A dynamic calibration method for multi-view images

Abstract

A dynamic calibration method for multi-view images is used to solve the inconvenience of known static calibration methods. It includes: each image capture unit captures images of multiple moving positions at each viewing angle, and generates corresponding image feature points corresponding to reference feature points in the actual scene; converts the image feature points into projection feature points; and calibrates the x, y and z axial inclination angles of each image capture unit according to the difference between a current distribution relationship of the projection feature points based on the current initial external parameters and an ideal distribution relationship based on the calibrated external parameters; and calibrates the x, y and z axial positions in the initial external parameters of each image capture unit, so that the projection feature points corresponding to the same reference feature point in each viewing angle overlap. The present invention can achieve the effect of conveniently and accurately calibrating the external parameters of the image capture units at each viewing angle.

Description

Technical Field

[0001] The present invention relates to an image processing technology, and in particular to a dynamic calibration method for multi-view images. Background Art

[0002] With the development of assisted driving or autonomous driving technology, the application of automotive images is becoming more and more widespread and popular; in particular, in order to capture images with accurate orientation, it is necessary to specially calibrate the position and angle of the image capture unit (such as a camera) installed on the driving vehicle, and then obtain the current external parameters of each image capture unit through the corresponding static calibration method. In detail, the steps of the static calibration method include: first installing the image capture unit to the target driving vehicle; then, moving the driving vehicle to a scene specifically used for static calibration; using a calibration plate with a predetermined pattern, and taking multiple images containing the calibration plate, especially images with different viewing angles and distances; finally, calculating the external parameters through multiple feature points in the captured images.

[0003] However, when a vehicle is statically calibrated, it must be done at a specific site, and the pattern of the calibration plate is easily affected by factors such as light and shadow, placement and angle accuracy in the site, and the image of the calibration plate must be captured at a specific position and distance. Therefore, the static calibration method is too dependent on the use of auxiliary equipment and the adjustment of the surrounding environment. It is not simple and convenient in the calibration process, and is constrained by equipment and environment.

[0004] In view of this, the known calibration methods still need to be improved. Summary of the invention

[0005] To solve the above problems, an object of the present invention is to provide a dynamic calibration method for multi-view images, which is applicable to various vehicles and can conveniently and accurately calibrate image capture units at each viewpoint.

[0006] The directions or their approximate terms described throughout the present invention, such as "front", "rear", "left", "right", "up (top)", "down (bottom)", "inside", "outside", "side", etc., are mainly with reference to the directions of the attached drawings. Each direction or its approximate terms are only used to assist in the explanation and understanding of the various embodiments of the present invention, and are not used to limit the present invention.

[0007] The quantifiers “a” or “an” used in the elements and components described throughout the present invention are only for convenience of use and to provide a general meaning of the scope of the present invention; in the present invention, they should be interpreted as including one or at least one, and a single concept also includes plural cases, unless it is obvious that it means otherwise.

[0008] The term "coupling" as used throughout the present invention includes direct or indirect electrical and / or signal connections, which can be selected by those skilled in the art according to usage requirements.

[0009] The similar terms such as "combination", "assembly", "assembly" and "setting" mentioned throughout the present invention mainly include the connection that can be separated without destroying the components, or the connection that makes the components inseparable, and can be selected by those with ordinary knowledge in the field according to the materials of the components to be connected or the assembly requirements.

[0010] The term "Extrinsic Parameters" used in the present invention refers to the position and angle of the image capture unit in the 3D scene, and in particular refers to the three-dimensional coordinates and angle values ​​of the image capture unit installed on the vehicle; in addition, the term "Internal Parameters" refers to the information of the optical center and focal length of the image capture unit. Through the external parameters, a reference feature point in the real world can be converted to an image feature point in the captured image; through the internal parameters, the image feature point can be mapped to the top view plane / projection image.

[0011] The present invention provides a dynamic calibration method for multi-view images, based on installing multiple image capture units of multiple viewpoints on a vehicle, and executing the following steps, including: moving the vehicle along a predetermined path to different motion positions, each of the image capture units capturing images corresponding to the multiple motion positions at each of the viewpoints; generating four corresponding image feature points corresponding to four reference feature points in an actual scene in the captured images of each of the motion positions at each of the viewpoints; and converting the corresponding image feature points into the image feature points according to the internal parameters of each of the image capture units at each of the viewpoints. The image feature points are converted into projection feature points in a top-down view; and the initial external parameters of each image capture unit are calibrated: the angle external parameters are calibrated: according to the difference between a current distribution relationship of the projection feature points based on the current initial external parameters and an ideal distribution relationship of the projection feature points based on the calibrated external parameters, the x-axis inclination angle, y-axis inclination angle and z-axis inclination angle of each image capture unit are calibrated; and then, the position external parameters are calibrated: the x-axis position, y-axis position and z-axis position in the initial external parameters of the image capture unit of each viewing angle are calibrated so that the projection feature points corresponding to the same reference feature point in each viewing angle overlap.

[0012] Accordingly, a dynamic calibration method for multi-view images of the present invention uses image feature points in images captured at multiple different motion positions, and through the relationship between image feature points, reference feature points, and projection feature points, to first calibrate angle external parameters and then calibrate position external parameters, thereby achieving the effect of accurately and conveniently calibrating the external parameters of each view image capture unit.

[0013] When calibrating the position external parameters, the x-axis position, y-axis position and z-axis position of the image capture unit of one viewing angle are used as references to calibrate the x-axis position, y-axis position and z-axis position of the image capture unit of other viewing angles, so that the projection feature points corresponding to the same reference feature point in each viewing angle overlap. In this way, the effect of quickly and accurately calibrating the position external parameters of the image capture unit of each viewing angle can be achieved.

[0014] When calibrating the angle external parameter, the x-axis tilt angle is calibrated first. In this way, by calibrating the x-axis tilt angle first, the calibration of the y-axis and z-axis tilt angles can be facilitated, which has the effect of quickly calibrating the angle external parameter.

[0015] When calibrating the x-axis tilt angle, two projection feature points with a spacing in the x-axis direction are selected from each projection feature point of each viewing angle, and a first x-axis spacing of the two selected projection feature points in the x-axis direction when they are in the first motion position is calculated, and a second x-axis spacing of the two selected projection feature points in the x-axis direction when they are in the second motion position is calculated, and the corresponding x-axis tilt angle is adjusted so that the first x-axis spacing is equal to the second x-axis spacing. In this way, the effect of quickly and accurately calibrating the x-axis tilt angle can be achieved through feature points at different motion positions (dynamic).

[0016] Wherein, when calibrating the y-axis tilt angle, the projection feature points converted from the captured images at each of the viewing angles each have a corresponding reference feature point; there is a first reference distance between two points in the reference feature points, and there is a second reference distance between the other two points in the reference feature points, the first reference distance is equal to the second reference distance, and there is a specific proportional relationship between the first reference distance and the second reference distance; a first projection distance is defined between the two projection feature points corresponding to the first reference distance, and a second projection distance is defined between the two projection feature points corresponding to the second reference distance, and the corresponding y-axis tilt angle is adjusted so that the first projection distance and the second distance meet the specific proportional relationship. In this way, through the above method of calibrating the y-axis tilt angle, the effect of quickly and accurately calibrating the y-axis tilt angle can be achieved.

