Workpiece pose estimation method and system based on outer contour and laser rays

Abstract

The invention belongs to the related technical field of machine vision pose estimation, and discloses a workpiece pose estimation method and system based on an outer contour and a laser ray, and the method comprises the steps: S1, building a secondary database in which the workpiece pose corresponds to the contour and the laser ray through a simulation or experiment means; s2, acquiring an actually measured contour and an actually measured laser line of the workpiece, matching the actually measured contour and the actually measured laser line in the secondary database, and coarsely screening out a coarsely estimated attitude; s3, acquiring space coordinates of the actually measured laser ray and the laser ray corresponding to the coarse estimation attitude; carrying out attitude optimization by taking the minimum distance between the two as a target to obtain an optimized estimated attitude; and S4, obtaining estimated position coordinates of the workpiece according to the camera pose and the actually measured contour. According to the method, the workpiece pose is estimated based on cooperation of the workpiece contour and the laser rays projected on the workpiece surface, the method does not depend on object surface textures, the occupied space of a database is small, the matching speed is high, and the estimation accuracy is high.

Description

Technical Field

[0001] The present invention belongs to the technical field related to machine vision pose estimation, and more specifically, relates to a workpiece pose estimation method and system based on an outer contour and a laser line. Background Art

[0002] In the industrial field, scenarios where the pose of a target needs to be measured often occur to facilitate controlling a robotic arm to grasp parts. Industrial parts have the characteristic of weak texture and less surface information, resulting in easy failure of traditional feature descriptors to extract features and causing difficulties in pose estimation.

[0003] Existing technologies are mainly divided into two categories. One category is to extract the point cloud on the surface of an object and register it with a template point cloud through algorithms such as ICP to achieve pose estimation. This method is restricted by the quality of point cloud extraction and often has a slow speed. The other category of methods is through deep learning. By taking RGB-D images or point cloud images of the actual scene and directly or indirectly estimating the pose of the part by training a neural network model. This method requires a large amount of data for training, has a high implementation cost and poor versatility. Summary of the Invention

[0004] In view of the above defects or improvement requirements of the existing technology, the present invention provides a workpiece pose estimation method and system based on an outer contour and a laser line, which is used to solve the problem of difficult pose estimation of existing industrial parts in processing and production.

[0005] To achieve the above object, according to one aspect of the present invention, a workpiece pose estimation method based on an outer contour and a laser line is provided, including:

[0006] Offline database establishment stage:

[0007] S1: Construct a sampling system including a camera and a laser through simulation or experimental means. At each pose, use the laser to scan the workpiece and obtain the workpiece image from the camera's perspective, extract the contour sample and the laser line sample, and establish a secondary database corresponding to the pose, contour, and laser line respectively;

[0008] Online estimation stage:

[0009] S2: Scan the workpiece with the laser, obtain the workpiece image through the camera and extract the measured contour and the measured laser line of the workpiece, and match them with the samples in the secondary database to obtain a rough estimated pose;

[0010] S3: Based on the camera pose, the workpiece image, and the laser scanning plane, respectively obtain the spatial coordinates of the measured laser line and the laser line corresponding to the roughly estimated pose; take the minimum distance between the measured laser line and the laser line corresponding to the roughly estimated pose as the objective, optimize the roughly estimated pose, and obtain the optimized estimated pose;

[0011] S4: Based on the camera pose and the measured contour, obtain the estimated position coordinates of the workpiece.

[0012] According to the workpiece pose estimation method based on the outer contour and the laser line provided by the present invention, the offline database establishment stage and the online estimation stage respectively further include: establishing a part coordinate system and a camera coordinate system, where the Z-axis of the part coordinate system coincides with the optical axis of the camera; representing the pose of the workpiece using the Euler angles in the order of ZXZ axes of the part coordinate system;

[0013] In S1, the sampling system is used to sample the information of the workpiece under different poses, specifically including:

[0014] Determine the distribution of the sampling viewpoints using the spherical Fibonacci sampling method, where the first two Euler angle components of the pose can be determined according to the sampling viewpoints;

[0015] At each sampling viewpoint, rotate the workpiece around the Z-axis of the part coordinate system by one week at a preset angular step to perform information sampling, and obtain multiple poses with different third Euler angle components at each sampling viewpoint.

[0016] According to the workpiece pose estimation method based on the outer contour and the laser line provided by the present invention, in S1, the laser is used to scan the workpiece at each pose and the workpiece image is obtained from the camera's perspective. According to the workpiece image, the contour sample and the laser line sample of the workpiece are extracted, specifically including:

[0017] Extract the workpiece contour corresponding to each pose according to the workpiece image;

[0018] At each pose, move the laser so that different positions of the workpiece are scanned by the laser multiple times, respectively obtain the laser scanning images and extract the laser lines, and obtain multiple laser lines corresponding to different positions of the workpiece;

[0019] Perform normalization processing on the workpiece contour and the laser line;

[0020] For the workpiece contours corresponding to multiple poses with different third Euler angle components at each sampling viewpoint, select one of them as the representative contour, extract the feature descriptor from the normalized representative contour to form a contour sample, and each sampling viewpoint corresponds to one contour sample; extract the feature descriptor from the normalized laser line to form a laser line sample, and each pose corresponds to multiple laser line samples.

[0021] According to the workpiece pose estimation method based on the outer contour and the laser line provided by the present invention, the contour sample includes the Fourier descriptor of the contour, and the laser line sample includes the normal vector feature of the laser line; specifically forming the laser line sample includes:

[0022] For any laser line, segment it according to continuity. For each segment of the laser line, downsample to obtain representative points, and fit the normal vector with the representative points within the preset neighborhood range of any representative point. The set of normal vectors corresponding to all representative points on any laser line forms the laser line sample corresponding to the any laser line.

[0023] According to the workpiece pose estimation method based on the outer contour and the laser line provided by the present invention, S2 specifically includes:

[0024] Normalize the measured contour and the measured laser line, and then extract the feature descriptors respectively;

[0025] Match the measured contour with the contour sample through the feature descriptors, and determine k contour samples with smaller distances as k candidate representative contours according to the Euclidean distance between the feature descriptors;

[0026] According to the angular deviation in the circumferential direction between the measured contour and the candidate representative contour, perform the matching of the third Euler angle component among the multiple poses corresponding to each candidate representative contour to determine the candidate poses, and a total of k candidate poses are determined;

[0027] Match the measured laser line with the laser line samples corresponding to the k candidate poses through the feature descriptors to obtain the rough estimated pose.

