Multi-camera calibration method based on tracer particles

Abstract

The invention provides a multi-camera calibration method based on tracer particles, and the method comprises the steps: continuously adding sparse tracer particles in an overlapped view field region, illuminating the tracer particles based on a high-brightness light source, carrying out the synchronous shooting of multiple cameras, and obtaining a smear-free tracer particle image at different moments, carrying out interpolation calculation on sub-pixel-level pixel coordinates of tracer particle projection points in the image; estimating the projection depth according to the obtained pixel coordinates and constructing a measurement matrix; performing singular value decomposition on the measurement matrix to obtain projective motion and a projective structure; a camera internal reference hypothesis is introduced, a homography matrix is solved, and projective reconstruction is converted into measurement reconstruction; and adding three-dimensional coordinate points defined in a world coordinate system in the overlapped view field area, and aligning the measurement reconstruction with the world system of the calibration scene. The multi-camera calibration method is not limited by the requirement of the calibration plate for multi-camera arrangement, the problem of error accumulation caused by the step of correlating adjacent cameras in a traditional multi-camera calibration method is solved, and the precision of the multi-camera calibration method is improved.

Description

Technical Field

[0001] The invention relates to the technical field of flow field optical measurement, and in particular to a multi-camera calibration method based on tracer particles. Background Art

[0002] With the popularization of cameras and the gradual maturity of stereo vision, multi-camera systems play an important role in optical 3D measurement of flow fields, such as: Particle Image Velocimetry (PIV) uses multiple cameras to capture images of the same flow field from different perspectives, and the obtained particle motion data in the fluid is used to calculate the velocity field and 3D flow structure of the fluid; Particle Tracking Velocimetry (PTV) uses multiple cameras to track the motion trajectory of particles in 3D space, thereby calculating the velocity field and acceleration field in the fluid; Tomographic Background Oriented Schlieren (TBOS) uses multiple cameras and background plates to simultaneously capture the changes in the flow field from different angles, and finally uses a 3D reconstruction algorithm to obtain 3D data of the flow field. The key prerequisite for 3D reconstruction is to calibrate all cameras. The purpose of multi-camera calibration is mainly to restore the 3D structure of the scene from a 2D image, and the accuracy of multi-camera calibration directly affects the accuracy of 3D reconstruction.

[0003] The multi-camera calibration method often uses a calibration plate for calibration. Due to the large angle between the cameras, it is usually difficult to ensure that all cameras can capture the calibration plate at the same time, which makes it impossible to obtain the external parameters of all cameras through one calibration. The multi-camera calibration method needs to be decomposed into first performing binocular calibration on adjacent cameras separately, and then correlating each group of cameras. This method not only requires multiple calibrations of local camera pairs, but also the errors will accumulate after multiple conversions, directly affecting the calibration accuracy.

[0004] At present, with the increasing requirements for the accuracy of three-dimensional reconstructed flow fields, the number of cameras used is gradually increasing to more than 8. If the traditional multi-camera calibration method is used, the operation is cumbersome. Therefore, it is very important to develop a new method that can quickly calibrate multi-camera calibration parameters and improve the accuracy of calibration results. Summary of the invention

[0005] The purpose of the present invention is to provide a multi-camera calibration method based on tracer particles, which is not limited by the requirements of the calibration plate for the arrangement of multiple cameras, avoids the error accumulation problem caused by the interrelated steps of adjacent cameras in the traditional multi-camera calibration method, is simpler and more flexible, and has a wider range of calibration scenarios; and at the same time, the constraint relationship between multiple cameras is used for calibration, which improves the robustness and accuracy of the multi-camera calibration method.

[0006] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0007] A multi-camera calibration method based on tracer particles, the multi-camera calibration method comprising the following steps:

[0008] Step A, building a scene of a multi-camera system to be calibrated, the scene including N cameras to be calibrated, N ≥ 3;

[0009] Step B, continuously adding sparse tracer particles in the overlapping field of view, illuminating the tracer particles based on a high-brightness light source, synchronously shooting with multiple cameras to obtain images of tracer particles without smear at different times, and interpolating and calculating the sub-pixel pixel coordinates of the projection points of the tracer particles in the image;

[0010] Step C, estimating the projection depth according to the obtained sub-pixel pixel coordinates and constructing a measurement matrix, performing singular value decomposition on the measurement matrix to obtain the projective motion P and the projective structure X;

[0011] Step D: introduce the camera intrinsic parameter hypothesis, use the geometric constraints and the camera intrinsic parameter hypothesis to calculate the homography matrix, and transform the projective reconstruction into the metric reconstruction; perform RQ decomposition on the metric motion to obtain the camera intrinsic parameter matrix K i And the rotation matrix R in metric space i ;

[0012] Step E, introduces the actual three-dimensional coordinate points in the world coordinate system defined in the overlapping field of view flow field area, aligns the metric reconstruction with the world system of the scene to be calibrated, and maps the metric reconstruction to the world coordinate system.

