Pose combination calibration method in 10kV hot-line work safety distance detection scene
The three-dimensional coordinates of the pixel point in the center of the light spot were extracted through binocular stereoscopic vision algorithm and Gaussian fitting method, combined with the single-point laser distance measurement value, and the position combination parameters of the binocular camera and the single-point laser distance measuring instrument were obtained using the LM optimization algorithm, which solved the problem of the solution. In the prior art, the problem of insufficient accuracy and parameter coupling has achieved high-precision distance measurement at the 10kV live operation site.
Patent Information
- Application Number
- CN202410876246.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-02
- Publication Date
- 2024-11-15
AI Technical Summary
The existing pose combination calibration method has insufficient measurement accuracy and applicability at the 10kV live operation site, especially the method based on binocular cameras and single-point laser rangefinders has problems of parameter coupling and high algorithm complexity.
The binocular stereoscopic vision algorithm is used to extract the three-dimensional coordinates of the pixel point in the center of the light spot through Gaussian fitting method, and the initial value of the pose combination calibration is established by combining the single-point laser ranging value, and the constraint function is optimized through the LM optimization algorithm to obtain independent combination parameters.
Within the measurement range of 15-20m, the relative error rate of ranging is 2.01%, which meets the ranging accuracy requirements of 10kV live operation site. The error fluctuation is significantly reduced, improving the applicability and accuracy of the system.
Smart Images

Figure CN118967823A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of electric power detection and discloses a posture combination calibration method in a 10kV live working safety distance detection scenario. Background Art
[0002] Live working is one of the important measures to ensure power supply reliability and improve people's satisfaction with electricity use. During the operation, it is necessary to monitor in real time whether the distance between the key parts of the personnel and the live body meets the requirements of safety regulations. However, the common manual supervision methods in actual production are highly subjective, so the automation technology for safe distance detection has become a research hotspot in academia, mainly including methods based on binocular vision, laser ranging, and the combination of the two. The distance detection accuracy based on binocular vision is restricted by variable environmental factors and cannot meet the medium and long distance distance measurement requirements of 10kV live working sites. The method based on laser ranging has problems such as high equipment cost, poor portability, and insufficient real-time performance. The method of combining the two uses the high measurement accuracy of the single-point laser rangefinder in the medium and long distance to correct the binocular camera ranging system, which is more suitable for the safe distance detection task in the live working scene. However, the key to accurate correction is to establish the spatial position relationship constraint between the two, so the combined calibration of their posture is extremely important.
[0003] The existing pose combination calibration methods mainly include four types of methods: target-based, targetless, hand-eye calibration, and deep learning. In the target-based method, Chen Yuan et al. first grouped the point clouds and improved the calibration accuracy by calculating the total average value of the point clouds; Han Zhengyong et al. [9]The external parameter calibration is converted into a problem of solving a three-dimensional space matrix; Wei Kequan et al. eliminated the influence of the position by changing the placement of the target. In the targetless method, Song Wensong et al. extracted the same-name feature points of the camera image and the point cloud grayscale image and then performed matching calculations, but the error increased when the number of data frames was small; Levinson et al. matched the discontinuous point cloud depth boundary and image edge and then searched. This method is only suitable for small-scale tasks. In the hand-eye calibration method, Yang Chao et al. established a nonlinear model between observation and prediction to optimize the solution of external parameters, but the accuracy was insufficient in unstructured environments; Taylor et al. calibrated the continuous frames of each sensor and then calibrated the external parameters between sensors, which also had the problem of insufficient accuracy. Most of the existing multi-sensor pose calibration methods based on deep learning are unsupervised or semi-supervised learning. This type of method only obtains approximate results, and its applicability and accuracy need to be improved. The measurement accuracy of the safety distance at the live working site is high and the image acquisition frame rate is low, so the pose combination calibration method based on the target is more suitable for this scenario. There have been relevant studies on the combined calibration method of single-point laser rangefinder and binocular camera based on target. Chu Zhaobi et al. first roughly solved the calibration parameters, and then established constraint equations for optimization. The calibration accuracy of this method depends on the number of points involved in iterative optimization and the scene adaptability is insufficient. Ma Haoran et al. obtained the combined parameters by solving the equation of the laser light in the camera coordinate system. This method obtained the direction vector of the light after obtaining the optical center coordinates. The mutual dependence between different parameters caused the calibration results to not converge. Tang Qingbo et al. used the coordinates of the intersection of the camera optical center-laser point and laser light as the spot coordinates and used the calibration algorithm to obtain the combined parameters. This method increased the complexity of the algorithm because it required multiple calculations of coordinate parameters. Summary of the invention
[0004] In order to solve the above technical problems, the present invention provides a method for calibrating the posture combination of a binocular camera and a single-point laser rangefinder applied to a 10kV live working scene. The three-dimensional coordinates of the center pixel of the light spot obtained by the Gaussian fitting method in the camera coordinate system are obtained by a binocular stereo vision algorithm, and the initial value of the posture combination calibration is obtained based on the distance value between the three-dimensional coordinates and the single-point laser. A constraint function of the spatial position parameters is established and the calibration plate is moved to obtain the three-dimensional coordinates of multiple groups of pixel points. The constraint function is optimized by the LM algorithm to obtain mutually independent posture combination calibration parameters. During the calibration process, the internal parameters of the binocular camera are obtained by Zhang Zhengyou's calibration method, and no further optimization is required.