[0017] Among them, the reference feature points include a first reference feature point, a second reference feature point, a third reference feature point, a fourth reference feature point, a fifth reference feature point, a sixth reference feature point, a seventh reference feature point and an eighth reference feature point; the multiple image capture units include a first image capture unit, a second image capture unit, a third image capture unit and a fourth image capture unit; the four projection feature points obtained by the first image capture unit at each of the movement positions correspond to the first reference feature point, the second reference feature point, the third reference feature point and the fourth reference feature point respectively; the four projection feature points obtained by the second image capture unit at each of the movement positions correspond to the first reference feature point, the third reference feature point, the fifth reference feature point and the The seventh reference feature point; the four projection feature points acquired by the third image capture unit at each of the movement positions correspond to the second reference feature point, the fourth reference feature point, the sixth reference feature point and the eighth reference feature point respectively; the four projection feature points acquired by the fourth image capture unit at each of the movement positions correspond to the fifth reference feature point, the sixth reference feature point, the seventh reference feature point and the eighth reference feature point respectively; the distance between the first reference feature point and the third reference feature point is equal to the distance between the second reference feature point and the fourth reference feature point; the distance between the fifth reference feature point and the seventh reference feature point is equal to the distance between the sixth reference feature point and the eighth reference feature point. In this way, through the effect of equal distances between the above-mentioned partial reference feature points, the effect of quickly and accurately calibrating the y-axis inclination angle can be achieved.

[0018] When calibrating the z-axis tilt angle, the projected feature points converted from the captured images at each of the viewing angles each have a corresponding reference feature point; the reference feature points have two preset reference feature points, and the x-coordinates and one of the y-coordinates of the two preset reference feature points are the same; the corresponding z-axis tilt angle is adjusted so that the x-coordinates or y-coordinates of the two projected feature points corresponding to the two preset reference feature points are the same. In this way, through the above method of calibrating the z-axis tilt angle, the effect of quickly and accurately calibrating the z-axis tilt angle can be achieved.

[0019] Wherein, the reference feature points include a first reference feature point, a second reference feature point, a third reference feature point, a fourth reference feature point, a fifth reference feature point, a sixth reference feature point, a seventh reference feature point and an eighth reference feature point; the plurality of image capture units include a first image capture unit, a second image capture unit, a third image capture unit and a fourth image capture unit; the four projection feature points acquired by the first image capture unit at each of the movement positions correspond to the first reference feature point, the second reference feature point, the third reference feature point and the fourth reference feature point respectively; the four projection feature points acquired by the second image capture unit at each of the movement positions correspond to the first reference feature point, the third reference feature point, the fifth reference feature point and the seventh reference feature point respectively; the four projection feature points acquired by the third image capture unit at each of the movement positions correspond to the second reference feature point respectively. , the fourth reference feature point, the sixth reference feature point and the eighth reference feature point; the four projection feature points acquired by the fourth image capture unit at each of the movement positions correspond to the fifth reference feature point, the sixth reference feature point, the seventh reference feature point and the eighth reference feature point respectively; the first reference feature point has the same y coordinate as the second reference feature point, the third reference feature point has the same y coordinate as the fourth reference feature point, the fifth reference feature point has the same y coordinate as the sixth reference feature point, and the seventh reference feature point has the same y coordinate as the eighth reference feature point; the first reference feature point has the same x coordinate as the fifth reference feature point, the second reference feature point has the same x coordinate as the sixth reference feature point, the third reference feature point has the same x coordinate as the seventh reference feature point, and the fourth reference feature point has the same x coordinate as the eighth reference feature point. In this way, through the configuration of the same x coordinate or y coordinate between the above reference feature points, the effect of quickly and accurately calibrating the z-axis inclination angle can be achieved.

[0020] Figure 1 A schematic top view of a scene for executing a dynamic calibration method according to the present invention is shown.

[0021] Figure 2 A three-dimensional schematic diagram showing first and second viewing angle images captured by the first and second image capturing units.

[0022] Figure 3 A schematic diagram showing two first-perspective images captured at two different motion positions.

[0023] Figure 4 The corresponding image feature points are converted into overhead projection feature points.

[0024] Figure 5 A schematic diagram showing the calibration principle of the x-axis inclination angle.

[0025] Figure 6 A schematic diagram showing the calibration principle of the y-axis inclination angle.

[0026] Figure 7 A schematic diagram showing the calibration principle of the z-axis tilt angle.

[0027] Figure 8 A schematic diagram showing projection images generated at various viewing angles with calibrated angular extrinsic parameters.

[0028] Fig. 9 A schematic diagram showing the selection of a position extrinsic parameter of a single viewing angle as a basic extrinsic parameter.

[0029] Fig.10 A schematic diagram showing the calibration of position extrinsic parameters for other viewing angles.

[0030] Fig.11 A schematic flow chart showing the dynamic calibration method for multi-view images of the present invention is shown.

[0031] Wherein: 1: vehicle, 2: image capture unit, 2F: first image capture unit, 2L: second image capture unit, 2R: third image capture unit, 2B: fourth image capture unit, WP: reference feature point, WP1: first reference feature point, WP2: second reference feature point, WP3: third reference feature point, WP4: fourth reference feature point, WP5: fifth reference feature point, WP6: sixth reference feature point, WP7: seventh reference feature point, WP8: Eighth reference feature point, x, y, z: axial, CI: captured image, CIF: first-angle captured image, CIF#1: first-angle captured image of the first motion position, CIF#2: first-angle captured image of the second motion position, CIL: second-angle captured image, CP: image feature point, CP1-F: first image feature point, CP1-F#1: image feature point of the first motion position, CP1-F#2: image feature point of the second motion position, PI: projection image, PIF: first projection image, PIL: second projection image, PIR: third projection image, PIB: Fourth projection image, PP: projection feature point, PP1-F, PP1-L: first projection feature point, PP2-F, PP2-R: second projection feature point, PP3-F, PP3-L: third projection feature point, PP4-F, PP4-R: fourth projection feature point, PP5-L, PP5-B: fifth projection feature point, PP6-R, PP6-B: sixth projection feature point, PP7-L, PP7-B: seventh projection feature point, PP8-R, PP8-B: eighth projection feature point, PP1-F#1, PP1-L#1: first projection feature point of the first motion position , PP2-F#1, PP2-R#1: the second projection feature point of the first motion position, PP3-F#1, PP3-L#1: the third projection feature point of the first motion position, PP4-F#1, PP4-R#1: the fourth projection feature point of the first motion position, PP5-L#1, PP5-B#1: the fifth projection feature point of the first motion position, PP6-R#1, PP6-B#1: the sixth projection feature point of the first motion position, PP7-L#1, PP7-B#1: the seventh projection feature point of the first motion position, PP8-R#1, PP8-B#1: the first The eighth projection feature point of the motion position, P1-F#2, PP1-L#2: the first projection feature point of the second motion position, PP2-F#2, PP2-R#2: the second projection feature point of the second motion position, PP3-F#2, PP3-L#2: the third projection feature point of the second motion position, PP4-F#2, PP4-R#2: the fourth projection feature point of the second motion position, PP5-L#2, PP5-B#2: the fifth projection feature point of the second motion position, PP6-R#2, PP6-B#2: the sixth projection feature point of the second motion position, PP7-L#2,PP7-B#2: seventh projection feature point of the second motion position, PP8-R#2, PP8-B#2: eighth projection feature point of the second motion position, dx#1: first x-axis spacing, dx#2: second x-axis spacing, dβ1: first distance, dβ2: second distance, λ1: first value, λ2: second value, S1, S2, S3: steps. , DETAILED DESCRIPTION