[0013] Step A further comprises:

[0014] Arrange a high brightness light source at a designated position so that it can effectively illuminate the tracer particles in the overlapping field of view;

[0015] During calibration, adjust the aperture of all camera lenses to a smaller value, determine the exposure time during calibration according to the moving speed of the tracer particles, and gradually reduce the exposure time until there is no smear of the tracer particles on the image.

[0016] The tracer particles are reflective or fluorescent; the tracer particles include but are not limited to helium bubbles and glass beads.

[0017] The high brightness light source adopts an incoherent or coherent light source to illuminate the imaging effect of the tracer particles in space, including but not limited to LED lights or lasers.

[0018] Step B further comprises:

[0019] Multiple cameras are used to synchronously shoot to obtain the images of the tracer particles without smear at different times, and the images shot by all cameras at the same time are recorded as the same set of images;

[0020] A threshold is set according to the brightness of the projection point, and each image is converted into a binary image using the threshold. Pixels in the image that are greater than the preset threshold are set to 1, and the remaining pixels are set to 0. The generated binary image is then marked to identify all connected areas, and the centroid coordinates of each connected area are calculated, and the number of detected connected areas is recorded.

[0021] Compare the number of connected areas of each image in the same set of images and retain the same number of image sets; sort the centroid coordinates of the particles in each image in the same set of images by height, and the particles with the same sequence number correspond to the same tracer particle in space;

[0022] The range of the sub-image is determined according to the centroid coordinates of each connected region and the predefined region size; the sub-image is extracted from the image, the resolution of the sub-image is increased by an interpolation algorithm, and the sub-pixel pixel coordinates of the particle image are determined.

[0023] Step C further comprises:

[0024] The projection depth is estimated using the constraints of epipolar geometry. The projection depth is the depth from the camera center to the three-dimensional point in the direction of the principal ray, and then the projection mapping equation is constructed:

[0025]

[0026] in, is the projection depth of the jth projection point observed by the i-th camera, j=1, ..., n, i=1, ..., m; is the coordinate of the jth projection point observed by the i-th camera; P i is the projection matrix of the i-th camera; X j is the coordinate of the jth tracer particle;

[0027] Assuming there are m cameras and n tracer particles, the measurement matrix equation is:

[0028]

[0029] Among them, W s is the measurement matrix, and The coordinates of the projection point of the j-th 3D point projected to the i-th camera.

[0030] Step D further comprises:

[0031] The homography matrix H is calculated using geometric constraints and camera intrinsic parameter assumptions, and then the metric motion is calculated. and metric structure

[0032] Perform RQ decomposition on the metric motion matrix to obtain the camera intrinsic parameter K i And the rotation matrix R in metric space i , in, is the metric matrix of the i-th camera; K i , R i ,t i are the intrinsic parameters, rotation matrix and translation vector of the i-th camera respectively.

[0033] Step E further comprises:

[0034] Add the three-dimensional coordinates defined in the world coordinate system in the overlapping field of view area, and project them onto each camera to obtain the projection point coordinates;

[0035] The three-dimensional coordinates of each set of projected points in the metric space are calculated by triangulation according to the metric motion, and their real positions in the world coordinate system are matched to calculate the rigid transformation including rotation, translation and scaling.

[0036] All cameras undergo a rigid transformation and the resulting metric reconstruction is aligned to the world system of the scene to be calibrated.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] First, the multi-camera calibration method based on tomographic background schlieren of the present invention based on tracer particles does not require individual calibration between two cameras. In contrast, the calibration of two cameras may require a lot of manual operations to complete the calibration of each pair of cameras. The present invention can calibrate a large number of cameras at the same time as long as tracer particles are applied in the overlapping field of view, simplifying the calibration process.

[0039] Second, in the multi-camera calibration method of the present invention based on tracer particles, the overall calibration result may be affected by the failure of calibration of some camera pairs in the pairwise camera calibration. The method of the present invention improves the overall robustness by using the common data of all cameras. Even if the calibration results of some cameras are poor, they can still be corrected using the information of other camera pairs. At the same time, the cumulative error caused by the correlation between cameras is avoided, and the calibration accuracy is improved.