[0005] Preferably, the three-dimensional coordinates of the light spot hit by the single-point laser on the calibration plate are obtained through the stereo vision algorithm of the binocular camera, the three-dimensional coordinates of another light spot are obtained by moving the calibration plate, the initial value of the combination parameter is calculated based on the distance between the two points measured by the single-point laser, the spatial position constraint relationship is established, and the final posture combination parameters are solved by optimizing the constraint relationship.
[0006] Preferably, the calibration process involves the conversion of multiple coordinate systems, and the following coordinate system is established:
[0007] 1) Camera coordinate system: camera optical center O c Establish the camera coordinate system O as the origin c X c Y c Z c , the X-axis is parallel to the mirror surface to the right, the Y-axis is along the camera mirror surface downward, and the Z-axis is perpendicular to the camera mirror surface outward; the coordinates of the light spot Q in this coordinate system are expressed as Q(X q , Y q , Z q );
[0008] 2) Pixel coordinate system: The image pixel coordinate system OUV is established with the upper left corner of the imaging plane as the origin, with the U axis parallel to the image plane and horizontal to the right, and the V axis perpendicular to the U axis and downward; the coordinates of the light spot Q in this coordinate system are expressed as q(u, v);
[0009] 3) Physical coordinate system: Establish the physical coordinate system O” with the intersection of the camera optical axis on the imaging plane as the origin c X” c Y" c Z” c , the X-axis is parallel to the U-axis of the pixel coordinate system, and the Y-axis is parallel to the V-axis of the pixel coordinate system;
[0010] The coordinate transformation relationship between the pixel coordinate system and the physical coordinate system satisfies equation (1);
[0011]
[0012] Where dx and dy are the physical sizes of a single pixel in the U and V axis directions. 0 , V 0 ) are the principal point coordinates of the image;
[0013] The coordinate transformation relationship between the physical coordinate system and the camera coordinate system satisfies equation (2);
[0014]
[0015] Where f is the focal length of the camera, (X q , Y q , Z q ) is the coordinate of point Q in the camera coordinate system, Z q is the depth value of point Q in the camera coordinate system;
[0016] Combining equation (1) and equation (2) gives equation (3), which is the transformation relationship that the binocular camera pixel coordinate system and the camera coordinate system must satisfy;
[0017]
[0018] Where s is the torsion factor.
[0019] Preferably, the three-dimensional position coordinates of the imaging point of the single-point laser rangefinder in the camera coordinate system satisfy equation (4);
[0020]
[0021] where θ x ,θ y ,θ z are the angles between the laser light and the positive direction of the camera coordinate system axis, L x , L y , L z are the offsets of the laser optical center relative to the origin of the camera coordinate system, and H is the distance measured by the single-point laser rangefinder;
[0022] According to the spatial position relationship, the angle θ between the light and the three coordinate axes is x ,θ y ,θ z Should satisfy formula (5);
[0023] cos 2 θ x +cos 2 θ y+ cos 2 θ z =1 (5)
[0024] Combining equation (4) and equation (5) we get equation (6);
[0025]
[0026] Among them, L x , L y , L z ,θ x ,θ y That is, the pose combination parameters need to be calibrated and are independent of each other.
[0027] Preferably, the Gaussian fitting method is used to extract the coordinates of the pixel point at the center of the light spot;
[0028] The intensity distribution of the laser beam is approximately Gaussian distribution, and its Gaussian function is expressed as formula (5);
[0029]
[0030] Where M is the peak parameter of Gaussian distribution, (σ x , σ y ) are the distribution parameters in the x and y directions respectively;
[0031] Taking the logarithm of both ends of equation (7) and expanding the square term, we can get equation (8);
[0032]
[0033] The least squares method is used to perform Gaussian fitting on the spot area in the image, and the fitting extreme point (x 0 ,y 0 ) as the center pixel of the light spot.
[0034] Preferably, the calibration of the posture combination parameters is achieved by the following steps:
[0035] The first step: fix the relative position of the camera and the single-point laser rangefinder to ensure that the imaging point can be seen on the calibration plate in the camera's field of view, use the Gaussian fitting method to extract the pixel coordinates of the center of the imaging spot, combine the distance length measured by the single-point laser, and calculate its three-dimensional coordinates in the camera coordinate system. Move the calibration plate to obtain the three-dimensional coordinate value of another point, and solve the initial value of the combined parameters together; the second step: obtain multiple sets of three-dimensional coordinate values by moving the calibration plate multiple times, establish a constraint function, and optimize through an iterative algorithm to obtain the final parameters.