[0028] According to the workpiece pose estimation method based on the outer contour and the laser line provided by the present invention, establishing the secondary databases respectively corresponding to the pose, the contour, and the laser line further includes:

[0029] For any pose, extract the circumscribed circle of the normalized workpiece contour, calculate the centroid of the normalized laser line, and record the ratio rela of the distance from the centroid of any laser line to the center of the circumscribed circle to the radius of the circumscribed circle in the secondary database data ;

[0030] Match the measured laser line with the laser line samples corresponding to the k candidate poses through the feature descriptors to obtain the rough estimated pose, specifically including:

[0031] Extract the circumscribed circle of the normalized measured contour, calculate the centroid of the normalized measured laser line, and obtain the ratio rela of the distance from the centroid of the measured laser line to the center of the circumscribed circle of the measured contour to the radius of the circumscribed circle real ;

[0032] Let relareal the rela corresponding to all laser lines in k candidate poses data Compare them, and select the laser lines with the difference less than the preset threshold as candidate laser lines;

[0033] Calculate the similarity between the candidate laser lines and the measured laser lines, and select the candidate laser line with the highest similarity as the matching laser line, and the corresponding pose as the rough estimated pose;

[0034] Among them, the laser line distance is used to compare the similarity. Specifically:

[0035] For each first representative point obtained by downsampling the measured laser line, place it at the same pixel point of the image where the candidate laser line is located and search for the second representative point obtained by downsampling the candidate laser line within the preset neighborhood range, and calculate the deviation d between the representative points point , the deviation d point The calculation is as follows:

[0036] If a unique second representative point is searched, calculate the angle between the normal vectors corresponding to the first representative point and the second representative point and the Euclidean distance between the two points, and sum them with weights as the deviation d point ;

[0037] If multiple second representative points are searched, calculate the deviation d between the first representative point and each second representative point point , and retain the smallest deviation d point ;

[0038] If no second representative point is searched, take a relatively large constant as the deviation d point ;

[0039] Sum the deviations d calculated for all the first representative points point as the distance, i.e., the similarity, between the measured laser line and the candidate laser line.

[0040] According to the workpiece pose estimation method based on the outer contour and the laser line provided by the present invention, in S3, according to the camera pose, the workpiece image, and the laser scanning plane, the spatial coordinates of the measured laser line and the laser line corresponding to the rough estimated pose are respectively obtained, specifically including:

[0041] Obtain the camera pose through calibration, and determine the spatial coordinates of the camera optical center according to the camera pose;

[0042] For any point on the measured laser line, obtain the corresponding spatial coordinates of this point on the camera imaging plane according to the pixel coordinates and the camera pose, establish a ray equation passing through the camera optical center and this point on the camera imaging plane, and obtain the spatial coordinates of the measured laser line by simultaneously solving the ray equation corresponding to the point on the measured laser line and the laser scanning plane corresponding to the measured laser line;

[0043] By simultaneously solving the ray equation corresponding to the point on the laser line in the rough estimated pose and the laser scanning plane corresponding to the laser line in the rough estimated pose in the secondary database, obtain the spatial coordinates of the laser line corresponding to the rough estimated pose.

[0044] According to the workpiece pose estimation method based on the outer contour and the laser line provided by the present invention, in S3, taking the minimum distance between the measured laser line and the laser line corresponding to the rough estimated pose as the target, optimize the rough estimated pose, and obtain the optimized estimated pose, which specifically includes:

[0045] Interpolate according to the spatial coordinates of multiple laser lines corresponding to the rough estimated pose to obtain a spatial point cloud including these multiple laser lines;

[0046] Set a compensation rotation matrix and a translation vector, and take the minimum sum of the distances between all points on the measured laser line after being corrected by the compensation rotation matrix and the translation vector and the spatial point cloud as the goal to obtain the optimized compensation rotation matrix;

[0047] Use the optimized compensation rotation matrix to correct the rough estimated pose to obtain the optimized estimated pose.

[0048] According to the workpiece pose estimation method based on the outer contour and the laser line provided by the present invention, S4 specifically includes:

[0049] Extract the center of the circumscribed circle in the contour image corresponding to the measured contour;

[0050] According to the camera pose and the pixel coordinates of the center of the circumscribed circle of the measured contour, obtain the spatial coordinates of the center of the circumscribed circle of the measured contour on the camera imaging plane, and establish a ray equation passing through the camera optical center and the center of the circumscribed circle of the measured contour on the camera imaging plane;

[0051] Obtain the intersection point by solving the ray equation from the camera optical center to the center of the circumscribed circle of the measured contour on the camera imaging plane and the actual laser scanning plane, and obtain the spatial coordinates of a point on the workpiece as the position coordinates for workpiece estimation.

[0052] According to another aspect of the present invention, there is provided a workpiece pose estimation system based on an outer contour and a laser line. The system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it executes the workpiece pose estimation method based on an outer contour and a laser line as described in any one of the above.

[0053] Generally speaking, compared with the prior art by the above technical solutions conceived by the present invention, the workpiece pose estimation method and system based on an outer contour and a laser line provided by the present invention:

[0054] 1. It is proposed to estimate the workpiece pose by collaborating the workpiece contour and the laser line projected on the workpiece surface. The pose estimation can be more accurately realized through two judgment criteria; specifically, the laser line on the three-dimensional contour surface of the object is used as a feature, which has high brightness, good monochromaticity, is less affected by ambient light, has a small beam divergence, and the concentration of measurement points is good, which is beneficial to improving the estimation accuracy; in addition, it is proposed to construct a two-level database of pose, contour, and laser line, which improves the retrieval efficiency and saves storage space; it is also proposed to optimize and obtain the optimized pose on the basis of matching the rough estimated pose using the secondary database, which further improves the pose estimation accuracy; the method of the present invention does not depend on the surface texture of the object, and the database occupies less space, has a fast matching speed, high estimation accuracy, and low hardware requirements;

[0055] 2. The specific pose sampling method when constructing the secondary database is proposed, that is, first determine the sampling view point direction and then perform rotational sampling under each sampling view point, which is beneficial to improving the sampling uniformity and sampling efficiency; the construction method of the contour sample and the laser line sample in the secondary database is proposed. Specifically, first normalize the contour and the laser line, and use a scale-invariant descriptor to describe the contour and the laser line, which improves the accuracy at different scales;