[0032] In order to make the above contents and other purposes, features and advantages of the present invention more clearly understood, the preferred embodiments of the present invention are specifically cited below and described in detail with the accompanying drawings; in addition, the same symbols in different drawings are considered the same and their descriptions are omitted. In particular, Figures 1 to 10 are used to illustrate the mechanism and principle of the dynamic calibration method of multi-view images of the present invention, and Figure 11 is a flow chart of the dynamic calibration method of multi-view images of the present invention proposed based on Figures 1 to 10.

[0033] It should be noted that, in order to facilitate the explanation of the principle of the calibration method of the present invention, corresponding codes are given to specific technical features in the specification, and the codes are only partially shown in the drawings. The meaning and numbering rules of the codes are explained below: 1. The code "WP" stands for "reference feature point". The codes "WP1~WP8" are subordinate codes of the code "WP" and stand for "the first to the eighth reference feature points".

[0034] 2. The code "CI" means "captured image". The codes "CIF, CIL, CIR, CIB" are the first-order subordinate codes of the code "CI", which mean "first to fourth angles of view captured images". The codes "CIF#i, CIL#i, CIR#i, CIB#i" are the second-order subordinate codes of the code "CI", which mean "first to fourth angles of view captured images of the i-th motion position".

[0035] 3. The code “CP” represents “image feature point” and is generated from the corresponding reference feature point WP. The code “CP1~CP8” followed by “-F”, “-L”, “-R” and “-B” are the first-order subordinate codes of the code “CP”, indicating the “first to eighth image feature points” generated by the corresponding first, second, third and fourth image capture units corresponding to the first to eighth reference feature points WP1~WP8. The code “CP1~CP8” is first followed by “-F”, “-L”, “-R” and “-B” and then “#i”, which is the second-order subordinate code of the code “CP”, and the code “#i” is used to represent different motion positions.

[0036] 4. The code "PI" stands for "projection image", which is generated by converting the captured image CI through the internal reference K. The codes "PIF, PIL, PIR, PIB" are the first-order subordinate codes of the code "PI", which stand for "first to fourth projection images" and correspond to the first to fourth perspective captured images CIF, CIL, CIR, CIB. Code.

[0037] 5. The code “PP” represents “projected feature points”, which are generated by the image feature points CP through the conversion of the internal reference K. The code “PP1~PP8” is followed by “-F”, “-L”, “-R” and “-B”, which are the first-order subordinate codes of the code “PP”, indicating the “first to eighth projected feature points” generated by the corresponding first, second, third and fourth image capture units corresponding to the corresponding image feature points CP. The code “PP1~PP8” is first followed by “-F”, “-L”, “-R” and “-B”, and then “#i”, which is the second-order subordinate code of the code “PP”, and the code “#i” is used to indicate different motion positions.

[0038] Please refer to FIG. 1, which shows a schematic top view of a scene in which the dynamic calibration method is performed. In the scene, an image capture unit 2 corresponding to each viewing angle of a vehicle 1 selects at least four reference feature points WP; in each captured image CI of each viewing angle captured by the image capture unit 2, the reference feature point WP captured presents a corresponding image feature point CP. Optionally, the specific position of each reference feature point WP, in particular the coordinate information [X W , Y W ] is pre-configured and determined. Optionally, the specific position of each reference feature point WP can also be selected from two adjacent captured images CI of each viewing angle with the same feature by using a pre-trained image recognition model, and the corresponding coordinate information [X W , Y W ] can also be obtained through the pre-trained image recognition model; it should be noted that the pre-trained image recognition model can be implemented using various neural networks or other machine learning technologies, for example, it can be implemented using known object recognition models or semantic segmentation models and other technologies.

[0039] In detail, in terms of the image capture and conversion mechanism, each of the image capture units 2 has a corresponding initial external parameter R and an internal parameter K; the initial external parameter R and the coordinate information of the corresponding reference feature point WP [X W , Y W ] After the inner product, the corresponding image feature point CP and coordinate information [X C , Y C ]; the coordinate information of the internal parameter K and the corresponding image feature point CP [X C , YC ] After the inner product, the corresponding top-view projection feature point PP and coordinate information [X P , Y P The above conversion of different coordinate positions through the initial external parameter R and the internal parameter K is understandable to those with ordinary knowledge in the field of the present invention, so it will not be repeated here.

[0040] For example, the vehicle 1 has four image capture units 2 to construct the corresponding surround image. The image capture units 2 can be further divided into a first image capture unit 2F, a second image capture unit 2L, a third image capture unit 2R and a fourth image capture unit 2B. The first image capture unit 2F is used to capture a first perspective capture image CIF in a first perspective direction (e.g., a front perspective); the second image capture unit 2L is used to capture a second perspective capture image CIL in a second perspective direction (e.g., a left perspective); the third image capture unit 2R is used to capture a third perspective capture image CIR in a third perspective direction (e.g., a right perspective); and the fourth image capture unit 2B is used to capture a fourth perspective capture image CIB in a fourth perspective direction (e.g., a rear perspective). Corresponding to the first to fourth perspective directions, the scene has at least eight reference feature points WP in total, which can be defined as the first to eighth reference feature points WP1 to WP8, respectively.

[0041] In the first-perspective captured image CIF, the first reference feature point WP1, the second reference feature point WP2, the third reference feature point WP3 and the fourth reference feature point WP4 in the world coordinates are photographed by the first image capture unit 2F and converted by the initial external parameter R of the first image capture unit 2F to generate corresponding first image feature points CP1-F, second image feature points CP2-F, third image feature points CP3-F and fourth image feature points CP4-F respectively.

[0042] In the second-perspective captured image CIL, the first reference feature point WP1, the third reference feature point WP3, the fifth reference feature point WP5 and the seventh reference feature point WP7 in the world coordinates are captured by the second image capture unit 2L and converted by the initial external parameter R of the second image capture unit 2L to generate the corresponding first image feature point CP1-L, third image feature point CP3-L, fifth image feature point CP5-L and seventh image feature point CP7-L respectively.