[0040] Third, the tomographic background schlieren multi-camera calibration method based on tracer particles of the present invention does not need to place calibration objects multiple times in the calibration scene, so it can be applied in a wider range of scenes and is not limited by the size of the calibration plate and the arrangement of multiple cameras. This makes the present invention more flexible in various application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the multi-camera calibration method of tomographic background schlieren based on tracer particles of the present invention;

[0042] Figure 2 It is a schematic diagram of the structure of multi-camera calibration based on tracer particles of the present invention. DETAILED DESCRIPTION

[0043] The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0044] See also Figure 1 The embodiment of the present invention discloses a multi-camera calibration method based on a tomographic background schlieren method of tracer particles, and the multi-camera calibration method based on a tomographic background schlieren method of tracer particles comprises the following steps:

[0045] Step A, building a scene of a multi-camera system to be calibrated, where the scene includes N cameras to be calibrated, where N ≥ 3;

[0046] Step B, continuously adding sparse tracer particles in the overlapping field of view, illuminating the tracer particles based on a high-brightness light source, synchronously shooting with multiple cameras to obtain images of tracer particles without smear at different times, and interpolating and calculating the sub-pixel pixel coordinates of the projection points of the tracer particles in the image;

[0047] Step C, estimating the projection depth according to the obtained sub-pixel pixel coordinates and constructing a measurement matrix, performing singular value decomposition on the measurement matrix to obtain the projective motion P and the projective structure X;

[0048] Step D: introduce the camera intrinsic parameter hypothesis, use the geometric constraints and the camera intrinsic parameter hypothesis to calculate the homography matrix, and transform the projective reconstruction into the metric reconstruction; perform RQ decomposition on the metric motion to obtain the camera intrinsic parameter matrix K i And the rotation matrix R in metric space i ;

[0049] Step E: adding the actual three-dimensional coordinate points defined in the world coordinate system in the overlapping field of view area, aligning the metric reconstruction with the world system of the scene to be calibrated, and mapping the metric reconstruction to the world coordinate system.

[0050] See also Figure 2 The multi-camera system to be calibrated includes: the camera and lens to be calibrated, tracer particles, calibration plate, transmission line, and image acquisition system; the multi-camera system scene is built according to the experimental requirements; in order to increase the distinction between tracer particles and background and increase the depth of field range of the camera, the aperture number of all camera lenses is adjusted to a smaller value during calibration; the exposure time selected during calibration is determined by the moving speed of the tracer particles, and the exposure time can be gradually reduced until there is no smear of the tracer particles on the image;

[0051] A person skilled in the art can select suitable tracer particles according to the scene to be calibrated, etc. The tracer particles may be reflective or fluorescent, with a moderate particle size so that they can form clear projection points in the camera image and have good dispersibility so that they are evenly distributed in the medium. If a tracer particle generator device is used, the bubble outlet device is placed vertically upward below the overlapping field of view area and does not appear within the camera field of view. The tracer particles include but are not limited to helium bubbles, glass beads, etc.

[0052] The steps to calculate sub-pixel pixel coordinates are as follows:

[0053] (1) Multiple cameras are used to synchronously capture images of particle tracers without smear at different times. The images captured by all cameras at the same time are recorded as the same set of images.

[0054] (2) Use a preset threshold to convert each image into a binary image, set the pixels in the image that are greater than the preset threshold to 1, and set the remaining pixels to 0; then mark the generated binary image, identify all connected areas, calculate the centroid coordinates of each connected area, and record the number of detected connected areas;

[0055] (3) Compare the number of connected areas of each image in the same set of images and retain the same number of image sets; sort the centroid coordinates of each image in the same set of images by height, and the particles with the same sequence number correspond to the same tracer particle in space.

[0056] (4) Determine the range of the sub-image according to the coordinates of the centroid of each connected region and the predefined region size; extract the sub-image from the image, increase the resolution of the sub-image through an interpolation algorithm, and determine the sub-pixel pixel coordinates of the particle image through a specific algorithm.

[0057] The high brightness light source is arranged at an appropriate position to ensure that it can effectively illuminate the tracer particles in the overlapping field of view. The image acquisition system controls the camera to synchronously shoot and record the image. The above light source can be an incoherent or coherent light source, including but not limited to LED lights or lasers.

