A high-precision AR-HUD distortion calibration method
The high-precision AR-HUD distortion calibration method addresses errors in existing methods by assigning code values to projection dot matrices, determining corresponding points, and calculating pre-distortion parameters, resulting in improved accuracy and speed over structured light-based solutions.
Patent Information
- Application Number
- CN202410934462.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-07-12
AI Technical Summary
Existing AR-HUD distortion calibration methods suffer from unavoidable errors in reprojection-based solutions, difficulties in estimating and eliminating errors in homography-based solutions, and a speed-accuracy tradeoff in structured light-based solutions.
A high-precision AR-HUD distortion calibration method that assigns code values to each point in the projection dot matrix, uses these values to determine corresponding points in the sample and reference dot matrices, and calculates pre-distortion parameters based on positional relationships, bypassing reprojection and homography calculations.
This method achieves higher accuracy and speed compared to structured light-based methods by eliminating complex encoding and decoding processes, while maintaining precision without the errors associated with reprojection and homography-based approaches.
Smart Images

Figure CN118505576B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of image processing, and in particular relates to a high-precision AR-HUD distortion calibration method. Background Art
[0002] Distortion calibration of existing AR-HUD can be divided into three categories: reprojection-based, homography-based, and structured light-based. However, due to the existence of distortion, the reprojection-based solution has unavoidable errors, the homography-based solution may have errors that are difficult to estimate and eliminate, and the structured light-based solution has the contradiction between speed and accuracy, such as high accuracy but slow speed, low accuracy but fast speed.
[0003] like Figure 5 As shown, the optical path of AR-HUD is complex, with multiple transmissions and reflections. In view of the uniformity of transmission, reflection and materials such as windshield, the tolerance of the installation of various devices and the free-form surface of the windshield, the final imaging usually has distortion that needs to be calibrated.
[0004] Existing AR-HUD distortion calibration solutions can be divided into three categories:
[0005] 1) Reprojection-based schemes. This type of scheme estimates the reprojection relationship by using the correspondence between the projection lattice and the sample lattice, and uses this relationship to transfer the reference lattice to the original projection image. Then, the pre-distortion parameters are calculated based on the difference between the reference lattice and the projection lattice. However, due to the existence of distortion, there are unavoidable errors in the calculation of the reprojection relationship, as shown in the following example: Figure 2 The projection lattice and Figure 3 As shown in the sample dot matrix in : the projection dot matrix is regularly arranged, while the sample dot matrix is irregularly distributed, and it is difficult to find an accurate relationship to ensure point-to-point overlap between the two.
[0006] 2) Based on homography. This type of solution usually determines the homography relationship by using the correspondence between the reference dot matrix and the projection dot matrix, and uses this relationship to transfer the sample dot matrix to the original projection image, and then uses the difference between it and the projection dot matrix to calculate the pre-distortion parameters. However, this type of solution may have errors that are difficult to estimate and eliminate, such as Figure 3 As shown in the figure: the reference point matrix is within the area determined by the sample point matrix, and the homography relationship can usually ensure accuracy within and near the area where the points for calculating the homography relationship are located. Therefore, using the homography relationship to convert the sample point matrix outside the reference point matrix area may result in errors that are difficult to estimate and thus difficult to eliminate.
[0007] 3) Based on structured light. CN117872602A, which is the most similar to this patent, is an AR-HUD distortion calibration method based on structured light. It uses the decoded value of structured light to determine the correspondence of the reference dot matrix on the projected original image, and calculates the distortion calibration value of the AR-HUD based on the difference between these corresponding points and the projected dot matrix. Although this type of solution can bypass the calculation of the reprojection relationship and homography relationship in the first two types of solutions, there is still a contradiction between accuracy and speed as a whole. Structured light can be roughly divided into spatial structured light and temporal structured light, among which,
[0008] a. Spatial structured light is encoded and decoded according to the special grayscale distribution in the image space (such as pseudo-random array). Its advantage is that it can complete encoding and decoding with the help of a picture. Its disadvantage is that its minimum encoding and decoding unit needs to be determined by a local area, so its encoding density is difficult to be the same as the size of the original image. In addition, due to the introduction of environmental noise, it is difficult to obtain its precise coordinates when decoding the area. Since it can use a single image for encoding and decoding, it has a speed advantage in AR-HUD distortion calibration that requires processes such as projecting images, acquiring images, and decoding.
