vector are used to map the center points of the reference color blocks corresponding to all row and column indexes of the standard color card to the point cloud overhead view, to generate a complete point cloud color block center prediction result.

[0021] The actual color value of the missing color block is calculated by extracting the neighborhood pixel color information centered on the predicted center point.

[0022] Preferably, the two-dimensional rigid transformation model is

[0023] wherein, R represents the rotation matrix, T represents the translation vector; Ci represents the i-th effective color block center point coordinate, Ci represents the reference color block center point coordinate in the standard color card matched with the color feature of the i-th effective color block.

[0024] The point cloud calibration system for plant phenotype extraction provided by the application comprises:

[0025] A point cloud data acquisition module is configured to arrange a 3D printed cylindrical disc with a known size and a clear plane normal feature and a standard color card in a plant point cloud acquisition scene and acquire point cloud data.

[0026] A point cloud scale and direction calibration module is configured to extract a cylindrical disc region point cloud, calculate a scale factor based on the two-dimensional plane projection of the cylindrical disc region point cloud to complete point cloud scale calibration, and perform point cloud direction calibration based on the plane normal feature of the cylindrical disc region point cloud.

[0027] A point cloud color calibration module is configured to perform height direction clipping on the point cloud after scale and direction calibration to obtain a color card region point cloud, generate a point cloud overhead view, perform color block detection on the point cloud overhead view image, extract effective color blocks through edge detection and geometric constraints, construct a matching relationship between the reference color blocks of the standard color card and the extracted point cloud effective color block center points, calculate the rotation matrix and the translation vector between the reference color block and the point cloud color block based on a two-dimensional rigid transformation model, map the standard color card reference color block to the point cloud overhead view, and predict the missing color block to obtain complete color blocks in the color card. The color values of the complete color blocks are extracted, a color correction matrix is constructed, and the color correction matrix is applied to all color vectors of the plant point cloud to realize global color calibration.

[0028] A computer readable storage medium according to the present application, the computer readable storage medium has a computer program stored thereon, the program is executed by the processor to realize the point cloud calibration method for plant phenotype extraction.

[0029] An electronic device according to the present application, comprising a processor and a memory, the memory has a computer program stored thereon, the program is executed by the processor to realize the point cloud calibration method for plant phenotype extraction.

[0030] Advantages: Compared with the prior art, the present application has the following obvious advantages: an automatic point cloud calibration method based on standard geometric body and color card joint calibration is proposed, by arranging a 3D printed cylindrical disc with a known diameter and a plane orientation in the acquisition scene, the scale recovery and direction alignment of the point cloud are realized, then combined with the standard color card detection in the overhead view of the calibrated object point cloud after scale and direction calibration, the rotation angle is directly estimated through the geometric features of the color block, avoiding the dependence on complete corner point detection in the traditional method, the calculation process is stable and efficient; through the rigid transformation model combining rotation correction and translation mapping, even in the case of local missing of the color card, arbitrary change of the shooting angle or uneven distribution of the detected color block, reliable extraction of all color blocks in the color card can still be realized, significantly improving the color card recognition robustness in the case of insufficient point cloud density or partial occlusion; further, accurate measurement color information is extracted, a color correction matrix (CCM) is constructed, and unified correction of the RGB channels of the object point cloud and the plant point cloud is completed. This method is completely based on the geometric and color features of the collected data itself, without any external equipment or prior information, and can realize the consistency improvement of the scale, direction and color of the point cloud, with high universality and automation characteristics. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 The method flowchart of the present application;

[0032] Figure 2 The precision evaluation diagram of the point cloud set after scale and direction calibration of the present application; wherein, figure (a) represents the fitting result of the height direction predicted value and the measured value, figure (b) represents the fitting result of the width direction predicted value and the measured value, and figure (c) represents the fitting result of the length direction predicted value and the measured value. DETAILED DESCRIPTION

[0033] The technical solutions of the present application will be further described below in combination with the drawings.