[0058] The preset threshold is set according to the brightness of the projection point. Each image is converted into a binary image using the preset threshold. The pixels in the image that are greater than the preset threshold are set to 1, and the remaining pixels are set to 0. Then the generated binary image is marked, all connected areas are identified, and the centroid coordinates of each connected area are calculated, and the number of detected connected areas is recorded.

[0059] Compare the number of connected areas of each image in the same set of images and retain the same number of image sets; sort the centroid coordinates of the particles in each image of the same set of images by height, and the particles with the same sequence number correspond to the same tracer particle in space.

[0060] The predefined area can be set according to the scale of the connected area. The centroid coordinates of each connected area and the predefined area size are used to extract this area from the image to obtain a sub-image. The resolution of the sub-image is increased through an interpolation algorithm. Bilinear interpolation, bicubic interpolation and other methods can be selected according to needs.

[0061] A local area is divided with the centroid coordinates as the center, and the sub-pixel coordinates are calculated using the quadratic interpolation method. The specific process of the quadratic interpolation method is as follows: Define the quadratic interpolation model: I(x, y) = ax 2 +by 2 +cxy+dx+ey+f, where l(x, y) is the intensity value of the pixel, (x, y) is the coordinate of the pixel in the local area, and a, b, c, d, e, f are the coefficients to be determined. By solving the quadratic polynomial, the partial derivatives of the quadratic polynomial are set to zero to solve the sub-pixel coordinates. Right now:

[0062]

[0063] in The projection point coordinates of the j-th 3D point projected onto the i-th camera (j=1, ...., n, i=1, ...., m), which are used for subsequent camera calibration.

[0064] According to the needs, Gaussian fitting method, phase correlation method, optical flow method or sub-pixel edge detection method can be selected to calculate the sub-pixel pixel coordinates of the projection point.

[0065] Using the constraints of epipolar geometry to estimate the projection depth, let the coordinates of all projection points of camera c be The projection depth is For camera i≠c, if cameras i and c have enough corresponding points to calculate a unique fundamental matrix F ic , then calculate their basic matrix, and then we can have

[0066]

[0067] The above projection depth λ is the depth from the camera center to the three-dimensional point in the direction of the main ray, and then constructs the projection mapping equation:

[0068]

[0069] in, is the projection depth of the jth projection point observed by the i-th camera; is the coordinate of the jth projection point observed by the ,th camera; P i is the projection matrix of the th camera; X j is the coordinate of the jth tracer particle.

[0070] Assuming there are m cameras and n tracer particles, the following measurement matrix equation W is written by combining the projection point coordinates and the corresponding projection point depth: s :

[0071]

[0072] Perform singular value decomposition on the above measurement matrix to obtain the projective motion and projective structure. Furthermore, W s =UDV T , due to W s The matrix rank is 4, and the first 4 columns of U are selected to form U 3m×4 , select the first 4 singular values ​​of D to form D 4×4 , select V T The first 4 lines consist of Projective Movement and projective structures

[0073] To convert the projective reconstruction into a metric reconstruction, we need to solve the homography matrix H, which is defined as H 4×4 =[A 4×3 b 4×1 ],and Combined, we get: μ i K i [R i t i ]=P i [A 4×3 b 4×1 ],in

[0074] First, solve b and define the origin of the world coordinate system as the center of mass of all tracer particles. Then, we can obtain the equation about b. 4×1 Solve 2m equations.

[0075] By μ i K i R i =P i A Launch Define a 4×4 symmetric matrix Q=AA T , the camera intrinsic parameter assumption is imposed: the camera pixels are square, the principal point coincides with the origin of the image coordinate system, and the intrinsic parameters of all cameras are the same, that is, Then we can get:

[0076]

[0077] Then we can obtain 4m linear equations to solve the matrix Q of 10 unknowns, and finally factorize Q to obtain A, which can be combined into the homography matrix H, and then obtain the metric motion and metric structure

[0078] Perform RQ decomposition on the metric motion matrix to obtain the camera intrinsic parameter K i And the rotation matrix R in metric space i , where t i is the translation vector in the metric space.

[0079] in, is the metric matrix of the i-th camera; K i , R i ,t i are the intrinsic parameters, rotation matrix and translation vector of the i-th camera respectively.

[0080] Add the 3D coordinates defined in the world coordinate system in the overlapping field of view area and project them onto each camera to obtain the coordinates of the projected points. The 3D coordinates of each set of projected points in the metric space are calculated by triangulation according to the metric motion, and then their real positions in the world coordinate system are matched to calculate the rigid transformation (including rotation, translation and scaling); all cameras undergo the same rigid transformation, and then align the obtained metric reconstruction with the world system of the scene to be calibrated.