[0036] Preferably, the Gaussian fitting method is used to extract Q 1 Center pixel coordinate q 1 (x q1 ,y q1 ), point Q is obtained through binocular stereo vision algorithm 1 The depth value Z q1 , Substitute the coordinate value and depth value into formula (3) to obtain point Q 1 Coordinate Q in the camera coordinate system 1 (X q1 , Y q1 , Z q1 ), using the laser rangefinder to obtain the distance value H 1 ; Keep the relative position of the sensor unchanged and move the calibration plate to S 2 Position, at this time the laser is at S 2 The imaging point of the plane is Q 2 , using Gaussian fitting method to extract Q 2 Center pixel coordinate q 2 (x q2 ,y q2 ), the binocular camera gets the point Q2 The depth value is Z q2 , and similarly obtain Q 2 The three-dimensional coordinate Q in the camera coordinate system 2 (X q2 , Y q2 , Z q2 ), and the distance value H 2 ;
[0037] Substitute the above values into equation (9), solve the simultaneous equations to obtain the initial values of the combined parameters, and the solution formula is shown in equation (9);
[0038]
[0039] Preferably, the coordinates Q of multiple groups of light spots are obtained by moving the calibration plate. i And the corresponding laser distance value H i , establish the constraint function f(L x , L y , L z ,θ x ,θ y ), the constraint function is as shown in formula (10); when the function f is the smallest, the combined parameters are optimal, as shown in formula (11);
[0040]
[0041] Where Q 1 is the coordinate of the point obtained by the binocular stereo vision algorithm, H 1 is the laser distance measurement value, Q' 1 The coordinates of the point are calculated from the calibration combination parameters and the laser ranging value.
[0042]
[0043] in The combined parameters when the optimization function f takes the minimum value;
[0044] Input the initial value of the combined parameters, move the calibration plate and use the obtained coordinate data for iterative optimization. The optimization ends when the calibration accuracy requirement is met or the number of iterations reaches the maximum value.
[0045] The LM optimization algorithm execution steps are as follows:
[0046] 1) Input the initial value of the combined parameter x 0 , set the initial trust radius μ 0 =0.05, accuracy ε=1×10 -10 , the maximum number of iterations C = 500;
[0047] 2) Calculate the step size Δx at the nth iteration n , the solution formula is shown in formula (12), and the parameter Δx is updated n+1 =x n +Δx n , the number of iterations C = C + 1;
[0048] Δx n =-(J(x n ) T J(x n )+μn I) -1 J(x n ) T f(x n ) (12)
[0049] 3) If f(x n+1 )<f(x n ) and Δx n When <ε, or the number of iterations C>500, stop the iteration and output the optimization result. Go to step 2);
[0050] 4) If f(x n+1 )≥f(x n ), then take μ n+1 =5μ n Go to step 2).
[0051] The existing pose combination calibration method between camera and laser rangefinder is not applicable enough in the 10kV live working safety distance detection scenario. A relative pose calibration method between binocular camera and single-point laser rangefinder based on the fusion of Gaussian fitting and LM algorithm is proposed. The center pixel of the spot in the binocular image is extracted by Gaussian fitting method, and then the three-dimensional coordinates of the pixel in the camera coordinate system are obtained by stereo vision algorithm. The initial value of pose combination calibration is obtained by combining the laser ranging value and the coordinate value. The constraint equation of the pose combination parameters of the above two sensors is established, and the calibration plate is moved to obtain multiple groups of pixel point three-dimensional coordinate values and corresponding laser ranging values. The dual sensor combination calibration parameters are optimized by iterative algorithm. The measured results show that after the method is used to calibrate the binocular camera and single-point laser rangefinder system, the relative error rate of the system is 2.01% within the measurement range of 15-20m at the work site, and the error fluctuation is significantly reduced, which meets the ranging accuracy requirements of the 10kV live working site. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 It is a schematic diagram of the relative positions of the two sensors.
[0053] Figure 2 It is a schematic diagram of the relative positions of the established coordinate systems.
[0054] Figure 3 It is a schematic diagram of the calibration model.
[0055] Figure 4 It is a schematic diagram of the equipment composition.
[0056] Figure 5 is the coordinate value of the light spot in the camera coordinate system.
[0057] Figure 6 is the error corresponding to different numbers of iteration points.
[0058] Figure 7 is the error corresponding to different experimental distances.
[0059] Figure 8 This is the binocular camera viewing angle diagram in the laboratory ranging experiment.
[0060] Fig. 9 This is a comparison chart of measurement errors.
[0061] Fig.10 This is a schematic diagram of the work site environment and device installation; (a) work site environment (b) measurement device installation location. DETAILED DESCRIPTION
[0062] 1. Principle of dual-sensor posture combination parameter calibration
[0063] The single-point laser light is a ray with the laser optical center as the endpoint in the camera coordinate system. The spatial position relationship between the camera and the laser is calibrated. Its ultimate goal is to determine the offset of the laser optical center relative to the origin in the camera coordinate system and the angle between the laser light and the positive direction of the camera coordinate system coordinate axis. The three-dimensional coordinates of the light spot hit by the single-point laser on the calibration plate are obtained through the stereo vision algorithm of the binocular camera. The three-dimensional coordinates of another light spot are obtained by moving the calibration plate. The initial value of the combination parameter is calculated based on the distance between the two points measured by the single-point laser, and a spatial position constraint relationship is established. The final posture combination parameters are solved by optimizing the constraint relationship. 1.1 Sensor installation and coordinate transformation principle
[0064] Fix the binocular camera and the single-point laser rangefinder on a tripod. The dotted line represents the laser beam. Their relative positions are shown in Figure 1 shown.