[0009] b. Temporal structured light is to encode the code value of the same position in the image into different images (i.e., the coding sequence diagram). When decoding, it is necessary to obtain the sequence image and decode it according to certain rules. This method overcomes the shortcomings of spatial structured light, can achieve a decoding density that is the same as the image resolution, and with the help of interpolation and other methods, its accuracy can reach sub-pixel level. The disadvantage is that it usually requires multiple sequence images to complete. In AR-HUD distortion calibration that requires projection images, image acquisition and decoding processes, the process is relatively time-consuming. Summary of the invention
[0010] The object of the present invention is to provide a high-precision AR-HUD distortion calibration method, which can skip the calculation of reprojection parameters and homography parameters to solve the problems in the prior art raised in the above background technology.
[0011] To achieve the above object, the present invention adopts the following technical solution: a high-precision AR-HUD distortion calibration method, comprising:
[0012] Loading information, including the original projection image and its feature type, the rows, columns and coordinate information of the projection dot matrix, and the calibration relationship between the viewpoint camera and the virtual image in front of the vehicle or the reference dot matrix;
[0013] Give each point in the projection lattice a code value of a specific rule to form a code value mapping of the projection lattice;
[0014] Projecting an image, displaying the original projection image;
[0015] Shooting an image, collecting a projection image through a viewpoint camera, and obtaining a sample image including a sample dot matrix;
[0016] Detect the image and identify the sample dot matrix in the sample image;
[0017] Determine the target area, and determine the display area on the virtual image in front of the vehicle or the sample image based on the detected sample dot matrix;
[0018] Fitting or interpolating the dot matrix in the target area, and generating a code value matrix for the entire target area based on the known dot matrix code values;
[0019] Find the corresponding point matrix, use the code value matrix in the target area to match the coordinates of the sample point matrix or the reference point matrix, and determine the corresponding point matrix of the sample point matrix or the reference point matrix on the projection original image;
[0020] The calibration parameters are calculated, and based on the position difference between the corresponding dot matrix and the projected dot matrix, the pre-distortion parameters of the AR-HUD are calculated, and finally the pre-distortion matrix is obtained.
[0021] Preferably, the code value assignment method of the specific rule includes: directly mapping the position of the point in the dot matrix to a predefined code value sequence.
[0022] Preferably, the method further comprises: coding other dot matrices outside the projection dot matrix according to a rule corresponding to the projection dot matrix or a different rule, so as to maintain the corresponding relationship of the code values.
[0023] Preferably, the sample image is obtained by photographing the image displayed by the AR-HUD with a viewpoint camera.
[0024] Preferably, the findCirclesGrid function in the OpenCV library is used to perform spot detection in the detection image to identify the sample dot matrix.
[0025] Preferably, the determining of the target area also includes calculating the intersection of the origin of the viewpoint camera coordinate system and the line connecting each point in the sample dot matrix with the virtual image plane, so as to determine the corresponding position of the sample dot matrix on the virtual image in front of the vehicle.
[0026] Preferably, the fitting or interpolation uses a cubic spline interpolation method to perform two-dimensional space interpolation on the dot code values in the target area.
[0027] Preferably, in the step of finding the corresponding lattice, the largest inscribed rectangle is found within the envelope of the intersection lattice, and the reference lattice is determined accordingly.
[0028] Preferably, in the calculation of the calibration parameters, a bilinear interpolation method is used to interpolate the pre-distortion parameters to cover every point of the entire projection original image.
[0029] Preferably, it also includes: using the obtained pre-distortion matrix to perform real-time or periodic distortion compensation on the AR-HUD system.