[0081] The multi-camera calibration method proposed in the present invention is applied in combination with specific examples as follows.

[0082] In order to compare the calibration results of the present invention with the actual camera intrinsic parameters and extrinsic parameters, this embodiment adopts a numerical simulation method. Figure 2 The multi-camera calibration scene in the embodiment is shown. The scene contains 7 cameras, and 3D points are randomly generated from the inside of a sphere. These points are transformed by a preset camera matrix to obtain their projected point coordinates in each camera. Subsequently, Gaussian noise is added to each projected point coordinate to simulate the measurement error in actual applications.

[0083] (1) The seven cameras designed have the same internal parameters, with a pixel size of 3.45 μm × 3.45 μm, a resolution of 2448 × 2048, and a lens focal length of 50 mm;

[0084] Table 1 shows the actual internal parameter matrix of the multi-camera calibration internal parameters in this embodiment.

[0085] Table 1. Multi-camera calibration internal parameter results and true internal parameter matrix

[0086]

[0087] A chessboard is placed vertically in the overlapping field of view area, and the center of the chessboard is defined as the origin of the world coordinate system. The z-axis is vertically upward, and the x-axis is perpendicular to the chessboard plane. The y-axis is determined according to the right-hand rule, and rotated around the z-axis to ensure that any two adjacent cameras can observe the chessboard at the same time. The two cameras each take a picture of the chessboard, and then the three-dimensional coordinates of the corner points of the chessboard in the world coordinates and the coordinates of the projection points on the two cameras are obtained; the three-dimensional coordinates of each group of projection points in the metric space are calculated by the triangulation method according to the metric motion, and then their real positions in the world coordinate system are matched, and the rigid transformation (including rotation, translation and scaling) is calculated. All cameras undergo the same rigid transformation, and then the obtained metric reconstruction is converted to the world coordinate system.

[0088] Table 2 gives the camera center coordinates and true values ​​of the multi-camera calibration in this embodiment.

[0089] Table 2 Multi-camera calibration camera center coordinates and real center coordinates (unit: mm)

[0090] Camera No. True value Calculated Value Camera 1 <![CDATA[[1000.00 0.00 0.00] T ]]> <![CDATA[[999.52 0.00 0.04] T ]]> Camera 2 <![CDATA[[642.79 766.04 0.00] T ]]> <![CDATA[[642.45 765.55 0.05] T ]]> Camera 3 <![CDATA[[-220.45 954.88 0.00] T ]]> <![CDATA[[-220.20 953.54 0.00] T <!-- 7 -->]]> Camera 4 <![CDATA[[-898.79 438.37 20.00] T ]]> <![CDATA[[-898.21 438.02 20.00] T ]]> Camera 5 <![CDATA[[-880.82 -429.60 20.00] T ]]> <![CDATA[[-881.45 -429.86 19.99] T ]]> Camera 6 <![CDATA[[-224.95 -974.37 20.00] T ]]> <![CDATA[[-225.25 -975.64 20.01] T ]]> Camera 7 <![CDATA[[611.00 -766.21 20.00] T ]]> <![CDATA[[611.36 -766.53 20.01] T ]]>

[0091] The experimental data in Table 3 are the average reprojection errors of the accuracy of the internal and external parameter calibration results of this embodiment. It can be seen from Table 3 that the average reprojection error of the calibration results of the 7 cameras in this embodiment is less than 0.2 pixels, indicating that the calibration results of this embodiment are accurate.

[0092] Table 3 Average reprojection error of multi-camera calibration (unit: pixel)

[0093] Camera No. Average reprojection error Standard deviation of the reprojection error Camera 1 0.11 0.06 Camera 2 0.11 0.06 Camera 3 0.11 0.06 Camera 4 0.11 0.06 Camera 5 0.11 0.06 Camera 6 0.11 0.06 Camera 7 0.11 0.06 Camera 8 0.11 0.06