[0065] The calibration process involves the conversion of multiple coordinate systems. The relative positions of the coordinate systems to be established are as follows: Figure 2 As shown, the coordinate system is established as follows: 1) Camera coordinate system: The camera optical center O c Establish the camera coordinate system O as the origin c X c Y c Z c , the X axis is parallel to the mirror surface to the right, the Y axis is along the camera mirror surface downward, and the Z axis is perpendicular to the camera mirror surface outward. The coordinates of the light spot Q in this coordinate system are expressed as Q(X q , Y q , Z q ).
[0066] 2) Pixel coordinate system: The image pixel coordinate system OUV is established with the upper left corner of the imaging plane as the origin, with the U axis parallel to the image plane and horizontal to the right, and the V axis perpendicular to the U axis downward. The coordinates of the light spot Q in this coordinate system are expressed as q(u, v).
[0067] 3) Physical coordinate system: Establish the physical coordinate system O” with the intersection of the camera optical axis on the imaging plane as the origin c X” c Y" c Z” c , the X-axis is parallel to the U-axis of the pixel coordinate system, and the Y-axis is parallel to the V-axis of the pixel coordinate system.
[0068] The coordinate transformation relationship between the pixel coordinate system and the physical coordinate system satisfies equation (1).
[0069]
[0070] Where dx and dy are the physical sizes of a single pixel in the U and V axis directions. 0 , V 0 ) is the coordinate of the principal point of the image. The coordinate transformation relationship between the physical coordinate system and the camera coordinate system satisfies equation (2).
[0071]
[0072] Where f is the focal length of the camera, (X q , Y q , Z q ) is the coordinate of point Q in the camera coordinate system, Z q is the depth value of point Q in the camera coordinate system.
[0073] Combining equation (1) and equation (2) gives equation (3), which is the transformation relationship that the binocular camera pixel coordinate system and the camera coordinate system must satisfy.
[0074]
[0075] Where s is the torsion factor.
[0076] 1.2 Principle of combined parameter calibration
[0077] The three-dimensional position coordinates of the imaging point of the single-point laser rangefinder in the camera coordinate system satisfy equation (4).
[0078]
[0079] where θ x ,θ y ,θ z are the angles between the laser light and the positive direction of the camera coordinate system axis, L x , L y , L z are the offsets of the laser optical center relative to the origin of the camera coordinate system, and H is the distance measured by the single-point laser rangefinder.
[0080] According to the spatial position relationship, the angle θ between the light and the three coordinate axes is x,θ y ,θ z Formula (5) should be satisfied.
[0081] cos 2 θ x +cos 2 θ y+ cos 2 θ z =1 (5)
[0082] Combining equation (4) and equation (5) we get equation (6).
[0083]
[0084] Among them, L x , L y , L z ,θ x ,θ y That is, the present invention requires calibrated and mutually independent posture combination parameters.
[0085] 1.3 Gaussian fitting method to extract the pixel coordinates of the center of the spot
[0086] The single-point laser rangefinder will form a divergent light spot on the calibration plate. The calibration process of the present invention needs to obtain the coordinates of the pixel point at the center of the light spot, so it is necessary to locate the center of the laser light spot. Traditional methods are mainly divided into grayscale-based positioning algorithms and edge-based positioning algorithms. Considering the irregular shape of the light spot and the poor edge quality in actual situations, the present invention uses the Gaussian fitting method to extract the coordinates of the pixel point at the center of the light spot.
[0087] The intensity distribution of the laser beam can be approximated as a Gaussian distribution, and its Gaussian function is expressed as formula (5).
[0088]
[0089] Where M is the peak parameter of Gaussian distribution, (σ x , σ y ) are the distribution parameters in the x and y directions respectively.
[0090] Taking the logarithm of both ends of equation (7) and expanding the square term, we can get equation (8).
[0091]
[0092] The least squares method is used to perform Gaussian fitting on the spot area in the image, and the fitting extreme point (x 0 ,y 0 ) as the center pixel of the light spot.
[0093] 2. Dual sensor posture combination parameter calibration process
[0094] The present invention realizes the calibration of posture combination parameters through two steps. The first step: fix the relative position of the camera and the single-point laser rangefinder to ensure that the imaging point can be seen on the calibration plate in the camera field of view, use the Gaussian fitting method to extract the pixel coordinates of the center of the imaging spot, combine the distance length measured by the single-point laser, calculate its three-dimensional coordinates in the camera coordinate system, move the calibration plate to obtain the three-dimensional coordinate value of another point, and solve the initial value of the combination parameter jointly; the second step: obtain multiple groups of three-dimensional coordinate values by moving the calibration plate multiple times, establish a constraint function, optimize through an iterative algorithm, and obtain the final parameters.