[0030] Technical effects and advantages of the present invention: Compared with the prior art, the high-precision AR-HUD distortion calibration method proposed in the present invention has the following advantages:
[0031] The present invention has the same advantages as the scheme based on structured light: the calculation of reprojection parameters and homography parameters can be skipped. Compared with the scheme based on structured light, the present scheme does not require the use of complex encoding and decoding based on grayscale distribution differences, and has an accuracy advantage over spatial structured light and a speed advantage over temporal structured light. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 It is a schematic diagram of an AR-HUD system in the prior art;
[0033] Figure 2 This is an example of a projection original image (circular spot array image) in the prior art;
[0034] Figure 3 This is an example diagram of a dot matrix under a viewpoint camera in the prior art;
[0035] Figure 4 This is an example diagram of pre-distortion variables in the prior art;
[0036] Figure 5 This is an example diagram of the AR-HUD optical path in the prior art;
[0037] Figure 6 This is a flow chart of the high-precision AR-HUD distortion calibration method of the present invention;
[0038] Figure 7 This is an example diagram of the display area of the present invention;
[0039] Figure 8 It is a schematic diagram of a code value matrix obtained by interpolation in a display area according to the present invention. DETAILED DESCRIPTION
[0040] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0041] The present invention provides a high-precision AR-HUD distortion calibration method, which requires that the positional relationship between each point in the projection dot matrix in the projection original image is determined, and then with the help of this relationship, the complicated encoding and decoding process can be eliminated to directly determine the order relationship between these points. In implementation, the projection dot matrix can be directly assigned with a certain rule to obtain its code value, and the reference dot matrix or sample dot matrix on the sample image, the virtual image in front of the vehicle, or other planes that have a certain relationship with the virtual image plane in front of the vehicle is assigned the same value or assigned according to other rules to obtain its code value; based on the positional relationship of each point in the sparse reference dot matrix or sample dot matrix, the code values at the coordinates in the area determined by these dot matrices are obtained by fitting or interpolation, and a code value matrix is obtained; finally, according to the coordinates of each point in the sample dot matrix or the reference dot matrix, their code values are found in the code value matrix, and according to the code value, their corresponding dot matrix on the projection original image is found. The details are as follows:
[0042] like Figure 6 As shown, the high-precision AR-HUD distortion calibration method in this embodiment includes the following steps:
[0043] The main process of the program is as follows Figure 6 As shown, it mainly includes:
[0044] 1) Loading information: Loading the information required for AR-HUD distortion calibration, mainly including: the original projection image, information such as feature type, projection dot matrix rows and columns and coordinates in the original projection image, and the calibration relationship between the viewpoint camera and the virtual image in front of the vehicle (or reference dot matrix, etc.).
[0045] 2) Obtain the dot matrix code value and assign a value to each point in the current projection dot matrix according to a certain rule. This value is the code value. The rule and the obtained code value need to help determine the correspondence between the sample dot matrix and the projection dot matrix on the projection original image.
[0046] 3) Projected image, i.e., the original projection image loaded in step 1 of the AR-HUD display.
[0047] 4) Capture an image, that is, the viewpoint camera captures the image displayed in step 3 to obtain a sample image.
[0048] 5) Detecting the image, that is, detecting the sample image according to the content information of the original projection image loaded in step 1 to obtain a sample dot matrix.
[0049] 6) Determine the target area, that is, determine the display area according to the sample dot matrix information detected in step 5. According to different requirements, the display area can be determined on the sample image, on the plane where the front virtual image of the vehicle (or the reference dot matrix) is located, etc. At this time, the sample dot matrix, the reference dot matrix and other dot matrices in the target area are only sparse and discrete coordinate points in the area, and their code values are also sparse and discrete, and cannot cover the entire display area.
[0050] 7) Fitting or interpolation: First, use certain rules to code the reference lattice or sample lattice (this operation can also be completed in step 2). Since the point-to-point correspondence between the projection lattice, sample lattice and reference lattice is certain, different coding rules can still ensure the correspondence between the code values of the projection lattice and the code values of the projection lattice. Then, according to the code values at the lattice, fit or interpolate the code values at each coordinate point in the entire target area to obtain a code value matrix with a clear correspondence with the target area.
[0051] 8) Find the corresponding point matrix, that is, obtain the corresponding code value according to the coordinates of each point in the sample point matrix or the reference point matrix on the target area, the clear correspondence between the target area and the code value matrix, and further find the corresponding point matrix on the projection original image based on these code values.
[0052] 9) Calculate the calibration parameters, that is, calculate the pre-distortion parameters of the AR-HUD based on the difference between the corresponding dot matrix and the projected dot matrix. First, calculate the pre-distortion variables at these dot matrices, and then fit or interpolate the pre-distortion variables at these points to obtain the pre-distortion variables of each point on the original projection image, and obtain the final pre-distortion matrix.