[0034] As shown in the drawings, Figure 1 A point cloud preprocessing method for plant phenotype extraction according to the present application, comprising the following steps:

[0035] (1) Data acquisition

[0036] A 3D printed cylinder disc and a standard color card of known size are laid out in the plant phenotype scanning area. The 3D printed cylinder disc is a combination of a cylinder and a disc with different diameters, and the disc diameter is greater than the cylinder, facilitating the identification of the normal vector, and the surface color of the 3D printed cylinder disc is significantly different from the background, facilitating color segmentation of the point cloud through RGB attributes. In this embodiment, the disc structure is that the disc is above the cylinder, and both the disc and the cylinder are blue.

[0037] Raw point cloud data with RGB color information is collected, including three-dimensional coordinate points , color information .

[0038] (2) Geometric feature analysis and calibration, specifically including:

[0039] (2.1) Disc area point cloud extraction

[0040] In order to identify the cylinder disc area, the color filtering range is set according to the color of the cylinder disc, and the point set belonging to the disc is screened out .

[0041] (2.2) Plane fitting and normal vector calculation

[0042] The least squares method is used to fit the plane equation , and the unit normal vector of the plane is calculated ; the Z-axis direction is determined in combination with the structural characteristics of the 3D printed cylinder disc. Specifically, if there are no blue point clouds below the disc enclosure frame, the downward direction is the positive direction of the Z-axis, and vice versa, the upward direction is the positive direction of the Z-axis. Preferably, the RANSAC algorithm is used in the plane fitting process to remove outliers, ensuring the robustness of the plane.

[0043] (2.3) Fitting the edge shape of the cylinder disc, calculating the scale factor and performing point cloud scale correction.

[0044] In the fitted plane, the point set of the disc is projected onto a two-dimensional plane. An algebraic model of the disc edge circle is constructed , and the center and radius are solved. The least squares method is used to solve the parameters , and the disc fitting center and the disc fitting radius are obtained, thereby obtaining the disc fitting diameter .

[0045] The scale factor is calculated, and all point coordinates in the point cloud are multiplied by , which completes the geometric scale normalization. Wherein, represents the actual diameter of the disc, represents the disc fitting diameter.​

[0046] (2.4) Direction alignment and rotation correction

[0047] After the scale correction is completed, direction calibration is needed to align the local coordinate system of the point cloud with the world coordinate system. The standard Z-axis direction is set as , and the unit normal vector of the disc plane is , the rotation axis between and is calculated , and the rotation axis is normalized; the rotation angle between the two is calculated , and the rotation matrix is constructed , and the entire point cloud is rotated around the origin to align the Z-axis with the world coordinate system, thereby completing the Z-axis direction consistency correction. This rotation not only makes multiple point clouds have a common direction reference system, but also provides an accurate basis for subsequent crop height, vertical layering, and other analyses.

[0048] (2.5) Geometric accuracy evaluation

[0049] To verify the scale and direction calibration effect, the calibrated point cloud is projected according to the axis, the projected geometric dimensions are extracted, and linear fitting is performed with the manual measurement values to calculate the determination coefficient .

[0050]

[0051] wherein, represents the true measurement value of the i th sample, is the corresponding predicted value, is the average value of all true values, is the total number of samples. The numerator part represents the error sum of squares (residual sum of squares) between the predicted value and the true value, and the denominator part is the total deviation sum of squares of the true value relative to its mean. The larger the value, the closer to 1, indicating that the model has stronger explanation ability for the true value and better fitting effect.

[0052] As shown in Figure 2 , (a) in Figure 2 represents the fitting result of the predicted value and the measured value in the height direction, (b) represents the fitting result of the predicted value and the measured value in the width direction, and (c) represents the fitting result of the predicted value and the measured value in the length direction. The length direction , the width direction , and the height direction are calculated, indicating that the calibration method proposed in the present application has high geometric accuracy.

[0053] (3) Point cloud color calibration, specifically including the following steps: ​

[0054] (3.1) Color card region point cloud preprocessing and initial color block detection.

[0055] The point cloud data after scale and direction calibration is cropped to retain only the region point cloud with a Z-axis height below a certain threshold (e.g., 10 cm), generating a point cloud overhead view image. This is because this region is usually connected to the ground or platform surface and can completely contain the entire pattern of the standard color card. However, due to the low reflectivity and high color difference of the color card material, its performance in the point cloud rendering image is more blurred than the real image, with lower edge contrast. Some color blocks in the color card may even be missing due to occlusion or insufficient point cloud density, resulting in poor or even ineffective results of traditional corner detection methods.