[0056] 3. A contour and laser line matching method based on the secondary database is proposed. Specifically, first perform the matching of contour similarity (using the Euclidean distance of the contour feature descriptor), then perform the matching of the third Euler angle component corresponding to the contour, then perform the screening of the laser line position and the matching of similarity, and finally obtain the rough estimated pose, which is beneficial to improving the matching efficiency and matching accuracy;

[0057] 4. A method of converting the laser line into a space curve and then optimizing the pose using the position of the space curve is proposed, which can make the laser line corresponding to the optimized pose closer to the measured laser line, thereby further improving the pose estimation accuracy; in addition, a method of estimating the workpiece position coordinates using the center of the circumcircle of the measured contour is proposed, which can obtain the estimated position of the workpiece and realize the determination of the pose. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1It is the overall flowchart of the workpiece pose estimation method based on the outer contour and the laser line provided by the present invention;

[0059] Figure 2 It is a schematic diagram of an example of the installation positions of the camera and the laser provided by the present invention;

[0060] Figure 3 It is a schematic diagram of the distribution of the end points of the normal vectors representing the sampling view point directions on the spherical surface provided by the present invention;

[0061] Figure 4 It is the overall flowchart of establishing the secondary database provided by the present invention;

[0062] Figure 5 It is the flowchart of obtaining the rough estimated pose through rough matching provided by the present invention;

[0063] Figure 6 It is the flowchart of the fine matching provided by the present invention. Specific embodiments

[0064] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0065] Please refer to Figure 1 , this embodiment provides a workpiece pose estimation method based on the outer contour and the laser line. The workpiece pose estimation method includes:

[0066] Offline database establishment stage:

[0067] S1: Construct a sampling system including a camera and a laser through simulation or experimental means. Use the sampling system to sample information of the workpiece in different poses. At each pose, use the laser to scan the workpiece and obtain the workpiece image from the camera's perspective. According to the workpiece image, obtain the contour sample and the laser line sample of the workpiece, and establish a secondary database corresponding to the pose, contour, and laser line respectively;

[0068] Online estimation stage:

[0069] S2: Scan the workpiece through the laser, and obtain the workpiece image through the camera; obtain the measured contour and the measured laser line of the workpiece according to the workpiece image, match the measured contour and the measured laser line with the samples in the secondary database, and roughly screen out the rough estimated pose;

[0070] S3: According to the camera pose, the workpiece image, and the laser scanning plane, respectively obtain the spatial coordinates of the measured laser line and the laser line corresponding to the roughly estimated pose; take the minimum distance between the measured laser line and the laser line corresponding to the roughly estimated pose as the objective, and use the roughly estimated pose as the initial pose to perform pose optimization to obtain the optimized estimated pose.

[0071] S4: According to the camera pose and the measured contour, obtain the estimated position coordinates of the workpiece.

[0072] In some specific embodiments, as Figure 2 shown, the offline database establishment stage and the online estimation stage respectively further include: establishing a part coordinate system and a camera coordinate system, where the Z-axis of the part coordinate system coincides with the camera optical axis, the X-axis is perpendicular to the shown plane and points outwards, and the Y-axis is determined according to the right-hand system; use the Euler angles in the order of ZXZ axes of the part coordinate system to represent the pose of the workpiece.

[0073] Furthermore, in S1, the relative position between the camera and the laser can be preset, including the angle between the camera optical axis and the laser light plane and the initial position of the laser. In this example, the laser can be made to emit laser vertically downward, and the camera optical axis forms a 45-degree angle with the laser plane. Establish a secondary database between the pose, the contour, and the laser line according to the known part model. The database can be generated by simulation software or obtained through experiments. In this embodiment, SOLIDWORKS software is used for modeling and simulation, and the fixed-axis Euler angles in the order of ZXZ of the part coordinate system are used to represent the rotation. Since the third rotation axis coincides with the camera optical axis, determining the first two Euler angles can determine the shape of the contour. For the uniqueness of pose representation, the ranges of the three rotations are [360°, 180°, 360°] respectively.

[0074] After the part, that is, the workpiece, undergoes the first two rotations, the viewpoints of the camera are located at different positions in the coordinate system fixed to the part. In S1, the sampling system is used to sample the information of the workpiece in different poses, specifically including:

[0075] Use the spherical Fibonacci sampling method to determine the distribution of the sampling viewpoints. Among them, the first two Euler angle components of the pose can be determined according to the sampling viewpoints;

[0076] At each sampling viewpoint, make the workpiece rotate one week around the Z-axis of the part coordinate system according to the preset angular step to perform information sampling, and obtain multiple poses with different third Euler angle components at each sampling viewpoint.

[0077] The sampling viewpoint is the workpiece image acquisition point, i.e., the point where the camera is located. Specifically, the spherical Fibonacci sampling method is used to determine the distribution of the intersections of the sampling viewpoint directions with the unit sphere, and then the sampling viewpoints are determined; the sampling viewpoint direction is the direction of the line connecting the sampling viewpoint and the origin of the part coordinate system; the distribution of the sampling viewpoint directions is shown as follows:

[0078]

[0079] where N fibo is the total number of sampling viewpoints, n = 0, 1, 2, …, N fibo - 1, is a constant, and the obtained x n , y n , z n are the endpoints of the normal vector representing the sampling viewpoint direction; this sampling method can make the given number of sampling points N fibo achieve an approximately uniform distribution. The distribution of the normal vector endpoints on the sphere is as Figure 3 shown.

[0080] Since the range of the second parameter of the Euler angles is [0°, 180°], in order to align the sampling viewpoints on the part with the camera optical axis, first rotate around the X-axis of the part coordinate system so that the sampling viewpoints are rotated into the positive half-plane corresponding to the Y-axis in the YOZ plane; then rotate around the X-axis of the part coordinate system so that the sampling points fall on the Z-axis of the part coordinate system. In this way, the correspondence between each sampling viewpoint and the first two parameters of the Euler angles is:

[0081]

[0082] The Euler angle of rotation around the X-axis of the part coordinate system determined according to the sampling viewpoints is:

[0083]

[0084] Since the range of atan2 is (-π, π], the calculated range of θ Z is Therefore, judge θ Z , if it is less than 0, add 2π to map the range to [0, 2π). Similarly, map the range of θ X to [0°, 180°).