[0053] In addition, in the above steps 6, 7, and 8, different methods can be formed by different planes where the target area is located and different coding dot arrays. Assuming that the target area is established on the sample image, and the reference dot array and the sample dot array are also on the sample image, at least the following two methods a and b can be formed:
[0054] a. If the code values of the reference lattice are to be the same as those of the projection lattice, it is necessary to fit or interpolate the code values of each point in the target area with the help of the reference lattice and its code values (obtain a code value matrix), and then find the code values corresponding to the sample lattice with the help of the coordinates of the sample lattice and the clear correspondence between the target area and the code value matrix, and further find the corresponding lattice of the sample lattice on the projection original image based on the code values.
[0055] b. If the code values of the sample lattice are to be the same as the code values of the projection lattice, it is necessary to fit or interpolate the code values of each point in the target area with the help of the sample lattice and its code values (to obtain a code value matrix), and then use the coordinates of the reference lattice and the clear correspondence between the target area and the code value matrix to find the code values corresponding to the reference lattice, and further find the corresponding lattice of the reference lattice on the original projection image based on the code values.
[0056] There are many feasible embodiments that can be composed based on the differences in steps 6, 7, 8 and other steps, which cannot be listed one by one. Here, an embodiment based on calculating the target area on the virtual image in front of the vehicle will be described in detail. The specific steps of this embodiment are as follows: Figure 6 The main process is the same as:
[0057] 1) Load information, including the original projection image, feature type, projection dot matrix, and the conversion relationship between the viewpoint camera and the virtual image in front of the vehicle. The original projection image is generated by Figure 2 The circular spot image shown in the figure has the characteristic type of circular spot, the projection dot matrix is the coordinate of the center of the circular spot in the figure, and the dot matrix rows and columns are 9 rows and 18 columns. The transformation relationship between the viewpoint camera and the front virtual image of the vehicle is the rotation between the two coordinate systems ( )Translate( ) relationship, in the calibration workshop, the relative position between the viewpoint camera and the virtual image is determined, and these parameters can be easily obtained based on the simulation data.
[0058] 2) Get the dot matrix code value, and directly use the coordinates of each point in the projection dot matrix as the code value of each point. Assume that the coordinates of a point in the projection dot matrix are Then the code value of this point is Since the projection dot matrix is the real coordinate point on the projection original image, the above rules can be used to directly determine the coordinate value of any coordinate on the projection original image and its plane, which is its code value.
[0059] 3) Projected image, i.e., the circular spot image loaded in step 1 is displayed on the AR-HUD.
[0060] 4) Capture an image, that is, the viewpoint camera captures the image displayed in step 3 to obtain a sample image.
[0061] 5) Detect the image, that is, based on the feature type, dot matrix row and column information loaded in step 1, use OpenCV's findCirclesGrid function to perform spot detection to obtain the sample dot matrix.
[0062] 6) Determine the target area, that is, obtain the target area on the virtual image in front of the vehicle according to the sample dot matrix. The specific steps are:
[0063] First, with the help of the relationship between the viewpoint camera and the virtual image coordinate system, the plane equation of the virtual image plane under the viewpoint camera is determined. Assume that in the virtual image coordinate system, the normal of the virtual image plane is , the point on the virtual image plane is , then its normal under the viewpoint camera is , point .
[0064] Then, the straight line determined by the origin of the viewpoint camera coordinate system and the points in the sample dot matrix and the intersection points of the straight line and the virtual image plane in the viewpoint camera coordinate system are calculated in sequence to obtain the intersection dot matrix. The intersection dot matrix is the correspondence of the projection dot matrix on the virtual image plane in front of the vehicle, and is also the correspondence of the sample dot matrix on the virtual image in front of the vehicle (that is, the sample dot matrix on the virtual image in front of the vehicle).
[0065] Finally, the display area is determined according to the intersection matrix. In order to improve the speed or accuracy of the subsequent acquisition of the code value matrix, a proportional coefficient is set here. First, the intersection matrix is scaled according to the proportional coefficient, and then the effective area obtained based on the scaled intersection matrix will become the original area. Times. , then the code value matrix to be obtained becomes smaller, which helps to improve the speed of this step; if , the code value matrix to be obtained becomes larger, and increasing the resolution of the matrix in disguised form helps to improve the accuracy. In addition, in order to facilitate the subsequent acquisition of the code value of the reference dot matrix, the area determined by the intersection dot matrix is extended to a certain extent. Assume that there are N points in total, their coordinates are two-dimensional, and the offset is and . Then the maximum and minimum values of the x-coordinate and y-coordinate are:
[0066] ;
[0067] ;
[0068] ;
[0069] ;
[0070] The determined area is the upper left point , bottom right The rectangular area.
