Global calibration method for multi-line laser sensor
By designing a global calibration method of multi-line laser sensors in PVC floor buckle type inspection, the transformation matrix is calculated using rigid transformation and bidirectional ICP algorithm, the problem of insufficient board waste and qualitative analysis is solved, and efficient and accurate multi-sensor calibration and digitization are achieved. Measurement.
Patent Information
- Application Number
- CN202411562409.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-05
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art has problems of waste of sheets and insufficient qualitative analysis in the PVC floor buckle accuracy test, and the multi-sensor global calibration method requires high-precision equipment or multiple moving targets, which is inefficient.
Design a global calibration method for multi-line laser sensors. By placing line laser sensors at four different angles, obtaining contour data, using coarse registration and precise registration methods, including rigid transformation and bidirectional ICP algorithms, the transformation matrix is calculated to achieve Global calibration of multi-sensors.
High-precision global calibration of multi-sensors is realized, measuring efficiency and accuracy is improved, plate waste is reduced, and digital contour information is provided.
Smart Images

Figure CN119063654A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of laser sensor detection, and in particular to a multi-line laser sensor global calibration method. Background Art
[0002] At present, the precision inspection of PVC floor buckle profile mainly relies on the film projection method, that is, randomly sampling samples for slicing during the production process, and comparing the slices with the film of the standard profile through a projector to detect the deviation of the buckle size. In actual operation, this method usually requires randomly sampling a piece of board for cutting and testing every 20 minutes, resulting in a large amount of board waste. In addition, the film projection method can only perform qualitative analysis and cannot provide digital profile information. The measurement process is also relatively time-consuming.
[0003] In order to achieve efficient and non-destructive topography measurement, multi-angle line laser sensors are used to solve the problem of high-precision measurement of PVC floor buckles, and different sensors need to be globally calibrated. At present, there are two main methods for multi-sensor global calibration: the calibration method using high-precision measuring equipment and the calibration method without high-precision measuring equipment. The calibration method of high-precision measuring equipment requires dual theodolites, laser trackers, etc. to complete global calibration. The multi-sensor global calibration method without high-precision measuring equipment requires the design of targets and multiple movement of targets to achieve calibration.
[0004] Therefore, a special high-precision three-dimensional target is designed to perform global calibration of multiple line laser sensors at one time. Summary of the invention
[0005] In order to solve the above technical problems and shortcomings: how to design a special high-precision three-dimensional target and perform global calibration on multiple line laser sensors at one time, the present invention provides a multi-line laser sensor global calibration method.
[0006] In order to achieve the above-mentioned purpose and other related purposes, the present invention adopts the following technical solutions:
[0007] A multi-line laser sensor global calibration method comprises the following steps:
[0008] Step 1: Line laser sensors are placed at four different angles and photographed along the calibration block. Each sensor obtains the contour data of the feature point intervals A, B, C, and D respectively, extracts the contour of the calibration block photographed by each sensor, and finds the common fields of view AB and CD in the contours of the image coordinate systems of sensor 1, sensor 2, sensor 3, and sensor 4 respectively. The calibration is completed by aligning the common fields of view of different sensors.
[0009] Step 2: After obtaining the coordinates of each feature point, the coordinate system of sensor 1 is used as the reference coordinate system, and the feature point vectors in different coordinate systems are roughly aligned by rigid transformation, and then the transformation matrix of each coordinate system after rough alignment is obtained, and the rotation angle of the coordinate system to be aligned relative to the reference coordinate system is calculated, and then the rotation matrix R of the coordinate system to be aligned relative to the reference coordinate system can be obtained. 21 , translation matrix t 21 ;
[0010] Step 3: Use the bidirectional ICP algorithm for precise alignment, and establish a double KD tree for bidirectional search to find matching point pairs more accurately.
[0011] Preferably, in step one, the coordinates of feature points in a common field of view under different sensors are determined based on local curvature, and the curvature of the two-dimensional contour point cloud is solved using a three-point method.
[0012] Preferably, in step one, the point cloud data is filtered using a Savitzky-Golay algorithm. The Savitzky-Golay filtering algorithm applies the least squares method in each moving window to perform polynomial fitting on the data points to obtain a smoothed contour, and the edge feature points are found by calculating the curvature of each point on the contour curve.