[0085] In this example, considering the size of the database, take N fibo = 200.

[0086] The third parameter of the Euler angles is the angle of rotation around the camera optical axis (Z-axis). In this example, the step size is taken as 18°, and the total number of sampled postures is 200×(360° / 18°) = 4000.

[0087] After determining the rotational posture, for each posture, the contour and the laser line are extracted. In this example, the rotational feature is added according to the Euler angle rotation sequence using the rotational function in SOLIDWORKS software, and then saved as an image to extract the contour. For each posture, in this example, the scanning of the part by the laser line structure of the laser is simulated by translating the reference plane in SOLIDWORKS software to intersect with the part. For each posture, the laser is moved in the direction perpendicular to the light plane, and the part is scanned multiple times with the laser plane, and the reference plane simulating the laser plane in the simulation software intersects with the part multiple times. There are some positions in the surface intersection line that will not be irradiated by the laser. In this example, each point on the intersection line is connected to the light source point, and it is judged whether there are other intersection points on the path. If so, the point is deleted, and finally the unobstructed part is retained.

[0088] Further, in S1, at each posture, the laser is used to scan the workpiece and the workpiece image is obtained from the camera view. The contour sample and the laser line sample of the workpiece are obtained according to the workpiece image, specifically including:

[0089] Extract the workpiece contour corresponding to each posture according to the workpiece image;

[0090] At each posture, the laser is moved so that different positions of the workpiece are scanned by the laser multiple times. The laser scan images are obtained respectively and the laser lines are extracted, and multiple laser lines corresponding to different positions of the workpiece are obtained;

[0091] Normalize the workpiece contour and the laser line;

[0092] For the workpiece contours corresponding to multiple postures with different third Euler angle components at each sampling viewpoint, one of them is selected as the representative contour. The feature descriptor is extracted from the normalized representative contour to form a contour sample, and each sampling viewpoint corresponds to one contour sample for multiple postures; the feature descriptor is extracted from the normalized laser line to form a laser line sample, and each posture corresponds to multiple laser line samples.

[0093] To eliminate the influence of size and position on the feature extraction of the contour and the laser line, the contours and laser lines in the database need to be normalized before calculating the features. In this example, the center of the circumscribed circle of the contour is used as the center for normalization and alignment. The advantage of doing this is that it is less affected by noise and ensures that the contour and the laser line are always within the circumscribed circle range for central rotation. Define the size of the normalized drawing as h normalize ×w normalize , h normalize and w normalize are the height and width of the normalized drawing respectively. In this example, 2550 and 3300 are taken respectively. For the workpiece contour corresponding to each posture, the center of the circumscribed circle (x c , y c ) and the radius R of the circumscribed circle in the pixel coordinate system are extracted c。

[0094] Normalize the workpiece contour and the laser line extracted for each pose, specifically including:

[0095] Perform normalization processing of translation and scaling on the workpiece contour and the laser line. The normalization translation amount (p x , p y ) is:

[0096] (p x , p y ) = (v0 - x c , u0 - y c );

[0097] Among them, (u0, v0) is the midpoint of the image, that is, the midpoint of the pixel coordinate system. (x c , y c ) is the coordinate of the center of the circumscribed circle of the workpiece contour in the pixel coordinate system. The scaling factor scale is:

[0098]

[0099] Among them, the size of the image is h normalize ×w normalize ; R c is the radius of the circumscribed circle of the workpiece contour in the pixel coordinate system. Translate and scale the workpiece contour corresponding to each pose with multiple laser lines. The translation amounts of the contour and the laser line are the same, and the normalized contour and laser line can be obtained.

[0100] After obtaining the normalized contour and laser line, calculate their feature descriptions. For multiple workpiece contours of multiple poses corresponding to one sampling viewpoint, one can be selected as the representative contour, and the feature descriptor of the representative contour is calculated and stored in advance; for the laser line, the feature descriptor needs to be calculated for each laser line. Since the contour shape is determined by the first two components of the Euler angle, in this example, before extracting features for each pose, the first two components are searched in the extracted table. If found, it means that the corresponding contour has been extracted with features, so the current pose is skipped; if not found, it means that the contour corresponding to this pose has not been extracted, then the contour features of the current pose are extracted, and after extraction, the first two components are recorded in the extracted table to indicate that the current contour has been extracted with features.

[0101] In this example, Fourier descriptors are used to describe the contour features, and normal vector features are used to describe the laser lines. That is, the contour samples include the Fourier descriptors of the contour, and the laser line samples include the normal vector features of the laser lines. For the extraction of contour features, in this example, the contour points are first downsampled, and each point is regarded as a point in the complex coordinate system with the coordinate x + yi, where i is the imaginary unit. Calculate its FFT, take the modulus, normalize it, and take the first 30 terms as the contour description, denoted as fd, and delete the DC component of the first term. Such descriptors have the invariance of translation, rotation, and scaling.

[0102] For the extraction of laser line features, that is, forming the laser line samples specifically includes:

[0103] For any laser line, segment it according to continuity. For each segment of the laser line, downsample to obtain representative points, and fit the normal vector with the representative points within the preset neighborhood range of any representative point. The set of normal vectors corresponding to all representative points on any laser line forms the laser line sample corresponding to the any laser line, and record this laser line sample in the database. In this example, it is specified that the direction of the normal vector is downward, which avoids the error caused by the direction representation. After that, for each pose, calculate the ratio rela of the distance from the centroid of each laser line to the center of the circumcircle of the workpiece contour to the radius of the circumcircle of the contour data , which represents the relative position between the laser line and the contour, and the value range is 0 - 1.

[0104] Establish and store the mapping relationship between the pose and various descriptors. In this example, a hash table is used to establish the mapping relationship. The overall process of establishing the secondary database is as Figure 4 shown.

[0105] Furthermore, S2 specifically includes:

[0106] As Figure 2 shown, build a pose estimation device. In this embodiment, the laser and the camera are installed on the same crossbar. The laser emits laser vertically downward, the optical axis of the camera forms a 45-degree angle with the laser plane, and the laser can be moved horizontally through the displacement stage. Scanning the part once with the laser leaves a laser line on the surface of the part, and take pictures of the part image and the laser line image and extract the measured contour and the measured laser line.