[0071] Since the intersection point matrix is the correspondence of the sample matrix on the virtual image in front of the vehicle, we can Figure 3 The sample lattice in is regarded as an example of the intersection lattice. , , When , the final display area obtained according to the intersection matrix is as follows Figure 7 As shown in the dashed box.
[0072] 7) Fitting or interpolation, first assign the code values of the projection dot matrix to the intersection dot matrix (the sample dot matrix on the virtual image plane in front of the vehicle) accordingly, and then interpolate their code values using the cubic spline interpolation method according to the coordinate relationship between the points of the intersection dot matrix obtained in step 6, obtain the code values at each position in the display area determined in step 6, and obtain a code value matrix of the same size as the area. Since the code value corresponds to the two-dimensional coordinates on the original projection image, it is necessary to interpolate these two dimensions separately, and the following are obtained respectively. Figure 8 The X-dimensional code value matrix shown on the left and Figure 8 The Y-dimensional code value matrix is shown on the right.
[0073] 8) Find the corresponding point matrix, that is, find the corresponding point matrix of the reference point matrix on the original projection image. The specific steps are:
[0074] First, find the largest inscribed rectangle within the envelope defined by the outermost points of the intersection lattice.
[0075] Then, according to the row and column information of the dot matrix, obtain the reference dot matrix in the maximum inscribed rectangle. Assuming that the width of the maximum inscribed rectangle is W and the height is H, and the dot matrix has M rows and N columns, the dot spacing in the dot matrix is W / (N-1) in width and H / (M-1) in height. The distribution of the obtained reference dot matrix is as follows: Figure 7 As shown by the solid circle in .
[0076] Finally, the corresponding code value is found on the code value matrix according to the coordinates of the reference point matrix. The specific operation is to find the x component of the code value at the reference point matrix of the X-dimensional code value matrix, and find the y component on the Y-dimensional code value matrix. The two are combined to form the final code value, which is the coordinate of the midpoint of the corresponding point matrix of the reference point matrix on the original projection image.
[0077] 9) Calculate the calibration parameters, that is, calculate the pre-distortion variables at these points based on the difference between the projected dot matrix and the corresponding dot matrix, and further use the bilinear interpolation method to obtain the pre-distortion variables of each point in the original projection image, that is, obtain the pre-distortion matrix. This step is included in the existing AR-HUD distortion calibration scheme, so it is an existing technology and will not be repeated here.
[0078] In summary, the present invention has the same advantages as the scheme based on structured light: the calculation of reprojection parameters and homography parameters can be skipped. Compared with the scheme based on structured light, this scheme does not require the use of complex encoding and decoding based on grayscale distribution differences, and has an accuracy advantage over spatial structured light and a speed advantage over temporal structured light.
[0079] The above-mentioned names are explained below with reference to the accompanying drawings:
[0080] AR-HUD (AR-Head-up-Display): The fusion of AR technology and HUD display reflects the image through the windshield, and the image is integrated with the environment outside the car, allowing the driver to see the picture that is a combination of the displayed content and the actual environment.
[0081] Viewpoint camera: Figure 1 As shown, a camera is placed at the location of the driver's eyes (viewpoint) to obtain the image seen by the human eye.
[0082] Original projection image: When dedistorting, the AR-HUD projects a checkerboard, circular dot array, and other dot matrix images. Figure 2 This is an example image where the original projection image is a circular speckle array.
[0083] Projection dot matrix: The dot matrix in the original projection image. If the original projection image is a chessboard, the projection dot matrix is composed of the corner point coordinates. If it is a circular dot matrix, the projection dot matrix is composed of the circle center coordinates.
[0084] Sample image: The image of the virtual image in front of the windshield captured by the viewpoint camera. The so-called virtual image is similar to Figure 1 The AR-HUD in the image is imaged in front of the windshield. After passing through the AR-HUD's complex optical path, the virtual image in the sample image is as follows Figure 3 As shown by the hollow circle in , there is a certain deformation (i.e. distortion) in the spatial arrangement.
[0085] Sample dot matrix: The dot matrix in the extracted sample image. Figure 3 The dot matrix formed by the centers of the hollow circles in the figure is an example of a sample dot matrix.