[0043] In the third-perspective captured image CIR, the second reference feature point WP2, the fourth reference feature point WP4, the sixth reference feature point WP6 and the eighth reference feature point WP8 in the world coordinates are photographed by the third image capture unit 2R and converted by the initial external parameter R of the third image capture unit 2R to generate corresponding second image feature points CP2-R, fourth image feature points CP4-R, sixth image feature points CP6-R and eighth image feature points CP8-R respectively.

[0044] In the fourth-perspective captured image CIB, the fifth reference feature point WP5, the sixth reference feature point WP6, the seventh reference feature point WP7 and the eighth reference feature point WP8 in the world coordinates are photographed by the fourth image capture unit 2B and converted by the initial external parameter R of the fourth image capture unit 2B to generate the corresponding fifth image feature point CP5-B, sixth image feature point CP6-B, seventh image feature point CP7-B and eighth image feature point CP8-B respectively.

[0045] It should be noted that the description of the above technical features uses "digital" numbers and image feature points "CP" to present CP1~CP8, so as to clearly correspond to the reference feature points WP1~WP8 at the corresponding positions in the scene; in addition, the numbers "-F", "-L", "-R" or "-B" are used to clearly show which image capture unit 2F, 2L, 2R or 2B is used to obtain each of the image feature points CP.

[0046] Please refer to FIG. 2, which shows the first and second perspective captured images CIF and CIL captured by the first and second image capture units 2F and 2L of the vehicle 1 in the corresponding scene. In particular, the first reference feature point WP1 will generate corresponding image feature points CP1-F and CP1-L in the first and second perspective captured images CIF and CIL, respectively. It should be noted that, in order to facilitate the explanation that any of the reference feature points WP will generate a corresponding image feature point CP in the captured image CI of the adjacent perspective, although FIG. 2 only shows the situation where the first and second perspective captured images CIF and CIL have image feature points CP1-F and CP1-L respectively corresponding to the first reference feature point WP1, the combination of the image feature points CP generated by each reference feature point WP in the captured image CI of the adjacent perspective can be presented as shown in Table 1 below.

[0047] Table 1: Combinations of image feature points CP corresponding to each reference feature point WP in adjacent view images.

[0048] Reference feature points in the scene Captured images from adjacent viewing angles Corresponding image feature points The first reference feature point WP1 First-person image CIF CP1-F Second Angle Image CIL CP1-L The second reference feature point WP2 First-person image CIF CP2-F (not shown) Third-person image CIR (not shown) CP2-R (not shown) The third reference feature point WP3 First-person image CIF CP3-F (not shown) Second Angle Image CIL CP3-L (not shown) Fourth reference feature point WP4 First-person image CIF CP4-F (not shown) Third-person image CIR (not shown) CP4-R (not shown) Fifth reference feature point WP5 Second Angle Image CIL CP5-L (not shown) Fourth angle image CIB (not shown) CP5-B (not shown) Sixth reference feature point WP6 Third-person image CIR (not shown) CP6-R (not shown) Fourth angle image CIB (not shown) CP6-B (not shown) Seventh reference feature point WP7 Second Angle Image CIL CP7-L (not shown) Fourth angle image CIB (not shown) CP7-B (not shown) Eighth reference feature point WP8 Third-person image CIR (not shown) CP8-R (not shown) Fourth angle image CIB (not shown) CP8-B (not shown) Table 1 Please refer to FIG. 3 , in order to implement the dynamic calibration method of the present invention, the vehicle 1 is moved in a specific manner to capture the first perspective captured image CIF, the second perspective captured image CIL, the third perspective captured image CIR and the fourth perspective captured image CIB corresponding to the movement position at at least two different movement positions. In detail, the specific manner means that the distance and direction of the vehicle movement can be calculated to determine the difference in distance and angle between different movement positions. In particular, to facilitate explanation of the dynamic calibration method of the present invention, FIG. 4 only uses the example of capturing two corresponding first-perspective captured images CIF of the vehicle 1 at two specific different motion positions; the two first-perspective captured images CIF can be further represented as a first-perspective captured image CIF#1 of the first motion position and a first-perspective captured image CIF#2 of the second motion position; correspondingly, the first reference feature point WP1 generates a corresponding image feature point CP1-F#1 of the first motion position in the first-perspective captured image CIF#1 of the first motion position; the first reference feature point WP1 generates a corresponding image feature point CP1-F#2 of the second motion position in the first-perspective captured image CIF#2 of the second motion position.

[0049] It should be noted that the description of the above technical features uses “#” plus “number” numbers to express different information obtained when the vehicle 1 moves to different movement positions, so that the technical features of the present invention can be more clearly explained and presented through the above numbers.

[0050] Based on the example shown in Figure 3, the vehicle 1 captures image information of N moving positions in total, and the corresponding image feature points generated by each reference feature point WP in the adjacent perspective images of different moving positions can be explained as follows (not specifically shown in the figure); wherein N is a positive integer not less than 2, and N is preferably 5; in addition, in the following description, the number "i" represents the relevant technical features obtained by the vehicle 1 at the corresponding i-th different moving position, and i is 1 to N.

[0051] For the first reference feature point WP1, the first-perspective captured image CIF#i of the i-th moving position is captured to generate the corresponding image feature point CP1-F#i of the i-th moving position, and the second-perspective captured image CIL#i of the i-th moving position is also captured to generate the corresponding image feature point CP1-L#i of the i-th moving position.

[0052] For the second reference feature point WP2, the first-angle captured image CIF#i of the i-th moving position is captured to generate the corresponding image feature point CP2-F#i of the i-th moving position, and the third-angle captured image CIR#i of the i-th moving position is also captured to generate the corresponding image feature point CP2-R#i of the i-th moving position.

[0053] For the third reference feature point WP3, the first-perspective captured image CIF#i of the i-th moving position is captured to generate the corresponding image feature point CP3-F#i of the i-th moving position, and the second-perspective captured image CIL#i of the i-th moving position is also captured to generate the corresponding image feature point CP3-L#i of the i-th moving position.

[0054] For the fourth reference feature point WP4, the first-perspective captured image CIF#i of the i-th moving position is captured to generate the corresponding image feature point CP4-F#i of the i-th moving position, and the third-perspective captured image CIR#i of the i-th moving position is also captured to generate the corresponding image feature point CP4-R#i of the i-th moving position.

[0055] For the fifth reference feature point WP5, the second-angle captured image CIL#i of the i-th moving position is captured to generate the corresponding image feature point CP5-L#i of the i-th moving position, and the fourth-angle captured image CIB#i of the i-th moving position is also captured to generate the corresponding image feature point CP5-B#i of the i-th moving position.