[0107] In this example, the contour and the laser line are extracted by the threshold method. First, convert the contour image to a grayscale image, then set a suitable light intensity threshold to extract the target part and extract the binary image representing the contour area. The laser used emits red light. By extracting the red channel and setting the light intensity threshold, filter out the laser lines exceeding the light intensity threshold, and then perform an AND operation with the binary image of the contour area to extract the laser lines inside the contour. Since the extracted laser lines have a certain width, the measured laser line can be thinned by the gray centroid method, and then calculate the centroid of the thinned measured laser line. The centroid of the laser line is the average value of the positions of all points on the laser line.

[0108] Then, the measured profile and the measured laser line need to be normalized and position-aligned with the database so that the profile is located at the same position in the drawing sheet. After that, feature descriptors are extracted respectively.

[0109] First, similar profiles are matched. In this example, the measured profile is matched with the profile samples through feature descriptors, and k profile samples with smaller distances are determined as k candidate representative profiles according to the Euclidean distance between the feature descriptors; specifically, the Euclidean distance is calculated between the descriptors extracted from the measured profile and the profile descriptors in the database, d bdr = ||fd r - fd i ||2, where fd r is the Fourier descriptor of the measured profile, and fd i is the descriptor of the profile sample. After calculating the distances, the calculation results are sorted in ascending order, and the first k positions are selected as the matching results. For example, in this example, k = 10 is taken.

[0110] Then, according to the angular deviation in the circumferential direction between the measured profile and the candidate representative profiles, the matching of the third Euler angle component is performed among multiple poses corresponding to each candidate representative profile to determine the candidate poses. One candidate pose is matched for each candidate representative profile, and a total of k candidate poses are determined;

[0111] That is, the third rotation angle, namely the third Euler angle component, is determined for the selected profiles. In this example, the angular deviation in the circumferential direction between the measured profile and the candidate representative profiles is calculated by using an improved shape context algorithm, which specifically includes:

[0112] Taking the center of the circumscribed circle of the profile as the center, the circle is divided into m regions. In each region, the average distance from the profile points to the center of the circumscribed circle of the profile is calculated as a feature parameter, and after normalization, a feature vector with a length of m is obtained. The feature vector of the measured profile is denoted as g r (m), and the feature vector of the candidate representative profile in the database is g j (m). For each candidate representative profile, g r (m) is cyclically shifted by 0 to m - 1 positions respectively, and the Euclidean distance is calculated with the feature vector of the candidate representative profile;

[0113] According to the cyclic shift amount n cir with the minimum Euclidean distance, the angular deviation in the circumferential direction between the measured profile and the candidate representative profile is determined, which is 360·n cir / m (°). In the pose corresponding to the candidate representative contour, use this angular deviation to correct the third Euler angle component, and select the pose with the closest third Euler angle component among multiple poses corresponding to the candidate representative contour as the candidate pose according to the corrected third Euler angle component. For example, in this instance, m = 8 is taken.

[0114] Specifically, when sampling different poses for constructing the secondary database, starting from the starting position at each sampling viewpoint, rotate counterclockwise and stop sampling according to a preset angular step size. Each stop, that is, for each pose, record the rotation amount relative to the initial position as the starting rotation amount corresponding to each pose. When using the improved shape context algorithm to match the third Euler angle component under the candidate representative contour, shift g r (m) cyclically to the right by 0 to m - 1 bits respectively, corresponding to the measured contour rotating counterclockwise by [0°~(m - 1)*360 / m (°)]. Calculate the Euclidean distance with the feature vector of the candidate representative contour in the database, and the displacement amount n cir corresponding to the minimum Euclidean distance, and the third rotation angle is 360·n cir / m (°), then the third rotation amount of the determined candidate pose is the starting rotation amount of the pose corresponding to the candidate representative contour minus [(360·n cir / m) / 18°]*18°, [] is the integer operation.

[0115] Match the measured laser line with the laser line samples corresponding to k candidate poses through a feature descriptor to obtain a rough estimated pose.

[0116] Furthermore, establishing a secondary database corresponding to poses, contours, and laser lines respectively further includes:

[0117] For any pose, extract the circumscribed circle of the normalized workpiece contour, calculate the centroid of the normalized laser line, and record the ratio rela of the distance from the centroid of any laser line to the center of the circumscribed circle to the radius of the circumscribed circle in the secondary database data .

[0118] Match the measured laser line with the laser line samples corresponding to k candidate poses through a feature descriptor to obtain a rough estimated pose, specifically including: for each candidate pose, select the laser line in the database with a relative position close to the actual laser line, and calculate the normal vector similarity with the measured laser line. Specifically:

[0119] Extract the circumscribed circle of the normalized measured contour, calculate the centroid of the normalized measured laser line, and obtain the ratio rela of the distance from the centroid of the measured laser line to the center of the circumscribed circle of the measured contour to the radius of the circumscribed circle real ;

[0120] Let relareal The rela corresponding to all laser lines in k candidate poses data are compared, and the laser lines with a difference less than a preset threshold are selected as candidate laser lines; if the difference is greater than a certain threshold, this laser line is discarded;

[0121] Calculate the similarity between the candidate laser line and the measured laser line, and select the candidate laser line with the highest similarity as the matching laser line, and the corresponding pose as the rough estimated pose;

[0122] Among them, the laser line distance is used to compare the similarity. Specifically:

[0123] For each first representative point obtained by downsampling the measured laser line, place it at the same pixel point of the candidate laser line image and search for the second representative point obtained by downsampling the candidate laser line within a preset neighborhood range, and calculate the deviation d between the representative points point , the deviation d point The calculation is as follows:

[0124] If a unique second representative point is searched, calculate the angle between the normal vectors corresponding to the first representative point and the second representative point and the Euclidean distance between the two points, and sum them with weights as the deviation d point ;

[0125] If multiple second representative points are searched, calculate the deviation d between the first representative point and each second representative point point , and retain the minimum deviation d point ;

[0126] If no second representative point is searched, it is considered that the two laser lines are not similar at this position, and a relatively large constant is taken as the deviation d point ;

[0127] Sum the deviations d calculated for all the first representative points point as the distance, i.e., the similarity, between the measured laser line and the candidate laser line.