[0095] 2.1 Calculation of initial values of combined parameters
[0096] According to the above process, a calibration model is established, such as Figure 3 shown.
[0097] Fix the binocular camera and single-point laser rangefinder to ensure the laser imaging point Q 1 Falling on the calibration plate S 1 Using Gaussian fitting method to extract Q 1 Center pixel coordinate q 1 (x q1 ,y q1 ), point Q is obtained through binocular stereo vision algorithm 1 The depth value Z q1 , Substitute the coordinate value and depth value into formula (3) to obtain point Q 1 Coordinate Q in the camera coordinate system 1 (X q1 , Y q1 , Z q1 ), using the laser rangefinder to obtain the distance value H 1 ; Keep the relative position of the sensor unchanged and move the calibration plate to S 2 Position, at this time the laser is at S 2 The imaging point of the plane is Q 2 , using Gaussian fitting method to extract Q 2 Center pixel coordinate q 2 (x q2 ,y q2 ), the binocular camera gets the point Q2 The depth value is Z q2 , and Q can be obtained by the same logic 2 The three-dimensional coordinate Q in the camera coordinate system 2 (X q2 , Y q2 , Z q2 ), and the distance value H 2 .
[0098] Substituting the above values into equation (9), the simultaneous equations are solved to obtain the initial values of the combined parameters. The solution formula is shown in equation (9).
[0099]
[0100] 2.2LM optimization algorithm
[0101] During the calibration process, when the obtained combined parameters and the single-point laser ranging value are accurate enough, the three-dimensional coordinates Q'(X' q , Y' q , Z' q ) should be consistent with the three-dimensional coordinates given by the binocular camera. The initial value of the combined parameter obtained by the method described in 2.1 of the present invention is only derived from two light spots, and there is a large error. In order to obtain accurate combined parameters, the present invention optimizes the initial value of the combined parameter by establishing a constraint function. The coordinates Q of multiple groups of light spots are obtained by moving the calibration plate i And the corresponding laser distance value H i , establish the constraint function f(L x , L y , L z ,θ x ,θ y ), the constraint function is as shown in formula (10). When the function f is the smallest, the combined parameters are optimal, as shown in formula (11).
[0102]
[0103] Where Q 1 is the coordinate of the point obtained by the binocular stereo vision algorithm, H 1 is the laser distance measurement value, Q' 1 The coordinates of the point are calculated from the calibration combination parameters and the laser ranging value.
[0104]
[0105] in The combined parameters when the optimization function f takes the minimum.
[0106] The LM (Levenberg-Marquard) algorithm is an algorithm widely used in camera calibration, visual SLAM and other fields. Its search method uses the trust region method, which provides a more robust and accurate step size at the expense of some convergence speed, and avoids the non-singularity and ill-conditioned problems of the coefficient matrix to a certain extent. This method is used to optimize the combined parameters. Enter the initial value of the combined parameters, move the calibration plate and use the obtained coordinate data for iterative optimization, and end when the calibration accuracy requirement is met or the number of iterations reaches the maximum value.
[0107] The LM optimization algorithm execution steps are as follows:
[0108] 1) Input the initial value of the combined parameter x 0, set the initial trust radius μ 0 =0.05, accuracy ε=1×10 -10 , the maximum number of iterations C = 500;
[0109] 2) Calculate the step size Δx at the nth iteration n , the solution formula is shown in formula (12), and the parameter Δx is updated n+1 =x n +Δx n , the number of iterations C = C + 1;
[0110] Δx n =-(J(x n ) T J(x n )+μ n I) -1 J(x n ) T f(x n ) (12)
[0111] 3) If f(x n+1 )<f(x n ) and Δx n When <ε, or the number of iterations C>500, stop the iteration and output the optimization result. Go to step 2);
[0112] 4) If f(x n+1 )≥f(x n ), then take μ n+1 =5μ n Go to step 2).
[0113] 3. Analysis of calibration and ranging experiment results
[0114] 3.1 Experimental equipment construction
[0115] The sensor equipment parameters are shown in Table 1 and Table 2.
[0116] Table 1 Parameters of single-point laser rangefinder
[0117]
[0118] Table 2 Binocular camera parameters
[0119]
[0120] During the experiment, the relative position of the binocular camera and the single-point laser rangefinder on the tripod is kept unchanged, and a servo gimbal controlled by a single-chip microcomputer is installed in the middle to rotate the fixed dual sensor parts in different directions as a whole to avoid changes in the relative position of the dual sensor parts caused by manual adjustment. Figure 4 shown.