[0056] In this embodiment, the color block outline in the color card overhead view is obtained by Canny edge detection, combined with rectangular features, area threshold, and aspect ratio to remove false detection regions, and effective color blocks are extracted.

[0057] (3.2) Missing color block prediction. Specifically, the following steps are included:

[0058] (3.2.1) Constructing a point cloud color block center point coordinate set and a standard color card reference color block center point coordinate set.

[0059] In the point cloud overhead view, for the detected effective color blocks, obtain their four corner point coordinates. Let the four corner point coordinates of the i-th effective color block be , , then the geometric center position of the effective color block in the point cloud overhead view is defined as the arithmetic mean of the four corner point coordinates . Then the center points of the detected effective color blocks form the point cloud color block center point coordinate set , , where represents the number of effective color blocks actually detected and extracted.

[0060] A standard color card reference coordinate system is established with the center point of the top-left color block of the actual standard color card as the origin, the row direction as the horizontal axis, and the column direction as the vertical axis. In this coordinate system, only the row and column indices of the color blocks are used to describe their spatial structure without introducing any color block size or spacing information. For any standard color card, let it be composed of N rows and M columns of regularly arranged color blocks. The reference center point coordinate of the j-th color block in the n-th row and m-th column in the standard color card is defined as , then the reference center point coordinates of all color blocks in the standard color card form the reference color block center point coordinate set , ; .

[0061] (3.2.2) Matching point cloud color block center points with reference color block center points based on color information.

[0062] For the detection of the extracted effective color block, the color feature values thereof are calculated respectively, and compared with the color features of each reference color block in the standard color card, matching is performed by minimizing the color difference, the corresponding row index and column index of each effective color block in the standard color card are determined, and a matching point pair set between the center points of the reference color blocks and the center points of the point cloud color blocks is formed:

[0063]

[0064] (3.2.3) Based on the color block center matching point, a two-dimensional rigid transformation model is used for rotation translation coefficient fitting.

[0065] The standard color card reference coordinate system and the point cloud overhead view coordinate system satisfy a two-dimensional rigid transformation relationship. Based on the matching point pair set , a two-dimensional rigid transformation model is constructed to determine the rotation matrix and the translation vector :

[0066]

[0067] (3.2.4) Point cloud color card missing color block center point prediction based on rotation translation coefficients.

[0068] For any color block in the standard color card, the reference center point coordinates are , the rotation matrix and the translation vector obtained by fitting are used to map it to the point cloud overhead view, and the predicted center position of the color block in the point cloud overhead view is obtained:

[0069]

[0070] By mapping all the row and column indexes in the standard color card, the center point prediction results of the complete N*M color card in the point cloud image can be obtained, including the color blocks that are not detected.

[0071] (3.2.5) Extraction of actual color values based on predicted point cloud color block center points.

[0072] The color information of the corresponding image pixels is extracted in the neighborhood range of the color block center points obtained by initial color block detection and missing color block prediction, and the actual color value of the color block is calculated.

[0073] The present application directly estimates the rotation angle through the geometric features of the color block, avoiding the dependence of the traditional method on the detection of complete corner points, and the calculation process is stable and efficient. Through the rigid transformation model combining rotation correction and translation mapping, even in the case of local missing of the color card, arbitrary change of the shooting angle or uneven distribution of the detected color blocks, reliable extraction of all color blocks in the color card can still be realized, and the robustness of the color card recognition under the condition of insufficient point cloud density or partial occlusion is significantly improved.

[0074] (3.3) Color correction matrix construction and color calibration

[0075] The RGB average values of all color blocks in the color card after the missing color block prediction are extracted and normalized to construct a measurement color matrix U, and a standard reference value matrix T of the color card is combined to fit a linear transformation matrix V by using a least square method, so that .

[0076] The color vectors of the full point cloud are linearly transformed , the color values are ensured to be in the interval [0, 1] by using np.clip, and the color calibration of the calibration object point cloud and the plant point cloud is completed.