[0056] For the sixth reference feature point WP6, the third-perspective image CIR#i of the i-th moving position is captured to generate the corresponding image feature point CP6-R#i of the i-th moving position, and the fourth-perspective image CIB#i of the i-th moving position is captured to generate the corresponding image feature point CP6-B#i of the i-th moving position.

[0057] For the seventh reference feature point WP7, the second-angle captured image CIL#i of the i-th moving position is captured to generate the corresponding image feature point CP7-L#i of the i-th moving position, and the fourth-angle captured image CIB#i of the i-th moving position is also captured to generate the corresponding image feature point CP7-B#i of the i-th moving position.

[0058] For the eighth reference feature point WP8, the third-perspective image CIR#i of the i-th moving position is captured to generate the corresponding image feature point CP8-R#i of the i-th moving position, and the fourth-perspective image CIB#i of the i-th moving position is captured to generate the corresponding image feature point CP8-B#i of the i-th moving position.

[0059] Based on the above description, as shown in FIG. 4, when the vehicle 1 moves in two different motion positions in a specific manner during dynamic calibration (N=2, i=1-2), taking the image feature points CP generated by the corresponding reference feature points WP in the captured images CI of each perspective as an example, the inner product of the internal parameter K of the captured images CI of each perspective and the coordinate information of each image feature point CP can be performed to obtain the corresponding projection feature point PP. The projection feature point PP converted from each image feature point CP of the captured images CI of each perspective at each motion position is described in detail as follows.

[0060] By capturing image CIF#1 from the first perspective of the first motion position, the first, second, third, and fourth projection feature points PP1-F#1, PP2-F#1, PP3-F#1, and PP4-F#1 of the first motion position can be obtained. By capturing image CIF#2 from the first perspective of the second motion position, the first, second, third, and fourth projection feature points PP1-F#2, PP2-F#2, PP3-F#2, and PP4-F#2 of the second motion position can be obtained.

[0061] By capturing image CIL#1 from the second viewing angle of the first motion position, the first, third, fifth, and seventh projection feature points PP1-L#1, PP3-L#1, PP5-L#1, and PP7-L#1 of the first motion position can be obtained. By capturing image CIL#2 from the second viewing angle of the second motion position, the first, third, fifth, and seventh projection feature points PP1-L#2, PP3-L#2, PP5-L#2, and PP7-L#2 of the second motion position can be obtained.

[0062] By capturing image CIR#1 from the third perspective of the first motion position, the second, fourth, sixth, and eighth projection feature points PP2-R#1, PP4-R#1, PP6-R#1, and PP8-R#1 of the first motion position can be obtained. By capturing image CIR#2 from the third perspective of the second motion position, the second, fourth, sixth, and eighth projection feature points PP2-R#2, PP4-R#2, PP6-R#2, and PP8-R#2 of the second motion position can be obtained.

[0063] By capturing image CIB#1 from the fourth perspective of the first motion position, the fifth, sixth, seventh, and eighth projection feature points PP5-B#1, PP6-B#1, PP7-B#1, and PP8-B#1 of the first motion position can be obtained. By capturing image CIB#2 from the fourth perspective of the second motion position, the fifth, sixth, seventh, and eighth projection feature points PP5-B#2, PP6-B#2, PP7-B#2, and PP8-B#2 of the second motion position can be obtained.

[0064] Please refer to Figures 5 to 10 for illustrating the method of dynamic calibration of multi-view images of the present invention; wherein, Figures 5 to 7 illustrate the calibration / optimization of the angle extrinsic parameter in the initial extrinsic parameter R of the image capture unit 2 of a single view; and Figures 8 to 10 illustrate the calibration / optimization of the position extrinsic parameter in the initial extrinsic parameter R of the image capture unit 2 of each view, so as to complete the method of dynamic calibration of multi-view images of the present invention to generate corresponding synthetic images.

[0065] As shown in FIG. 5, it shows the method of calibrating the initial external parameter R in the x-axis tilt angle proposed by the present invention, and the calibration is performed using the projection feature points PP obtained at different movement positions of the vehicle 1; in particular, the x-axis tilt angle refers to the angle of rotation of the corresponding image capture unit 2 about the x-axis in the three-dimensional coordinate system, and can be understood as the pitch angle of the image capture unit 2. In detail, based on the projection feature points PP converted from each perspective image obtained in FIG. 4, two points with a spacing in the x-axis direction are selected from each of the projection feature points PP, and a first and second x-axis spacing dx#1 and dx#2 of the selected two projection feature points PP in the x-axis direction at the first and second movement positions are calculated; therefore, in the ideal case where the x-axis tilt angle has been calibrated, and in particular, when the y-axis position of the vehicle 1 at different movement positions changes, the first and second x-axis spacings dx#1 and dx#2 will still be the same.

[0066] For example, FIG. 5 uses the first image capture unit 2F in the first viewing direction as an example to illustrate that the first and second projection feature points PP1-F#1 and PP2-F#1 acquired at the first moving position of the carrier 1 are used to calculate the first x-axis spacing dx#1, and the first and second projection feature points PP1-F#2 and PP2-F#2 acquired at the second moving position are used to calculate the second x-axis spacing dx#2. When the first x-axis spacing dx#1 is not equal to the second x-axis spacing dx#2, it means that the x-axis tilt angle is not calibrated, and the x-axis tilt angle needs to be adjusted so that the first x-axis spacing dx#1 is equal to the second x-axis spacing dx#2.

[0067] As shown in FIG. 6 , the method for calibrating the inclination angle of the initial external parameter R in the y-axis direction proposed by the present invention can be calibrated using the projection feature points PP obtained by the vehicle 1 at the same moving position. In particular, the y-axis inclination angle refers to the angle of rotation of the corresponding image capture unit 2 about the y-axis in the three-dimensional coordinate system, and can be understood as the roll angle of the image capture unit 2. In detail, the projection feature points PP converted from the captured image CI at each viewing angle are at least four (corresponding to the at least four reference feature points WP mentioned above), two points of the at least four reference feature points WP have a first reference distance, and another two points of the at least four reference feature points WP have a second reference distance, and the first reference distance and the second reference distance have a specific proportional relationship, and the specific proportional relationship is preferably equal (i.e., 1:1); wherein, a first projection distance dβ1 is defined between the two projection feature points PP corresponding to the first reference distance, and a second projection distance dβ2 is defined between the two projection feature points PP corresponding to the second reference distance. In an ideal case where the y-axis inclination angle has been calibrated, the first projection distance dβ1 and the second projection distance dβ2 meet the specific proportional relationship. Based on the ideal case where the calibration has been completed, when the first distance dβ1 and the second distance dβ2 do not meet the specific proportional relationship, the y-axis inclination angle should be adjusted so that the first projection distance dβ1 and the second projection distance dβ2 meet the specific proportional relationship.