[0013] Preferably, the step of fine registration in step three includes:
[0014] S31, double KD tree construction: construct KD trees for the source point cloud P and the target point cloud Q respectively;
[0015] S32, Bidirectional nearest neighbor search: Search for point p in the source point cloud P from the target point cloud Q i The nearest neighbor point of the source point cloud P i , search for its nearest neighbor point q in the target point cloud Q i , search for q in the target point cloud Q in the source point cloud P i The nearest neighbor point p j , if p i and p j is the same point, that is, p i =p j , then p i and q i Form a matching point pair with a one-to-one correspondence; otherwise, continue to search for the next nearest point q in the target point cloud Q (i+1) , and repeat the above steps until a point pair that meets the conditions is found, or all point cloud contours are traversed;
[0016] S33, point pair transformation matrix calculation correspondence: find a set of matching point pairs, and calculate the corresponding rotation matrix R and translation matrix t;
[0017] S34, Error calculation and iteration: Calculate the mean square error of the transformation matrix, determine the iteration termination condition, and set the threshold , and set the maximum number of iterations; when Or the maximum number of iterations is reached and the iteration is terminated. Otherwise, the calculated transformation matrix transforms the source point cloud and updates the point cloud P, and repeats steps S31 to S33. After iteration, the final transformation matrix R f and t f ,The point cloud after rough registration is accurately registered on the target point cloud.
[0018] In summary, the invention has at least one of the following beneficial technical effects:
[0019] 1. This method uses different line laser sensors to obtain the contour data of the three-dimensional calibration block in different directions, and extracts the common field of view data obtained by each sensor based on the curvature feature. The common field of view of each sensor is roughly and precisely aligned, and then the transformation matrix of each sensor is obtained to achieve global calibration of multiple sensors;
[0020] 2. Due to the influence of noise during the measurement process, some point cloud data is prone to floating, which may cause the point with the maximum curvature to not be the edge feature point of the calibration block, thus affecting the accuracy of edge coordinate extraction. In order to improve the calibration accuracy and accurately extract the feature point coordinates, the Savitzky-Golay algorithm is used to filter the point cloud data;
[0021] 3. The two-dimensional rigid transformation method can achieve a rough registration of two sets of contours, but its ability to handle nonlinear errors is limited. It can only handle simple rotations and translations and cannot correct more complex geometric mismatches. In order to improve the matching accuracy, the bidirectional ICP algorithm is used for fine registration. By establishing a double KD tree for bidirectional search, matching point pairs can be found more accurately. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 It is a layout diagram of multiple line laser sensors of an embodiment of the invention; Figure 2 (a) is a contour image of a calibration block taken by a sensor 1 according to an embodiment of the present invention; Figure 2 (b) is a contour image of the calibration block taken by the second sensor of the embodiment of the invention; Figure 2 (c) is a contour image of the calibration block taken by sensor 3 of the embodiment of the invention; Figure 2 (d) is a contour image of the calibration block taken by sensor 4 of the embodiment of the invention; Figure 3 It is a feature point extraction diagram of an embodiment of the present invention; Figure 4 It is the pose graph of the public field of view after calibration. DETAILED DESCRIPTION
[0023] The following is an explanation of the embodiments of the present invention by specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.
[0024] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention. The illustrations only show the components related to the present invention rather than being drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the type, quantity and proportion of each component can be changed at will, and the component layout type may also be more complicated.
[0025] The following is combined with Figures 1 to 4 The specific implementation methods of the present invention are further described.
[0026] Example:
[0027] The invention discloses a multi-line laser sensor global calibration method, which uses different line laser sensors to obtain contour data in different directions of a three-dimensional calibration block, and extracts the common field of view data obtained by each sensor based on the curvature feature. The common field of view of each sensor is roughly aligned and precisely aligned, and then the transformation matrix of each sensor is obtained to achieve global calibration of multiple sensors.
[0028] The specific steps include:
[0029] Step 1: Layout design of different sensors (line laser sensors) and design of 3D calibration blocks, e.g. Figure 1 The line laser sensors are placed at four different angles and photographed along the calibration block. Each sensor obtains the contour data of the feature point intervals A, B, C, and D, and extracts the contour of the calibration block photographed by each sensor, as shown in Figure 2 (a) to (d) in Figure 2 are the image coordinate system contours of sensor 1, sensor 2, sensor 3, and sensor 4. Find the common fields of view AB and CD in the contours of the image coordinate systems of sensor 1, sensor 2, sensor 3, and sensor 4, respectively, and complete the calibration by aligning the common fields of view of different sensors.