[0121] 3.2 Combined parameter calibration experiment
[0122] The sensor posture combination calibration process is as follows:
[0123] 1) Fixing device;
[0124] 2) Use the Gaussian fitting method to extract the pixel point at the center of the spot on the calibration plate. Substitute the coordinate value of this point into equation (3) to find its corresponding coordinate value in the camera coordinate system;
[0125] 3) Move the calibration plate along the Z axis and repeat step 2) to obtain the coordinate value of another light spot, and substitute the single-point laser ranging value into formula (9) to solve the initial value of the combined parameter;
[0126] 4) Repeat step 3) to obtain multiple sets of light spot coordinate values and laser ranging values, and use the LM algorithm to minimize the function f to optimize the combination parameters;
[0127] 5) Optimization is completed and the final combination parameters are output.
[0128] In order to verify the effectiveness of the pose combination calibration method, the Euclidean distance between the light spot coordinates obtained by the binocular stereo vision algorithm and the coordinates obtained by the method of the present invention is used as an evaluation criterion. The evaluation function is shown in formula (13).
[0129]
[0130] Wherein (x', y', z') is the coordinate of a certain point obtained by the method of the present invention, (x, y, z) is the coordinate of the point obtained by the binocular stereo vision algorithm, and ρ is the calibration accuracy.
[0131] The calibration plate is moved 200 times in the calibration process, and the coordinate values of these light spots in the camera coordinate system are extracted. The position of the light spot in the camera coordinate system is represented by ·, such as Figure 5 shown.
[0132] Select two of the light spots and substitute them into formula (9) to calculate the initial value of the combination parameter. The selected light spot values are shown in Table 3.
[0133] Table 3 Light spot numerical information of the initial value of the calibration combination parameters
[0134]
[0135] The above 200 light spot data are processed and then put into formula (8), and the initial values of the combined parameters in Table 3 are optimized using the LM algorithm.
[0136] The combination parameters when the number of light spots involved in the iteration is different are obtained respectively. Starting from 5 light spots, the number of spots is gradually increased. The combination parameters obtained by iterative optimization with different numbers of light spots are combined with the ranging value of the single-point laser to perform 3D reconstruction of the corresponding coordinate points. The 3D coordinates of the light spot read by the binocular camera are taken as the true value. The calibration accuracy is calculated according to formula (11). The average error curve between the 3D coordinates reconstructed with different numbers of iterative points and the true value of the coordinates is obtained as shown in the figure below: Figure 6 shown.
[0137] Depend on Figure 6 It can be seen that the average single-point error of the combined parameters obtained by optimizing only 5 light spots in the measurement exceeds 6 mm. As the number of light spots participating in the iteration increases, the average error gradually decreases. When the number reaches 50, the error is less than 1 mm. There is no obvious downward trend in the error after continuing to increase the number of points.
[0138] 3.3 Error analysis of ranging experiment results
[0139] Considering the actual situation of 10kV live working site, the distance between the device and the working position ranges from 15 to 20 meters. It is necessary to first test the accuracy of the combined parameters at different distances in a laboratory environment. Under the premise of ensuring good lighting conditions, gradually increase the measurement distance, and take the actual measured distance on site as the true value. The curve of the distance measurement error changing with distance is shown in the figure below. Figure 7 shown.
[0140] Depend on Figure 7 It can be seen that the distance measurement error is 3.02mm when the actual distance is 15m, and the error increases with the increase of distance. When the actual distance increases to 20m, the distance measurement error is 4.51mm. In the measurement range of 15-20m, the average distance measurement error is 3.41mm.
[0141] In order to further verify the reliability of the calibration method described in the present invention in distance measurement, the distance measurement accuracy of the calibration device was verified in a laboratory environment. Figure 8 The distance measurement of the binocular camera within the working range is corrected by utilizing the ranging advantage of the single-point laser, so that the binocular camera can obtain more accurate world coordinates of the target point, thereby reducing the error in measuring the distance between multiple objects.
[0142] The measurement results obtained by the method of the present invention, the system without laser correction and the system calibrated by the method described in Reference 1 are compared. The experiment is carried out in a measurement range of 15-20m. The distance between the device and the calibration plate is increased by 0.2m each time from 15m. The actual size of the calibration plate 0.5×0.5m is taken as the true value, and the size of the calibration plate obtained by the binocular camera is taken as the measured value. The error curve is shown in Fig. 9 shown.
[0143] Reference 1: Chu Zhaobi, Li Zipeng, Gao Jinhui. A calibration method based on binocular camera and single-point laser rangefinder[J]. Chinese Journal of Scientific Instrument, 2023, 44(05): 232-239.
[0144] from Fig. 9 It can be seen that under the same experimental environment, after calibration by the method described in the present invention, the system device can perform distance measurement, and the distance measurement effect of the single-point laser on the binocular camera is improved, and the measurement error is significantly reduced. Fig. 9 The mean and variance of the error data are shown in Table 4.
[0145] It can be seen from Table 4 that after the device calibrated by the calibration method of the present invention is used for ranging correction, smaller measurement error and variance can be obtained, the measurement accuracy is improved, and the error fluctuation is smaller than that without correction, which fully demonstrates the feasibility of the calibration method of the present invention.
[0146] Table 4 Mean and variance of experimental data
[0147]
[0148] The device calibrated by the method of the present invention is used to measure the safety distance of the actual live working site. The working site environment and device installation are as follows: Fig.10 shown.