[0128] In this example, the formula for the laser line distance d laser is:

[0129]

[0130] Among them, n p is the number of representative points on the measured laser line; θ normal , d normal are the normalization parameters for the normal vector angle difference and the distance in the pixel coordinate system. In this example, θ normal =π, d normal=500; α and β are weights, which are 500 and 300 respectively in this example; NOT_MATCH_DIST represents the constant when no representative point is matched, which is 10 in this example. 3 The similarity is the inverse of the distance, and the pose corresponding to the candidate laser line with the highest similarity is the rough estimated pose. n is the normal vector angle of the nth representative point of the measured laser line obtained when calculating the feature descriptor, p n is the coordinate of the nth representative point on the measured laser line, and ||v|| represents the second norm of vector v. θ x Represents the normal vector angle of the only second representative point, q x Represents the coordinates of the only second representative point. θ i When there are multiple second representative points in the neighborhood, the normal vector angle of the i-th second representative point, q i Indicates the coordinates of the ith second representative point when there are multiple second representative points in the neighborhood. When calculating the similarity between the measured laser line and the candidate laser line, it is based on the normalized measured laser line and the candidate laser line. The process of coarse matching to obtain the coarse estimated posture is as follows: Figure 5 shown.

[0131] Further, in S3, the spatial coordinates of the measured laser line and the laser line corresponding to the roughly estimated posture are respectively obtained according to the camera posture, the workpiece image and the laser scanning plane, which specifically includes:

[0132] The camera pose is obtained through calibration, and the spatial coordinates of the camera optical center are determined according to the camera pose;

[0133] For any point on the measured laser line, the spatial coordinates corresponding to the point on the camera imaging plane are obtained according to the pixel coordinates and the camera posture, a ray equation passing through the camera optical center and the point on the camera imaging plane is established, and the spatial coordinates of the measured laser line are obtained by combining the ray equation corresponding to the point on the measured laser line with the laser scanning plane corresponding to the measured laser line;

[0134] For any point on the laser line under the rough estimated posture, the spatial coordinates corresponding to the point on the camera imaging plane are obtained according to its pixel coordinates and the camera posture, and a ray equation passing through the optical center of the camera and the point on the camera imaging plane is established. The spatial coordinates of the laser line corresponding to the rough estimated posture are obtained by combining the ray equation corresponding to the point on the laser line under the rough estimated posture in the secondary database with the laser scanning plane corresponding to the laser line under the rough estimated posture.

[0135] Further, in S3, taking the minimum distance between the measured laser line and the laser line corresponding to the roughly estimated posture as the goal and taking the roughly estimated posture as the initial posture to perform posture optimization, and obtaining the optimized estimated posture, specifically includes:

[0136] Interpolate the spatial coordinates of multiple laser lines corresponding to the roughly estimated pose to obtain a spatial point cloud including these multiple laser lines; the spatial point cloud is a surface point cloud obtained based on the spatial coordinate differences of multiple laser lines corresponding to the roughly estimated pose, that is, a spatial surface point cloud including multiple laser lines corresponding to the roughly estimated pose.

[0137] Set a compensation rotation matrix and a translation vector, and aim to minimize the sum of the distances between all points on the measured laser line after being corrected by the compensation rotation matrix and the translation vector and the spatial point cloud, so as to obtain an optimized compensation rotation matrix.

[0138] Use the optimized compensation rotation matrix to correct the roughly estimated pose to obtain an optimized estimated pose.

[0139] Furthermore, when calculating the spatial coordinates of the measured laser line and the laser line corresponding to the roughly estimated pose, first determine the position of the light plane. When constructing the secondary database, according to the specific position of the established part coordinate system and the known initial position of the laser, the initial light plane can be determined. When performing multiple laser scans at each pose, the light plane at each scan can be determined according to the movement amount of the laser relative to the initial position. During the online estimation stage, the actual laser scan plane can also be determined according to the movement amount of the laser relative to the initial position during actual scanning.

[0140] For example, in some specific examples, assume that the initial light plane passes through the origin of the part coordinate system. According to the point-normal form, the equation of the initial light plane in the part coordinate system is:

[0141] y - z = 0

[0142] Assume that during actual pose estimation, the displacement of the laser along the horizontal direction is SL, with the direction towards the camera being positive. After translation, the light plane passes through the point Then the equation of the light plane in the part coordinate system after translation is:

[0143]

[0144] In this example, the pixel size of the used camera is dx = dy = 2.2 μm, and the number of pixels is 2592×1944. The parametric ray equation of the ray passing through the optical center and a point on the camera imaging plane in the part coordinate system is:

[0145]

[0146] where t is a parameter, with a range of t > 0, x C0 , y C0 , z C0 is the camera optical center coordinate in the part coordinate system, Let \((u, v)\) be the coordinates of a point on the camera imaging plane in the part coordinate system, and \((x, y, z)\) be the coordinates of a point on the ray in the part coordinate system. The coordinates in the part coordinate system are the spatial coordinates.

[0147] To determine the coordinates of the camera optical center in the part coordinate system, in this example, a calibration board is fixed on the platform, and a coordinate system is fixedly connected to the lower left corner point of the calibration board. The x-axis and y-axis coincide with the length and width of the calibration board, and the z-axis is perpendicular to the calibration board and points upward. The translation amount of the calibration board coordinate system relative to the part coordinate system is The rotation amount is:

[0148]

[0149] By taking a series of images of the calibration board in different poses with the camera, the internal and external parameters of the camera can be obtained. For the installed calibration board, the external parameter is Then the pose of the camera coordinate system in the part coordinate system is:

[0150]

[0151] Among them,

[0152]

[0153] From the calibrated external parameters, we can obtain and Then:

[0154]

[0155] In the above formulas, \(t\) is the translation amount, \(R\) is the rotation amount, \(T\) represents the homogeneous transformation matrix, Part represents the part, Board represents the calibration board, and Camera represents the camera.

[0156] The coordinates of the pixel point with pixel coordinates \((u, v)\) in the camera coordinate system are \(((u - u_0)dx, (v - v_0)dy, f)\), where \(u_0, v_0\) are the coordinates of the principal point of the image, \(f\) is the image distance, simplified by the focal length, and \(dx, dy\) are the pixel sizes in the x and y directions. Convert this coordinate to homogeneous form and left-multiply by to obtain the representation of the point on the camera imaging plane in the part coordinate system. The translation component of

[0157] By simultaneously solving the optical plane equation and the ray equation, the coordinates of the spatial point set of the measured laser line in the part coordinate system and the coordinates of the spatial point set of the laser line corresponding to the rough estimated pose in the database can be obtained respectively. In this example, the pixel coordinates used when calculating the spatial coordinates of the measured laser line and the laser line corresponding to the rough estimated pose are calculated based on the coordinates before normalization.