[0030] The line laser is projected into the contour point cloud of the calibration block section, and the curvature at the edge reaches the maximum value in its neighborhood. Therefore, the coordinates of the feature points in the common field of view under different sensors are determined based on the local curvature, and the three-point method is used to solve the curvature of the two-dimensional contour point cloud. As shown in the following formula.
[0031]
[0032]
[0033]
[0034] a, b, c are three adjacent points p i-1 (x i-1 ,y i-1 ), p i (x i ,y i ), p i+1 (x i+1 ,y i+1 ) is the side length of the triangle, l is the semi-perimeter of the triangle, the area S of the triangle can be obtained, and then the curvature k can be obtained.
[0035]
[0036] Due to the influence of noise in the measurement process, some point cloud data is prone to floating, which may cause the point with the maximum curvature to be not the edge feature point of the calibration block, thus affecting the accuracy of edge coordinate extraction. In order to improve the calibration accuracy and accurately extract the coordinates of the feature points, the Savitzky-Golay algorithm is used to filter the point cloud data. The Savitzky-Golay (SG for short) filter is a special low-pass filter, also known as a Savitzky-Golay smoother or a digital smoothing polynomial filter. The Savitzky-Golay filter is a filtering method based on local polynomial least squares fitting in the time domain. It smoothes the data by fitting a polynomial in a sliding window, thereby removing noise while retaining the details and peaks of the data to the greatest extent. Specifically, it uses the data points in the sliding window as the input of the polynomial, and fits the filtered output value through the least squares method to obtain the smoothed contour. The edge feature points are found by calculating the curvature of each point on the contour curve. Such as Figure 3 As shown, the curvature map of the contour after filtering by sensor 2 and the positioning of the feature points are the positioning of the corresponding contour and feature points.
[0037] Step 2: After obtaining the coordinates of each feature point, the coordinate system of sensor 1 is used as the reference coordinate system, and the feature point vectors in different coordinate systems are roughly aligned by rigid transformation, and then the transformation matrix of each coordinate system after rough alignment is obtained.
[0038] The conversion of each coordinate system can be performed using the following transformation formula:
[0039] in, is the coordinate in the base coordinate system. is the coordinate of sensor i in the coordinate system, R i is the rotation matrix, T i is the translation vector.
[0040] Assume that the feature point vectors of the reference coordinate system and the coordinate system to be registered are v1=(a1,b1) and v2=(a2,b2). Normalized feature vectors:
[0041]
[0042] Calculate the rotation angle of the coordinate system to be registered relative to the reference coordinate system.
[0043]
[0044] θ 21 They are the rotation angles of the unit vector u2 relative to u1, and then the rotation matrix R of the coordinate system to be registered relative to the reference coordinate system can be obtained 21 ,
[0045]
[0046] Translation matrix t 21 :
[0047]
[0048] Step 3: The two-dimensional rigid transformation method can achieve a rough registration of the two sets of contours, but its ability to handle nonlinear errors is limited. It can only handle simple rotations and translations and cannot correct more complex geometric mismatches. In order to improve the matching accuracy, the bidirectional ICP algorithm is used for fine registration. By establishing a double KD tree for bidirectional search, matching point pairs can be found more accurately. The contour data after rough registration is defined as P = {p i ∈R 2 |i=1,…,N P} is the source point cloud, the reference coordinate system contour data Q={q i ∈R 2 |j=1,…,N Q} is the target point cloud, where N P , N Q is the total amount of precisely registered point cloud data, and N P =N Q .
[0049] The steps of fine registration in step 3 include:
[0050] S31, double KD tree construction: construct KD trees for the source point cloud P and the target point cloud Q respectively;
[0051] S32, Bidirectional nearest neighbor search: Search for point p in the source point cloud P from the target point cloud Q i The nearest neighbor point of the source point cloud P i , search for its nearest neighbor point q in the target point cloud Q i ,
[0052]
[0053] Search for q in the target point cloud Q in the source point cloud P i The nearest neighbor point p j :
[0054]
[0055] If p i and p j is the same point, that is, p i =p j , then p i and q i Form a matching point pair with a one-to-one correspondence; otherwise, continue to search for the next nearest point q in the target point cloud Q i+1 , and repeat the above steps until a point pair that meets the conditions is found, or all point cloud contours are traversed;
[0056] S33, point pair transformation matrix calculation correspondence: find a set of matching point pairs, and calculate the corresponding rotation matrix R and translation matrix t;
[0057] S34, Error calculation and iteration: Calculate the mean square error of the transformation matrix:
[0058]
[0059] Determine the iteration termination condition and set the threshold , and set the maximum number of iterations; when Or the maximum number of iterations is reached and the iteration is terminated. Otherwise, the calculated transformation matrix transforms the source point cloud and updates the point cloud P, and repeats steps S31 to S33. After iteration, the final transformation matrix R f and t f , the roughly registered point cloud is precisely registered on the target point cloud. By registering the contours of the calibration blocks under each sensor, the global calibration of multiple sensors can be achieved. The poses of the common field of view of each sensor before and after calibration are shown in Figure 4 shown.