[0149] In the actual live working site safety distance measurement, the size of the calibration plate attached to the boom truck was measured. The measurement was repeated 20 times to take the average value. The calibration plate size obtained by the device was 49.07×48.92cm. The average error rate was 2.01% within the distance range of 15-20m, which can meet the error requirements of safety distance measurement.
[0150] The above experimental results show that when the device calibrated by the method of the present invention is used for distance measurement in a laboratory environment of 15-20m, the average error is 7.798mm and the variance is 0.175mm. 2 When the device is used in a 10kV live operation site, within the same operating range, the absolute error of the measurement is 0.01005m for a 0.5×0.5m calibration plate.
[0151] 4. Conclusion
[0152] (1) The method of combined calibration of the dual sensor posture of binocular camera and single-point laser rangefinder uses the Gaussian fitting method to quickly and accurately extract the center pixel of the spot, and uses the three-dimensional coordinates obtained by the binocular stereo vision algorithm and the laser ranging value to calculate the initial value of the combined parameter. The mobile calibration plate obtains multiple sets of point data, and the LM algorithm is used to optimize the initial value to obtain the final independent combined parameters, avoiding parameter coupling.
[0153] (2) After the dual-sensor posture combined calibration method is used to calibrate the binocular camera and single-point laser combined ranging system, the average error rate of measuring the calibration plate size is 1.56% in a laboratory environment with a measurement range of 15-20m; in a complex scenario of 10kV live working, with a measurement range of 15-20m, the average ranging error rate is 2.01%, which can meet the accuracy requirements of live working safety distance measurement.
Claims
1. A combined calibration method of binocular camera and single-point laser rangefinder applied to 10kV live working scene, characterized in that: The three-dimensional coordinates of the center pixel of the light spot obtained by the Gaussian fitting method in the camera coordinate system are obtained through the binocular stereo vision algorithm, and the initial value of the pose combination calibration is obtained based on the distance between the three-dimensional coordinates and the single-point laser measurement. The constraint function of the spatial position parameters is established and the calibration plate is moved to obtain the three-dimensional coordinates of multiple groups of pixel points. The constraint function is optimized by the LM algorithm to obtain mutually independent pose combination calibration parameters.
2. The combined calibration method of binocular camera and single-point laser rangefinder applied to 10kV live working scene according to claim 1 is characterized in that: The three-dimensional coordinates of the light spot hit by a single-point laser on the calibration plate are obtained through the stereo vision algorithm of the binocular camera. The three-dimensional coordinates of another light spot are obtained by moving the calibration plate. The initial value of the combination parameter is calculated based on the distance between the two points measured by the single-point laser, and the spatial position constraint relationship is established. The final posture combination parameters are solved by optimizing the constraint relationship.
3. According to claim 2, a combined calibration method of binocular camera and single-point laser rangefinder applied to 10kV live working scene is characterized in that: Create the following coordinate system: 1) Camera coordinate system: camera optical center O c Establish the camera coordinate system O as the origin c X c Y c Z c , the X-axis is parallel to the mirror surface and points to the right, the Y-axis is along the camera mirror surface and points downward, and the Z-axis is perpendicular to the camera mirror surface and points outward; The coordinates of the light spot Q in this coordinate system are expressed as Q(X q , Y q , Z q ); 2) Pixel coordinate system: The image pixel coordinate system OUV is established with the upper left corner of the imaging plane as the origin, with the U axis parallel to the image plane and horizontal to the right, and the V axis perpendicular to the U axis and downward; the coordinates of the light spot Q in this coordinate system are expressed as q(u, v); 3) Physical coordinate system: Establish the physical coordinate system O” with the intersection of the camera optical axis on the imaging plane as the origin c X” c Y" c Z” c , the X-axis is parallel to the U-axis of the pixel coordinate system, and the Y-axis is parallel to the V-axis of the pixel coordinate system; The coordinate transformation relationship between the pixel coordinate system and the physical coordinate system satisfies equation (1); Where dx and dy are the physical dimensions of a single pixel in the U-axis and V-axis directions, and (U0, V0) are the coordinates of the principal point of the image; The coordinate transformation relationship between the physical coordinate system and the camera coordinate system satisfies equation (2); Where f is the focal length of the camera, (X q , Y q , Z q ) is the coordinate of point Q in the camera coordinate system, Z q is the depth value of point Q in the camera coordinate system; Combining equation (1) and equation (2) gives equation (3), which is the transformation relationship that the binocular camera pixel coordinate system and the camera coordinate system must satisfy; Where s is the torsion factor.