[0077] It has been verified that in multi-crop (rice, soybean, Chinese cabbage, poplar) multi-batch data, the calibrated point cloud can accurately restore the true geometric size. The present application provides a solid data consistency foundation for point cloud data in multi-source fusion, time series analysis and phenotype modeling, and promotes the development of 3D phenotype precision and high throughput.

[0078] Based on the same inventive concept, the point cloud calibration system for plant phenotype extraction provided by the present application comprises:

[0079] A point cloud data acquisition module is used to arrange a 3D printed cylindrical disc with a known size and a clear plane normal feature and a standard color card in a plant point cloud acquisition scene and acquire point cloud data;

[0080] A point cloud scale and direction calibration module is used to extract the point cloud in the cylindrical disc region, calculate a scale factor based on the two-dimensional plane projection of the point cloud in the cylindrical disc region to complete the point cloud scale calibration, and simultaneously perform point cloud direction calibration based on the plane normal feature of the point cloud in the cylindrical disc region;

[0081] A point cloud color calibration module is used to perform height direction clipping on the point cloud after the scale and direction calibration to obtain point cloud region point cloud, generate a point cloud overhead view, perform color card identification on the overhead view image, extract effective color blocks of the color card through edge detection and geometric constraints, construct a matching relationship between the standard color card coordinate system and the center points of the extracted effective color blocks, correct rotation and translation errors by using a rigid body transformation model, realize complete reconstruction of the standard color card in the point cloud overhead view, predict missing color blocks to obtain complete color blocks in the color card, extract color values of the complete color blocks, construct a color correction matrix, apply the color correction matrix to all color vectors of the plant point cloud, and realize global color calibration.

[0082] Based on the same inventive concept, the present application further provides a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and the program is executed by a processor to realize the point cloud calibration method for plant phenotype extraction.

[0083] Based on the same inventive concept, the present application also provides an electronic device comprising a processor and a memory, wherein the memory stores a computer program, and the program is executed by the processor to implement the above-mentioned plant phenotype-oriented point cloud calibration method.

Claims

1. A point cloud calibration method for plant phenotypic extraction, characterized in that, Includes the following steps: (1) In the plant point cloud acquisition scene, set up a 3D printed cylindrical disk and a standard color card with known size and clear planar normal features, and collect point cloud data; (2) Extract point cloud data of the cylindrical disk region, and perform point cloud scale and orientation calibration based on planar fitting and normal vector calculation of the cylindrical disk region; (3) The point cloud data after scale and orientation calibration is cropped in the height direction to obtain the point cloud of the color card area and generate a top view of the point cloud; (4) Perform color block detection on the top view image of the point cloud, extract effective color blocks through edge detection and geometric constraints, construct the matching relationship between the center point of the standard color card reference color block and the extracted effective color block of the point cloud, calculate the rotation matrix and translation vector between the reference color block and the point cloud color block based on the two-dimensional rigid transformation model, map the standard color card reference color block to the top view of the point cloud, and perform missing color block prediction. (5) Extract the color values ​​of the complete color blocks, construct a color correction matrix, and apply it to all color vectors of the plant point cloud to achieve global color calibration.

2. The point cloud calibration method for plant phenotypic extraction according to claim 1, characterized in that, When extracting the point cloud of the cylindrical disk region, the candidate point set of the cylindrical disk is obtained by setting the color filter range according to the color of the 3D printed cylindrical disk.

3. The point cloud calibration method for plant phenotypic extraction according to claim 1, characterized in that, In step (2), point cloud scale calibration is performed based on planar fitting of the cylindrical disk region. This includes projecting the point cloud of the cylindrical disk region onto a two-dimensional plane, fitting the circular shape of the cylindrical disk edge, solving for the size of the fitted circle, calculating the scale factor based on the actual projected diameter of the cylindrical disk and the fitted diameter, and multiplying all point cloud coordinates with the calculated scale factor to complete the point cloud scale calibration.