[0094] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0095] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A multi-camera calibration method based on tracer particles, characterized in that: The multi-camera calibration method based on tracer particles comprises the following steps: Step A, building a scene of a multi-camera system to be calibrated, where the scene includes N cameras to be calibrated, where N ≥ 3; Step B, continuously adding sparse tracer particles in the overlapping field of view, illuminating the tracer particles based on a high-brightness light source, synchronously shooting with multiple cameras to obtain images of tracer particles without smear at different times, and interpolating and calculating the sub-pixel pixel coordinates of the projection points of the tracer particles in the image; Step C, estimating the projection depth according to the obtained sub-pixel pixel coordinates and constructing a measurement matrix, performing singular value decomposition on the measurement matrix to obtain the projective motion P and the projective structure X; Step D: introduce the camera intrinsic parameter hypothesis, use the geometric constraints and the camera intrinsic parameter hypothesis to calculate the homography matrix, and transform the projective reconstruction into the metric reconstruction; perform RQ decomposition on the metric motion to obtain the camera intrinsic parameter matrix K i And the rotation matrix R in metric space i ; Step E: Add the actual three-dimensional coordinate points defined in the world coordinate system in the overlapping field of view, align the metric reconstruction with the world system of the scene to be calibrated, and map the metric reconstruction to the world coordinate system.

2. The multi-camera calibration method based on tracer particles according to claim 1, characterized in that: Step A further comprises: Arrange a high brightness light source at a designated position so that it can effectively illuminate the tracer particles in the overlapping field of view; During calibration, adjust the aperture of all camera lenses to a smaller value, determine the exposure time during calibration according to the moving speed of the tracer particles, and gradually reduce the exposure time until there is no smear of the tracer particles on the image.

3. The multi-camera calibration method based on tracer particles according to claim 1, characterized in that: The tracer particles are reflective or fluorescent; the tracer particles include but are not limited to helium bubbles and glass beads.

4. The multi-camera calibration method based on tracer particles according to claim 1, characterized in that: The high brightness light source adopts an incoherent or coherent light source to illuminate the imaging effect of the tracer particles in space, including but not limited to LED lights or lasers.

5. The multi-camera calibration method based on tracer particles according to claim 1, characterized in that: Step B further comprises: Multiple cameras are used to synchronously shoot to obtain the images of the tracer particles without smear at different times, and the images shot by all cameras at the same time are recorded as the same set of images; A threshold is set according to the brightness of the projection point, and each image is converted into a binary image using the threshold. Pixels in the image that are greater than the preset threshold are set to 1, and the remaining pixels are set to 0. The generated binary image is then marked to identify all connected areas, and the centroid coordinates of each connected area are calculated, and the number of detected connected areas is recorded. Compare the number of connected areas of each image in the same set of images and retain the same number of image sets; sort the centroid coordinates of the particles in each image in the same set of images by height, and the particles with the same sequence number correspond to the same tracer particle in space; The range of the sub-image is determined according to the centroid coordinates of each connected region and the predefined region size; the sub-image is extracted from the image, the resolution of the sub-image is increased by an interpolation algorithm, and the sub-pixel pixel coordinates of the particle image are determined.

6. The multi-camera calibration method based on tracer particles according to claim 1, characterized in that: Step C further comprises: The projection depth is estimated using the constraints of epipolar geometry. The projection depth is the depth from the camera center to the three-dimensional point in the direction of the principal ray, and then the projection mapping equation is constructed: in, is the projection depth of the jth projection point observed by the i-th camera, j=1, ..., n, i=1, ..., m; is the coordinate of the jth projection point observed by the i-th camera; P i is the projection matrix of the i-th camera; X j is the coordinate of the jth tracer particle; Assuming there are m cameras and n tracer particles, the measurement matrix equation is: Among them, W s is the measurement matrix, and The coordinates of the projection point of the j-th 3D point projected to the i-th camera.

7. The multi-camera calibration method based on tracer particles according to claim 1, characterized in that: Step D further comprises: The homography matrix H is calculated using geometric constraints and camera intrinsic parameter assumptions, and then the metric motion is calculated. and metric structure Perform RQ decomposition on the metric motion matrix to obtain the camera intrinsic parameter K i And the rotation matrix in metric space in, is the metric matrix of the i-th camera; K i , R i ,t i are the intrinsic parameters, rotation matrix and translation vector of the i-th camera respectively.

8. The multi-camera calibration method based on tracer particles according to claim 1, characterized in that: Step E further comprises: Add the three-dimensional coordinates defined in the world coordinate system in the overlapping field of view area, and project them onto each camera to obtain the projection point coordinates; The three-dimensional coordinates of each set of projected points in the metric space are calculated by triangulation according to the metric motion, and their real positions in the world coordinate system are matched to calculate the rigid transformation including rotation, translation and scaling. All cameras undergo a rigid transformation and the resulting metric reconstruction is aligned to the world system of the scene to be calibrated.

Images