[0086] Reference dot matrix: A dot matrix generated based on parameters such as the arrangement, number of dots, and location. Its function is to provide a calibration benchmark for the sample dot matrix. Figure 3 The point matrix formed by the centers of the solid circles in the figure is an example of the reference point matrix.
[0087] Pre-distortion: According to the distortion characteristics of AR-HUD, the original projected image is reversely distorted in advance to offset the distortion of AR-HUD itself so that the human eye sees a normal picture.
[0088] Pre-distortion variables: Figure 4 As shown in the figure, assuming that the image width direction corresponds to the X direction and the height direction corresponds to the Y direction, the black solid circle is a point in the projection lattice, and the black hollow circle is a point in the sample lattice corresponding to the black solid circle and converted to the projection original image. If the point in the sample lattice corresponding to the black hollow circle is to be coincident with the corresponding point in the viewpoint reference lattice (i.e., dedistortion), the black hollow circle needs to be moved to the black solid circle on the projection original image, and the movement amount is the pre-distortion amount.
[0089] Pre-distortion matrix: Based on the relationship between the dot matrix, the pre-distortion variables at these dot matrix can be determined, and further interpolation can be performed to obtain the pre-distortion variables of each pixel on the original projection image. Its size is the same as that of the original projection image, and if it is represented by a matrix, it can be called a pre-distortion matrix.
[0090] Homography transformation: Mapping a point on a projective plane to another projective plane. After mapping, the straight line is still a straight line and has the line-preserving property. This transformation can be done with a The non-singular matrix representation of is a linear transformation relationship from one plane to another.
[0091] PGU: Image generation unit. Currently, PGU technologies mainly include TFT-LCD projection (Thin Film Transistor Liquid Crystal Display), DLP projection (Digital Light Processing), LBS laser scanning projection (Laser Beam Scanning), LCOS projection (Liquid Crystal on Silicon), etc.
[0092] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A high-precision AR-HUD distortion calibration method, characterized in that: include: Loading information, including the original projection image and its feature type, the rows, columns and coordinate information of the projection dot matrix, and the calibration relationship between the viewpoint camera and the virtual image in front of the vehicle or the reference dot matrix; Give each point in the projection lattice a code value of a specific rule to form a code value mapping of the projection lattice; The code value assignment methods of specific rules include: directly mapping the position of the point in the dot matrix to a predefined code value sequence; Projecting an image, displaying the original projection image; Shooting an image, collecting a projection image through a viewpoint camera, and obtaining a sample image including a sample dot matrix; Detect the image and identify the sample dot matrix in the sample image; use the findCirclesGrid function in the OpenCV library to perform spot detection in the detection image to identify the sample dot matrix; Determine the target area, and determine the display area on the virtual image in front of the vehicle or the sample image based on the detected sample dot matrix; determining the target area also includes calculating the intersection of the line connecting the origin of the viewpoint camera coordinate system and each point in the sample dot matrix with the virtual image plane to determine the corresponding position of the sample dot matrix on the virtual image in front of the vehicle; Fitting or interpolating the dot matrix in the target area, and generating a code value matrix for the entire target area based on the known dot matrix code values; Find the corresponding lattice, use the code value matrix in the target area to match the coordinates of the sample lattice or the reference lattice, and determine the corresponding lattice of the sample lattice or the reference lattice on the original projection image; find the largest inscribed rectangle in the envelope of the intersection lattice among the corresponding lattices, and determine the reference lattice accordingly; The calibration parameters are calculated, and based on the position difference between the corresponding dot matrix and the projected dot matrix, the pre-distortion parameters of the AR-HUD are calculated, and finally the pre-distortion matrix is obtained.
2. A high-precision AR-HUD distortion calibration method according to claim 1, characterized in that: Also includes: Other dot matrices outside the projected dot matrix are coded according to a rule corresponding to the projected dot matrix or a different rule to maintain the corresponding relationship of the code values.
3. The high-precision AR-HUD distortion calibration method according to claim 1, characterized in that: The sample image is obtained by photographing the image displayed by the AR-HUD through a viewpoint camera.
4. The high-precision AR-HUD distortion calibration method according to claim 1, characterized in that: The fitting or interpolation uses a cubic spline interpolation method to perform two-dimensional space interpolation on the dot code values in the target area.
5. The high-precision AR-HUD distortion calibration method according to claim 1, characterized in that: In the calculation of the calibration parameters, a bilinear interpolation method is used to interpolate the pre-distortion parameters to cover every point of the entire projection original image.