[0060] The above embodiments are only illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, any equivalent changes made to the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A multi-line laser sensor global calibration method, characterized in that: The following steps are included: Step 1: Line laser sensors are placed at four different angles and photographed along the calibration block. Each sensor obtains the contour data of the feature point intervals A, B, C, and D respectively, extracts the contour of the calibration block photographed by each sensor, and finds the common fields of view AB and CD in the contours of the image coordinate systems of sensor 1, sensor 2, sensor 3, and sensor 4 respectively. The calibration is completed by aligning the common fields of view of different sensors. Step 2: After obtaining the coordinates of each feature point, the coordinate system of sensor 1 is used as the reference coordinate system, and the feature point vectors in different coordinate systems are roughly aligned by rigid transformation, and then the transformation matrix of each coordinate system after rough alignment is obtained, and the rotation angle of the coordinate system to be aligned relative to the reference coordinate system is calculated, and then the rotation matrix R of the coordinate system to be aligned relative to the reference coordinate system can be obtained. 21 , translation matrix t 21 ; Step 3: Use the bidirectional ICP algorithm for precise alignment, and establish a double KD tree for bidirectional search to find matching point pairs more accurately.
2. The multi-line laser sensor global calibration method according to claim 1, characterized in that: In step one, the coordinates of feature points in the common field of view under different sensors are determined based on the local curvature, and the curvature of the two-dimensional contour point cloud is solved using the three-point method.
3. The multi-line laser sensor global calibration method according to claim 1 or 2, characterized in that: In step 1, the point cloud data is filtered using the Savitzky-Golay algorithm. The Savitzky-Golay filtering algorithm applies the least squares method in each moving window to perform polynomial fitting on the data points to obtain a smoothed contour, and finds edge feature points by calculating the curvature of each point on the contour curve.
4. The multi-line laser sensor global calibration method according to claim 1, characterized in that: The steps of fine registration in step 3 include: S31, double KD tree construction: construct KD trees for the source point cloud P and the target point cloud Q respectively; S32, Bidirectional nearest neighbor search: Search for point p in the source point cloud P from the target point cloud Q i The nearest neighbor point of the source point cloud P i , search for its nearest neighbor point q in the target point cloud Q i , search for q in the target point cloud Q in the source point cloud P i The nearest neighbor point p j , if p i and p j is the same point, that is, p i =p j , then p i and q i Form a matching point pair with a one-to-one correspondence; otherwise, continue to search for the next nearest point q in the target point cloud Q (i+1) , and repeat the above steps until a point pair that meets the conditions is found, or all point cloud contours are traversed; S33, point pair transformation matrix calculation correspondence: find a set of matching point pairs, and calculate the corresponding rotation matrix R and translation matrix t; S34, Error calculation and iteration: Calculate the mean square error of the transformation matrix, determine the iteration termination condition, and set the threshold , and set the maximum number of iterations; when Or the maximum number of iterations is reached and the iteration is terminated. Otherwise, the calculated transformation matrix transforms the source point cloud and updates the point cloud P, and repeats steps S31 to S33. After iteration, the final transformation matrix R f and t f ,The point cloud after rough registration is accurately registered on the target point cloud.
Citation Information
Patent Citations
Spatial pose calibration method for multi-line laser sensor combined measurement
CN111578866A
Multi-stage vehicle-mounted laser point cloud road change monitoring method
CN112330661A
Laser sensor calibration method, robot and storage medium
CN116007659A
Point cloud splicing and registering system and method based on Goctor sensor
CN118134973A
Calibration method of multi-line laser sensor
CN118274736A
Cited By
Multi-line laser sensor global calibration method based on calibration block
CN120740488A
Locomotive vibration suppression method for railway dynamic continuous profile measurement
CN120831063A
Three-dimensional skin waviness fitting analysis method
CN121498585A