[0068] For example, FIG. 6 uses the first image capture unit 2F in the first viewing direction as an example to illustrate, and the specific ratio relationship is 1:1, and the projection feature points PP1-F, PP2-F, PP3-F, and PP4-F are obtained at any moving position of the vehicle 1. Under the configuration of the reference feature point WP of the present invention (as shown in FIG. 1), in one example, the first projection feature point PP1-F and the third projection feature point PP3-F are used to calculate the first projection distance dβ1, and the second projection feature point PP2-F and the fourth projection feature point PP4-F are used to calculate the second projection distance dβ2. In another example, the first projection distance dβ1 is calculated using the first projection feature point PP1-F and the fourth projection feature point PP4-F, and the second projection distance dβ2 is calculated using the second projection feature point PP2-F and the third projection feature point PP3-F.

[0069] Based on the same mechanism, in the second image capture unit 2L in the second viewing direction, the first projection distance dβ1 can be defined by the distance between the first and third projection feature points PP1-L, PP3-L (or the first and seventh projection feature points PP1-L, PP7-L), and the second projection distance dβ2 can be defined by the distance between the fifth and seventh projection feature points PP5-L, PP7-L (or the third and fifth projection feature points PP3-L, PP5-L).

[0070] Based on the same mechanism, in the third image capture unit 2R in the third viewing direction, the first projection distance dβ1 can be defined by the distance between the second and fourth projection feature points PP2-R, PP4-R (or the second and eighth projection feature points PP2-R, PP8-R), and the second projection distance dβ2 can be defined by the distance between the sixth and eighth projection feature points PP6-R, PP8-R (or the fourth and sixth projection feature points PP4-R, PP6-R).

[0071] Based on the same mechanism, in the fourth image capture unit 2B in the fourth viewing direction, the first projection distance dβ1 can be defined by the distance between the fifth and seventh projection feature points PP5-B and PP7-B (or the fifth and eighth projection feature points PP5-B and PP8-B), and the second projection distance dβ2 can be defined by the distance between the sixth and eighth projection feature points PP6-B and PP8-B (or the sixth and seventh projection feature points PP6-B and PP7-B).

[0072] As shown in FIG. 7 , it shows the method for calibrating the z-axis inclination angle of the initial external parameter R proposed by the present invention, which can be calibrated using the projection feature points PP obtained by the vehicle 1 at the same moving position. In particular, the z-axis inclination angle refers to the angle of rotation of the corresponding image capture unit 2 about the z-axis in the three-dimensional coordinate system, and can be understood as the heading angle of the image capture unit 2. In detail, the number of projection feature points PP converted from the captured image CI at each viewing angle is at least four (corresponding to the aforementioned at least four reference feature points WP), and it is preferred that the reference feature points WP have two preset reference feature points WP, and the x-coordinates and one of the y-coordinates of the two preset reference feature points WP are the same. Therefore, in the ideal case where the z-axis inclination angle has been calibrated, the corresponding x-coordinates and y-coordinates of the two projection feature points PP generated corresponding to the two preset reference feature points WP with the same x-coordinate or y-coordinate will also be the same. Based on the ideal situation that the z-axis inclination angle has been calibrated, and especially under the premise that the x-axis inclination angle and the y-axis inclination angle are calibrated, the x-coordinate or y-coordinate of the two projection feature points PP generated by the two reference feature points WP with the same corresponding x-coordinate or y-coordinate are taken out and defined as the first value λ1 and the second value λ2 respectively. When the x-coordinate or y-coordinate of the corresponding two projection feature points PP are different, the z-axis inclination angle should be adjusted so that the first value λ1 is equal to the second value λ2.

[0073] For example, FIG. 7 uses the first image capture unit 2F in the first viewing direction as an example to illustrate that the projection feature points PP1-F, PP2-F, PP3-F, and PP4-F acquired by the vehicle 1 at any moving position respectively take out the y coordinates of the two projection feature points PP generated by the two reference feature points WP with the same y coordinates, and define them as the first value λ1 and the second value λ2 respectively. Under the configuration of the reference feature point WP of the present invention (as shown in FIG. 1), in one example, the first value λ1 and the second value λ2 can be defined by the y coordinates of the first and second projection feature points PP1-F and PP2-F respectively; in another example, the first value λ1 and the second value λ2 can be defined by the y coordinates of the third and fourth projection feature points PP3-F and PP4-F respectively.

[0074] Based on the same mechanism, in the second image capture unit 2L in the second viewing direction, the first value λ1 and the second value λ2 can be defined by the x-coordinates of the first and fifth projection feature points PP1-L and PP5-L, or can be defined by the x-coordinates of the third and seventh projection feature points PP3-L and PP7-L.

[0075] Based on the same mechanism, in the third image capture unit 2R in the third viewing direction, the first value λ1 and the second value λ2 can be defined by the x-coordinates of the second and sixth projection feature points PP2-R and PP6-R, or can be defined by the coordinates of the fourth and eighth projection feature points PP4-R and PP8-R.

[0076] Based on the same mechanism, in the fourth image capture unit 2B in the fourth viewing direction, the first value λ1 and the second value λ2 can be defined by the y coordinates of the fifth and sixth projection feature points PP5-B and PP6-B, or can be defined by the y coordinates of the seventh and eighth projection feature points PP7-B and PP8-B.

[0077] After adjusting the above-mentioned x, y, and z axial inclination angles, the projection images PI corresponding to each viewing direction can be obtained as shown in FIG. 8, including the first projection image PIF corresponding to the first viewing angle captured image CIF, the second projection image PIL corresponding to the second viewing angle captured image CIL, the third projection image PIR corresponding to the third viewing angle captured image CIR, and the fourth projection image PIB corresponding to the fourth viewing angle captured image CIB. The image capture unit 2 corresponding to the captured image CI of each viewing direction has individual and uncalibrated / optimized external parameters at the x, y, and z positions.

[0078] It should be noted that according to the mechanism of calibrating the x, y, and z-axis inclination angles in the above Figures 5 to 7, the reference datum point WP pre-set and defined in the present invention can be configured according to actual needs, and the calibration of the above-mentioned x, y, and z-axis inclination angles can be achieved by recording the difference ratio between the two reference datum points in the x and y axes. Preferably, the configuration of each reference feature point WP can also be made as in Figure 1 of the present invention, so that the y coordinates of the first reference feature point WP1 and the second reference feature point WP2 are the same, the y coordinates of the third reference feature point WP3 and the fourth reference feature point WP4 are the same, the y coordinates of the fifth reference feature point WP5 and the sixth reference feature point WP6 are the same, and the y coordinates of the seventh reference feature point WP7 and the eighth reference feature point WP8 are the same; the x coordinates of the first reference feature point WP1 and the fifth reference feature point WP5 are the same, the x coordinates of the second reference feature point WP2 and the sixth reference feature point WP6 are the same, and the x coordinates of the third reference feature point WP3 and the seventh reference feature point WP8 are the same. The x-coordinate of the reference feature point WP7 is the same, so that the x-coordinates of the fourth reference feature point WP4 and the eighth reference feature point WP8 are the same; the distance between the first reference feature point WP1 and the third reference feature point WP3 is equal to the distance between the second reference feature point WP2 and the fourth reference feature point WP4; ​​the distance between the fifth reference feature point WP5 and the seventh reference feature point WP7 is equal to the distance between the sixth reference feature point WP6 and the eighth reference feature point WP8; preferably, the distance between the first reference feature point WP1 and the third reference feature point WP3 is also equal to the distance between the fifth reference feature point WP5 and the seventh reference feature point WP7.