[0158] After obtaining the spatial coordinates, the measured laser line and the laser lines corresponding to the rough estimated poses are respectively converted into spatial curves. Let the spatial curve corresponding to the measured laser line be L, and the spatial curves of the laser lines corresponding to the rough estimated poses in the database be C1, C2, …, Cn. The surface point cloud P is generated by interpolation. That is, the surface point cloud P can be obtained by interpolating the point sets of multiple laser lines corresponding to the rough estimated poses in the database. Through an optimization method, with the distance from L to P as the optimization function, the refined estimated pose is determined.

[0159] In this example, by adopting the optimization method of sqp, the optimization objective function s is:

[0160]

[0161] Among them, L(i) represents the coordinates of the i-th point in the measured laser line in the part coordinate system, ΔR(θ1, θ2, θ3) represents the compensation rotation matrix determined by θ1, θ2, θ3, and t(t1, t2, t3) represents the translation vector. Let p be a point on the measured laser line, and dist(p, P) represents the distance from point p to the point cloud P, which can be represented by the minimum value among the distances from point p to each point in the point cloud P. By optimizing to minimize s, the compensation rotation matrix ΔR and the translation vector t when s takes the minimum value can be obtained. Based on the rough estimated pose R0, the refined estimated pose is R f = R0ΔR. The pose optimization, that is, the refined matching process, is as Figure 6 shown.

[0162] Furthermore, S4 specifically includes:

[0163] Extract the center of the circumscribed circle in the contour image corresponding to the measured contour;

[0164] According to the camera pose and the pixel coordinates of the center of the circumscribed circle of the measured contour, obtain the spatial coordinates of the center of the circumscribed circle of the measured contour on the camera imaging plane, and establish a ray equation passing through the camera optical center and the center of the circumscribed circle of the measured contour on the camera imaging plane;

[0165] Find the intersection point of the ray equation from the camera optical center to the center of the circumscribed circle of the measured contour on the camera imaging plane and the actual laser scanning plane, and obtain the spatial coordinates of a point on the workpiece as the estimated position coordinates of the workpiece.

[0166] In this embodiment, the position of the part is estimated by actually measuring the spatial position of the laser line. First, the center within the light plane is determined. In this example, the center of the light plane is represented by the center of the circumscribed circle of the contour.

[0167] First, perform circumscribed circle fitting on the actually measured contour image to determine the coordinates of the center of the circumscribed circle in the pixel coordinate system. For the above method of obtaining spatial coordinates, convert the coordinates of the center of the circumscribed circle into the coordinates in the part coordinate system, and find the intersection point by the ray equation from the camera optical center to the center of the circumscribed circle and the actual light plane equation, so as to obtain the estimated center coordinates (x l, y l , z l ) in the light plane in the part coordinate system.

[0168] Specifically, in this embodiment, the actual laser scanning plane can be taken as the plane passing through the middle position in the horizontal direction of the workpiece, so that the obtained position coordinates are closer to the geometric center of the workpiece. For example, when the origin of the part coordinate system is set at the middle position in the horizontal direction of the workpiece, the actual laser scanning plane can be taken as the vertical plane passing through the origin.

[0169] Furthermore, this embodiment also provides a workpiece pose estimation system based on the outer contour and the laser line. The system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it executes the workpiece pose estimation method based on the outer contour and the laser line described in any one of the above.

[0170] The present invention provides a method and device for pose estimation based on an outer contour and a laser line. The method includes: First, construct a secondary database storing the laser line features and contour information on the contour surface through simulation or experimental methods, and record the relative position relationship between the laser line and the contour through parameters, and perform a rough estimation of the pose through contour similarity. Then, calculate the similarity between the laser lines, and select the contour with the highest similarity as the initial value of the rough estimation. After that, perform iterative optimization through an optimization method near the initial value to obtain a more accurate pose. After determining the pose of the part, estimate the position of the part in space based on the spatial position of the laser line and the contour.

[0171] Those skilled in the art can easily understand that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included within the protection scope of the present invention.

Claims

1. A workpiece pose estimation method based on the outer contour and the laser line, characterized in that, Including: Offline database establishment stage: S1: Construct a sampling system including a camera and a laser through simulation or experimental means. At each pose, use the laser to scan the workpiece and obtain the workpiece image from the camera's perspective. Extract the contour samples and laser line samples, and establish a secondary database corresponding to the pose and the contour and the laser line respectively; Online estimation stage: S2: Scan the workpiece with the laser, obtain the workpiece image through the camera and extract the measured contour and the measured laser line of the workpiece, and match them with the samples in the secondary database to obtain a rough estimated pose; S3: According to the camera pose, the workpiece image, and the laser scanning plane, respectively obtain the spatial coordinates of the measured laser line and the laser line corresponding to the rough estimated pose; taking the minimum distance between the measured laser line and the laser line corresponding to the rough estimated pose as the target, optimize the rough estimated pose to obtain the optimized estimated pose; S4: Obtain the estimated position coordinates of the workpiece according to the camera pose and the measured contour.

2. The workpiece pose estimation method based on the outer contour and the laser line according to claim 1, wherein The offline database establishment stage and the online estimation stage respectively further include: establishing a part coordinate system and a camera coordinate system, where the Z-axis of the part coordinate system coincides with the camera optical axis; representing the pose of the workpiece using the Euler angles in the order of ZXZ of the part coordinate system; In S1, the sampling system is used to sample information of the workpiece in different poses, specifically including: Using the spherical Fibonacci sampling method to determine the distribution of the sampling viewpoints. Among them, the first two Euler angle components of the pose can be determined according to the sampling viewpoints; At each sampling viewpoint, let the workpiece rotate one week around the Z-axis of the part coordinate system according to a preset angular step to perform information sampling, and obtain multiple poses with different third Euler angle components at each sampling viewpoint.