4. The point cloud calibration method for plant phenotypic extraction according to claim 1, characterized in that, Step (2) involves calibrating the point cloud orientation based on the calculation of the cylindrical disk region plane normal vector. This includes fitting the plane equation of the point cloud in the cylindrical disk region using the least squares method, calculating the unit normal vector of the cylindrical disk plane, and then calibrating the orientation based on the cylindrical disk plane normal vector and the standard Z-axis direction. Calculate the rotation axis and rotation angle to construct a rotation matrix, and apply the rotation matrix to achieve point cloud orientation calibration.

5. The point cloud calibration method for plant phenotypic extraction according to claim 1, characterized in that, The matching relationship between the standard color card reference color patch and the center point of the extracted point cloud effective color patch described in step (4) includes: (a) Establish a reference coordinate system for the standard color chart with the center point of the upper left corner color block as the origin, the horizontal axis along the row direction as the horizontal axis, and the vertical axis along the column direction as the vertical axis. Define the reference center coordinates of the color block in the nth row and mth column as follows: The coordinates of the reference center points of all color blocks constitute the set of reference color block center point coordinates. (b) By detecting the coordinates of the four corner points of the valid color blocks in the top view of the point cloud, the geometric center is calculated, and the coordinates of the center points of all valid color blocks constitute the set of center coordinates of the color blocks in the point cloud. (c) Color feature matching: Calculate the color feature values ​​of the effective color blocks extracted from the top view of the point cloud, perform color difference minimization matching with the reference color blocks of the standard color card, and establish the correspondence between the center points of the point cloud color blocks and the reference color blocks.

6. The point cloud calibration method for plant phenotypic extraction according to claim 5, characterized in that, The step of calculating the rotation matrix and translation vector between the reference color patch and the point cloud color patch based on the two-dimensional rigid transformation model, mapping the standard color card reference color patch onto the point cloud top view, and predicting missing color patches includes: Based on the set of matching point pairs, the rotation matrix and translation vector are solved by a two-dimensional rigid transformation model to minimize the error between the center of the point cloud color block and the predicted position of the standard color card reference coordinate system. Using the fitted rotation matrix and translation vector, the center points of the reference color blocks corresponding to all row and column indices of the standard color chart are mapped to the top view of the point cloud, generating a complete point cloud color block center prediction result; Extract color information of neighboring pixels centered on the predicted center point, and calculate the actual color value of the missing color block.

7. The point cloud calibration method for plant phenotypic extraction according to claim 6, characterized in that, The two-dimensional rigid transformation model is as follows: in, Represents the rotation matrix. Represents the translation vector; This represents the coordinates of the center point of the i-th valid color block. This represents the coordinates of the center point of the reference color patch in the standard color chart that matches the color characteristics of the i-th valid color patch.

8. A point cloud calibration system for plant phenotypic extraction, characterized in that, include: The point cloud data acquisition module is used to deploy 3D printed cylindrical disks and standard color cards with known dimensions and clear planar normal features in the plant point cloud acquisition scenario and acquire point cloud data. The point cloud scale and orientation calibration module is used to extract point cloud data of the cylindrical disk region and perform point cloud scale and orientation calibration based on planar fitting and normal vector calculation of the cylindrical disk region. The point cloud color calibration module is used to crop the point cloud in the height direction after scale and orientation calibration to obtain the point cloud of the color chart region and generate a top view of the point cloud. It then performs color block detection on the top view image, extracting valid color blocks through edge detection and geometric constraints. A matching relationship is established between the center points of the standard color chart reference color blocks and the extracted valid color blocks in the point cloud. Based on a two-dimensional rigid transformation model, the rotation matrix and translation vector between the reference color blocks and the point cloud color blocks are calculated. The standard color chart reference color blocks are mapped onto the top view of the point cloud, and missing color blocks are predicted to obtain complete color blocks in the color chart. Finally, the color values ​​of the complete color blocks are extracted, a color correction matrix is ​​constructed, and it is applied to all color vectors of the plant point cloud to achieve global color calibration.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the point cloud calibration method for plant phenotypic extraction as described in any one of claims 1-7.

10. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a computer program that, when executed by the processor, implements the point cloud calibration method for plant phenotypic extraction as described in any one of claims 1-7.

Images