[0079] Continuing with FIG. 8, as shown in FIG. 9, an external parameter corresponding to one of the projection images PI corresponding to each viewing direction is selected as a reference external parameter, and the position external parameter corresponding to the x, y, and z axial positions is optionally adjusted for the reference external parameter. For example, FIG. 9 uses the first projection image PIF as an example for explanation. Optionally, when the external parameter corresponding to the first projection image PIF or the fourth projection image PIB is selected as the reference external parameter, in addition to specifying the position external parameter corresponding to the y and z axial positions, the position external parameter corresponding to the x-axial position can be determined by making the x-axis average value of the projection feature point PP in the first or fourth projection image PIF, PIB equal to the center point position in the x direction of the first or fourth projection image PIF, PIB. Optionally, when the second projection image PIL or the third projection image PIR is selected as the reference external parameter, in addition to specifying the position external parameters of the corresponding x- and z-axial positions, the position external parameter of the corresponding y-axial position can be determined by making the y-axis average value of the projection feature point PP in the second or third projection image PIL, PIR equal to the center point position in the y direction of the second or third projection image PIL, PIR.

[0080] Next, as shown in FIG. 10, according to a projection image PI selected as a reference external parameter, the position external parameters of the x, y, and z axial positions in other projection images PI are adjusted accordingly to generate the superposition results of the projection images PI of each perspective in FIG. 10, so that the projection feature points PP corresponding to the same reference feature point WP can overlap, completing the dynamic calibration method of the multi-perspective image of the present invention. In particular, when the dynamic calibration method of the multi-perspective image of the present invention is completed, the captured images CI of the multi-perspective can be directly synthesized into a panoramic image.

[0081] According to the above description of the mechanism and principle of the dynamic calibration method of multi-view images of the present invention, FIG. 11 shows a flow chart of the dynamic calibration method of multi-view images of the present invention, including steps S1 to S3, and is described in detail as follows.

[0082] In step S1, captured images CI of multiple motion positions of each viewing angle are obtained. Specifically, based on multiple image capture units 2 of multiple viewing angles installed on the vehicle 1, captured images CI corresponding to multiple motion positions are captured in each viewing angle direction at different motion positions along a predetermined path; optionally, as shown in FIG. 1, the multiple image capture units 2 preferably include the first image capture unit 2F, the second image capture unit 2L, the third image capture unit 2R, and the fourth image capture unit 2B.

[0083] In detail, as shown in Figure 3, Table 1 and the aforementioned related contents, in the captured image CI of each motion position of each single perspective, each corresponds to four reference feature points WP in an actual scene, and corresponding four image feature points CP are generated; for example, in the first motion position and the second motion position of the first perspective, the first perspective captured image CIF#1 of the first motion position has the first to four image feature points CP1-F#1~CP4-F#1 of the first motion position corresponding to the first to four reference feature points WP1~WP4; the first perspective captured image CIF#2 of the second motion position has the first to four image feature points CP1-F#2~CP4-F#2 of the second motion position corresponding to the first to four reference feature points WP1~WP4.

[0084] In particular, in the same moving position, two of the four reference feature points WP in the actual scene generate corresponding two image feature points CP in a captured image CI of a perspective and another captured image CI of an adjacent perspective, respectively. For example, the first and third reference feature points WP1 and WP3 generate the first and third image feature points CP1-F and CP3-F in the first perspective captured image CIF, and the first and third reference feature points WP1 and WP3 generate the first and third image feature points CP1-L and CP3-L in the adjacent second perspective captured image CIL. In addition, the second and fourth reference feature points WP2 and WP4 generate the second and fourth image feature points CP2-F and CP4-F in the first perspective captured image CIF, and the second and fourth reference feature points WP2 and WP4 generate the second and fourth image feature points CP2-R and CP4-R in the adjacent third perspective captured image CIR.

[0085] In step S2 , according to the internal parameter K of each image capturing unit 2 at each viewing angle, the corresponding image feature point CP is converted into a top-view projection feature point PP (as shown in FIG. 4 ).

[0086] In step S3, the initial external parameters R of each image capture unit 2 are calibrated / optimized. Specifically, referring to FIGS. 5 to 7 and the aforementioned related contents, the angle external parameters (x-axis tilt angle, y-axis tilt angle and z-axis tilt angle) in the initial external parameters R of the image capture unit 2 of each viewing angle are first calibrated / optimized. In particular, the x-axis tilt angle, y-axis tilt angle and z-axis tilt angle of each image capture unit 2 are calibrated based on the difference between a current distribution relationship of the projection feature points PP based on the current initial external parameters and an ideal distribution relationship of the projection feature points PP based on the calibrated external parameters. Next, referring to FIGS. 8 to 10 and the aforementioned related contents, the position external parameters (x-axis position, y-axis position and z-axis position) in the initial external parameters R of the image capture unit 2 of each viewing angle are calibrated / optimized so that the projection feature points PP corresponding to the same reference feature point WP in each viewing angle can overlap, thereby completing the dynamic calibration method of multi-view images of the present invention. In particular, in the calibration / optimization of position extrinsic parameters, the position extrinsic parameters of the image capture unit 2 of one viewing angle are first selected and determined, and then the position extrinsic parameters of the image capture unit 2 of other viewing angles are calibrated accordingly to implement the dynamic calibration method of multi-view images of the present invention.

[0087] It should be noted that although the vehicle 1 shown in Figures 1 to 3 of the present invention is a vehicle, the dynamic calibration method of multi-view images of the present invention can be applied to various types of vehicles (such as airplanes or ships, etc.) and is not limited to vehicles.

[0088] In summary, the dynamic calibration method of multi-view images of the present invention obtains image feature points corresponding to each reference feature point through pre-configured reference feature points in conjunction with different positions of the vehicle on a predetermined path, and then determines whether the projection feature points conform to their ideal configuration relationship based on the principle of ideal external parameters to complete the calibration of the angle external parameters; then, the projection feature points pointing to the same reference feature point in each viewing direction are overlapped to complete the calibration of the position external parameters; in other words, through the calibration of the angle external parameters and the calibration of the position external parameters, the dynamic calibration method of multi-view images of the present invention can be implemented, and the images obtained from each viewing angle can be synthesized into a corresponding panoramic image.

[0089] Although the present invention has been disclosed using the above preferred embodiments, they are not intended to limit the present invention. Any person skilled in the art may make various changes and modifications to the above embodiments without departing from the spirit and scope of the present invention. The scope of protection of the present invention shall include all changes within the meaning and equivalent scope of the appended patent application.