3. The workpiece pose estimation method based on the outer contour and the laser line according to claim 2, wherein In S1, at each pose, use the laser to scan the workpiece and obtain the workpiece image from the camera's perspective. According to the workpiece image, extract the contour samples and laser line samples of the workpiece, specifically including: Extract the workpiece contour corresponding to each pose according to the workpiece image; At each pose, move the laser so that different positions of the workpiece are scanned by the laser multiple times, respectively obtain the laser scanning images and extract the laser lines, and obtain multiple laser lines corresponding to different positions of the workpiece; Normalize the workpiece contour and the laser line; For the workpiece contours corresponding to multiple poses with different third Euler angle components at each sampling viewpoint, select one of them as the representative contour, extract the feature descriptor from the normalized representative contour to form the contour sample, and each sampling viewpoint corresponds to one contour sample; extract the feature descriptor from the normalized laser line to form the laser line sample, and each pose corresponds to multiple laser line samples.

4. The workpiece pose estimation method based on the outer contour and the laser line according to claim 3, characterized in that The contour sample includes the Fourier descriptor of the contour, and the laser line sample includes the normal vector feature of the laser line; Specifically, forming the laser line sample includes: For any laser line, segment it according to continuity. For each segment of the laser line, downsample to obtain representative points, and fit the normal vector with the representative points within the preset neighborhood range of any representative point. The set of normal vectors corresponding to all representative points on any laser line forms the laser line sample corresponding to the any laser line.

5. The workpiece pose estimation method based on the outer contour and the laser line according to claim 3, characterized in that S2 specifically includes: Normalizing the measured contour and the measured laser line, and then extracting feature descriptors respectively; Matching the measured contour with the contour sample through feature descriptors, and determining k contour samples with smaller distances as k candidate representative contours according to the Euclidean distances between the feature descriptors; According to the angle deviation between the measured contour and the candidate representative contour in the circumferential direction, matching the third Euler angle component among multiple postures corresponding to each candidate representative contour is performed to determine a candidate posture, and a total of k candidate postures are determined; The measured laser line is matched with laser line samples corresponding to k candidate postures through feature descriptors to obtain a rough estimated posture.

6. The workpiece pose estimation method based on the outer contour and the laser line according to claim 5, wherein The secondary database corresponding to the posture, contour and laser line also includes: For any pose, an circumscribed circle is extracted from the normalized workpiece contour, and the centroid of the normalized laser line is calculated. The ratio rela of the distance from the centroid of any laser line to the center of the circumscribed circle to the radius of the circumscribed circle is recorded in the secondary database data ; The measured laser line is matched with the laser line samples corresponding to the k candidate postures through feature descriptors to obtain a rough estimated posture, specifically including: Extract the circumscribed circle of the measured profile after normalization, calculate the centroid of the measured laser line after normalization, and obtain the ratio rela of the distance from the centroid of the measured laser line to the center of the circumscribed circle of the measured profile to the radius of the circumscribed circle real ; Take rela real Compare it with the rela corresponding to all laser lines in k candidate poses data Select the laser lines with a difference less than a preset threshold as candidate laser lines; Calculating the similarity between the candidate laser line and the measured laser line, selecting the candidate laser line with the highest similarity as the matching laser line, and using the corresponding posture as the roughly estimated posture; Among them, the laser line distance is used to compare the similarity, specifically: For each first representative point obtained by downsampling the measured laser line, place it at the same pixel point of the image where the candidate laser line is located and search for the second representative point obtained by downsampling the candidate laser line within a preset neighborhood range, and calculate the deviation d between the representative points point , the deviation d point The calculation is as follows: If a unique second representative point is searched, calculate the angle between the normal vectors corresponding to the first representative point and the second representative point and the Euclidean distance between the two points, and sum them with weights as the deviation d point ; If multiple second representative points are found, calculate the deviation d between the first representative point and each second representative point point and retain the smallest deviation d point ; If no second representative point is searched, take a relatively large constant as the deviation d point ; Sum the deviations d calculated for all the first representative points point as the distance, i.e., similarity, between the measured laser line and the candidate laser line 7. The workpiece pose estimation method based on the outer contour and the laser line according to claim 1, wherein In S3, the spatial coordinates of the measured laser line and the laser line corresponding to the roughly estimated posture are respectively obtained according to the camera posture, the workpiece image and the laser scanning plane, which specifically includes: The camera pose is obtained through calibration, and the spatial coordinates of the camera optical center are determined according to the camera pose; For any point on the measured laser line, the spatial coordinates corresponding to the point on the camera imaging plane are obtained according to the pixel coordinates and the camera posture, a ray equation passing through the camera optical center and the point on the camera imaging plane is established, and the spatial coordinates of the measured laser line are obtained by combining the ray equation corresponding to the point on the measured laser line with the laser scanning plane corresponding to the measured laser line; The spatial coordinates of the laser line corresponding to the roughly estimated posture are obtained by combining the ray equations corresponding to the points on the laser line under the roughly estimated posture in the secondary database with the laser scanning plane corresponding to the laser line under the roughly estimated posture.

8. The workpiece pose estimation method based on the outer contour and the laser line according to claim 3, wherein, In S3, the rough estimated posture is optimized with the goal of minimizing the distance between the measured laser line and the laser line corresponding to the rough estimated posture, and the optimized estimated posture is obtained, which specifically includes: Obtaining a spatial point cloud including the plurality of laser lines by interpolation according to the spatial coordinates of the plurality of laser lines corresponding to the roughly estimated posture; A compensation rotation matrix and a translation vector are set, and an optimized compensation rotation matrix is obtained with the purpose of minimizing the sum of distances between all points on the measured laser line and the spatial point cloud after being corrected by the compensation rotation matrix and the translation vector; The roughly estimated attitude is corrected using the optimized compensation rotation matrix to obtain the optimized estimated attitude.

9. The workpiece pose estimation method based on the outer contour and the laser line according to claim 1, characterized in that, S4 specifically includes: Extracting the center of the circumscribed circle in the contour image corresponding to the measured contour; According to the camera posture and the pixel coordinates of the center of the circumscribed circle of the measured contour, the spatial coordinates of the center of the circumscribed circle of the measured contour on the camera imaging plane are obtained, and a ray equation passing through the camera optical center and the center of the circumscribed circle of the measured contour on the camera imaging plane is established; The ray equation from the optical center of the camera to the center of the circumcircle of the measured contour on the camera imaging plane is intersected with the actual laser scanning plane to obtain the spatial coordinates of a point on the workpiece as the estimated position coordinates of the workpiece.

10. A workpiece pose estimation system based on the outer contour and the laser line, characterized in that, The system includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it executes the workpiece pose estimation method based on the outer contour and the laser line according to any one of claims 1-9 above.

Images