4. The combined calibration method of binocular camera and single-point laser rangefinder applied to 10kV live working scene according to claim 3 is characterized in that: The three-dimensional position coordinates of the imaging point of the single-point laser rangefinder in the camera coordinate system satisfy equation (4); where θ x ,θ y ,θ z are the angles between the laser light and the positive direction of the camera coordinate system axis, L x , L y , L z are the offsets of the laser optical center relative to the origin of the camera coordinate system, and H is the distance measured by the single-point laser rangefinder; According to the spatial position relationship, the angle θ between the light and the three coordinate axes is x ,θ y ,θ z Should satisfy formula (5); cos 2 i x +cos 2 i y+ cos 2 i z =1 (5) Combining equation (4) and equation (5) we get equation (6); Among them, L x , L y , L z ,θ x ,θ y That is, the pose combination parameters need to be calibrated and are independent of each other.
5. The combined calibration method of binocular camera and single-point laser rangefinder applied to 10kV live working scene according to claim 4 is characterized in that: The Gaussian fitting method is used to extract the coordinates of the pixel point at the center of the light spot; The intensity distribution of the laser beam is approximately Gaussian distribution, and its Gaussian function is expressed as formula (5); Where M is the peak parameter of Gaussian distribution, (σ x , σ y ) are the distribution parameters in the x and y directions respectively; Taking the logarithm of both ends of equation (7) and expanding the square term, we can get equation (8); The least squares method is used to perform Gaussian fitting on the spot area in the image, and the fitting extreme point (x0, y0) is used as the center pixel of the spot.
6. The combined calibration method of binocular camera and single-point laser rangefinder applied to 10kV live working scene according to claim 1 or 5, characterized in that: The calibration of the pose combination parameters is achieved through the following steps: Step 1: Fix the relative position of the camera and the single-point laser rangefinder to ensure that the imaging point can be seen on the calibration plate in the camera's field of view. Use the Gaussian fitting method to extract the pixel coordinates of the center of the imaging spot. Combined with the distance length measured by the single-point laser, calculate its three-dimensional coordinates in the camera coordinate system. Move the calibration plate to obtain the three-dimensional coordinate value of another point, and solve the initial value of the combined parameters together. Step 2: Get multiple sets of 3D coordinate values by moving the calibration plate multiple times, establish constraint functions, and optimize through iterative algorithms to get the final parameters.
7. The combined calibration method of binocular camera and single-point laser rangefinder applied to 10kV live working scene according to claim 6 is characterized in that: The Gaussian fitting method is used to extract the central pixel coordinates of Q1 (q1(x q1 ,y q1 ), the depth value Z of point Q1 is obtained through binocular stereo vision algorithm q1 Substitute the coordinate value and depth value into equation (3) to obtain the coordinate Q1 (X q1 , Y q1 , Z q1 ), use the laser rangefinder to get the distance value H1; keep the relative position of the sensor unchanged, move the calibration plate to the S2 position, at this time the laser imaging point on the S2 plane is Q2, and use the Gaussian fitting method to extract the center pixel coordinates q2 (x q2 ,y q2 ), the binocular camera gets the point Q2 The depth value is Z q2 Similarly, we can obtain the three-dimensional coordinates of Q2 in the camera coordinate system (X q2 , Y q2 , Z q2 ), and the ranging value H2; Substitute the above values into equation (9), solve the simultaneous equations to obtain the initial values of the combined parameters, and the solution formula is shown in equation (9); 8. The combined calibration method of binocular camera and single-point laser rangefinder applied to 10kV live working scene according to claim 7 is characterized in that: Obtain the coordinates Q of multiple sets of light spots by moving the calibration plate i And the corresponding laser distance value H i , establish the constraint function f(L x , L y , L z ,θ x ,θ y ), the constraint function is as shown in formula (10); when the function f is the smallest, the combined parameters are optimal, as shown in formula (11); Where Q1 is the coordinate of the point obtained by binocular stereo vision algorithm, H1 is the laser ranging value, and Q'1 is the coordinate of the point calculated by the calibration combination parameters and the laser ranging value; in The combined parameters when the optimization function f takes the minimum value; Input the initial value of the combined parameters, move the calibration plate and use the obtained coordinate data for iterative optimization. The optimization ends when the calibration accuracy requirement is met or the number of iterations reaches the maximum value. The LM optimization algorithm execution steps are as follows: 1) Input the initial value of the combined parameter x0, set the initial trust radius μ0 = 0.05, and the accuracy ε = 1×10 -10 , the maximum number of iterations C = 500; 2) Calculate the step length △x at the nth iteration n , the solution formula is shown in formula (12), and the parameter △x is updated n+1 =x n +Δx n , the number of iterations C = C + 1; Δx n =-(J(x n ) T J(x n )+μ n I) -1 J(x n ) T f(x n ) (12) 3) If f(x n+1 ) <f(x n ) and Δx n <ε, or the number of iterations C>500, stop the iteration and output the optimization result; otherwise, Go to step 2); 4) If f(x n+1 )≥f(x n ), then take μ n+1 =5μ n Go to step 2).
Citation Information
Cited By
On-orbit calibration method of vision and laser fused space target pose measurement system
CN119146857A
Calibration-free large structure deformation laser vision fusion measurement method and system
CN120368870A
Calibration-free laser vision fusion measurement method and system for large-scale structure deformation
CN120368870B
Online measurement and compensation method for geometric error of numerical control machine tool
CN121680277A