Claims

1. A dynamic calibration method for multi-view images, based on installing multiple image capture units of multiple viewpoints on a vehicle, characterized in that: Perform the following steps, including: The vehicle is moved to different moving positions along a predetermined path, and each of the image capture units captures images corresponding to a plurality of moving positions at each of the viewing angles; in the captured images of each of the moving positions at each of the viewing angles, four corresponding image feature points are generated corresponding to four reference feature points in an actual scene; According to the internal parameters of the image capture unit at each viewing angle, the corresponding image feature points are converted into top-view projection feature points; and Calibrate the initial external parameters of each image acquisition unit: Calibrate angle extrinsic parameters: calibrate the x-axis inclination angle, y-axis inclination angle, and z-axis inclination angle of each image capture unit according to the difference between a current distribution relationship of the projection feature points based on the current initial extrinsic parameters and an ideal distribution relationship of the projection feature points based on the calibrated extrinsic parameters; and Next, the position external parameters are calibrated: the x-axis position, y-axis position and z-axis position in the initial external parameters of the image capture unit of each viewing angle are calibrated so that the projected feature points corresponding to the same reference feature point in each viewing angle overlap.

2. The method for dynamic calibration of multi-view images according to claim 1, characterized in that: When calibrating the position external parameters, the x-axial position, y-axial position and z-axial position of the image capture unit of one viewing angle are used as a reference to calibrate the x-axial position, y-axial position and z-axial position of the image capture unit of other viewing angles so that the projection feature points corresponding to the same reference feature point in each viewing angle overlap.

3. The method for dynamic calibration of multi-view images according to claim 1, characterized in that: When calibrating the angle external parameters, first calibrate the x-axis inclination angle.

4. The method for dynamic calibration of multi-view images according to claim 1, characterized in that: When calibrating the x-axis inclination angle, two projection feature points spaced in the x-axis direction are selected from the projection feature points of each of the viewing angles, a first x-axis distance between the two selected projection feature points in the x-axis direction at the first movement position is calculated, and a second x-axis distance between the two selected projection feature points in the x-axis direction at the second movement position is calculated, and the corresponding x-axis inclination angle is adjusted so that the first x-axis distance is equal to the second x-axis distance.

5. The method for dynamic calibration of multi-view images according to claim 3, characterized in that: When calibrating the x-axis inclination angle, two projection feature points spaced in the x-axis direction are selected from the projection feature points of each of the viewing angles, a first x-axis distance between the two selected projection feature points in the x-axis direction at the first movement position is calculated, and a second x-axis distance between the two selected projection feature points in the x-axis direction at the second movement position is calculated, and the corresponding x-axis inclination angle is adjusted so that the first x-axis distance is equal to the second x-axis distance.

6. A method for dynamic calibration of multi-view images according to any one of claims 1 to 5, characterized in that: When calibrating the y-axis tilt angle, the projected feature points converted from the captured images at each of the viewing angles each have a corresponding reference feature point; there is a first reference distance between two of the reference feature points, and there is a second reference distance between another two of the reference feature points, the first reference distance is equal to the second reference distance, and there is a specific proportional relationship between the first reference distance and the second reference distance; A first projection distance is defined between two projection feature points corresponding to the first reference distance, and a second projection distance is defined between two projection feature points corresponding to the second reference distance. The corresponding y-axis inclination angle is adjusted so that the first projection distance and the second distance conform to the specific proportional relationship.

7. The method for dynamic calibration of multi-view images according to claim 6, characterized in that: The reference feature points include a first reference feature point, a second reference feature point, a third reference feature point, a fourth reference feature point, a fifth reference feature point, a sixth reference feature point, a seventh reference feature point and an eighth reference feature point; the plurality of image capture units include a first image capture unit, a second image capture unit, a third image capture unit and a fourth image capture unit; the four projection feature points acquired by the first image capture unit at each of the movement positions correspond to the first reference feature point, the second reference feature point, the third reference feature point and the fourth reference feature point respectively; the four projection feature points acquired by the second image capture unit at each of the movement positions correspond to the first reference feature point, the third reference feature point, the fifth reference feature point and the seventh reference feature point respectively; The four projection feature points acquired by the third image capture unit at each of the movement positions correspond to the second reference feature point, the fourth reference feature point, the sixth reference feature point and the eighth reference feature point respectively; the four projection feature points acquired by the fourth image capture unit at each of the movement positions correspond to the fifth reference feature point, the sixth reference feature point, the seventh reference feature point and the eighth reference feature point respectively; The distance between the first reference feature point and the third reference feature point is equal to the distance between the second reference feature point and the fourth reference feature point; the distance between the fifth reference feature point and the seventh reference feature point is equal to the distance between the sixth reference feature point and the eighth reference feature point.

8. A method for dynamic calibration of multi-view images according to any one of claims 1 to 5, characterized in that: When calibrating the z-axis inclination angle, the projected feature points converted from the captured images at each of the viewing angles each have a corresponding reference feature point; the reference feature points have two preset reference feature points, and the x-coordinates and one of the y-coordinates of the two preset reference feature points are the same; the corresponding z-axis inclination angle is adjusted so that the x-coordinates or y-coordinates of the two projected feature points corresponding to the two preset reference feature points are the same.

9. The method for dynamic calibration of multi-view images according to claim 6, characterized in that: The reference feature points include a first reference feature point, a second reference feature point, a third reference feature point, a fourth reference feature point, a fifth reference feature point, a sixth reference feature point, a seventh reference feature point and an eighth reference feature point; the plurality of image capture units include a first image capture unit, a second image capture unit, a third image capture unit and a fourth image capture unit; the four projection feature points acquired by the first image capture unit at each of the movement positions correspond to the first reference feature point, the second reference feature point, the third reference feature point and the fourth reference feature point respectively; the four projection feature points acquired by the second image capture unit at each of the movement positions correspond to the first reference feature point, the third reference feature point, the fifth reference feature point and the seventh reference feature point respectively; The four projection feature points acquired by the third image capture unit at each of the movement positions correspond to the second reference feature point, the fourth reference feature point, the sixth reference feature point and the eighth reference feature point respectively; the four projection feature points acquired by the fourth image capture unit at each of the movement positions correspond to the fifth reference feature point, the sixth reference feature point, the seventh reference feature point and the eighth reference feature point respectively; The first reference feature point has the same y coordinate as the second reference feature point, the third reference feature point has the same y coordinate as the fourth reference feature point, the fifth reference feature point has the same y coordinate as the sixth reference feature point, and the seventh reference feature point has the same y coordinate as the eighth reference feature point; the first reference feature point has the same x coordinate as the fifth reference feature point, the second reference feature point has the same x coordinate as the sixth reference feature point, the third reference feature point has the same x coordinate as the seventh reference feature point, and the fourth reference feature point has the same x coordinate as the eighth reference feature point.

Images