A 3D point cloud-based intelligent cutting method with bones and meat

Abstract

The application discloses a kind of intelligent cutting methods with bone meat based on 3D point cloud, belong to intelligent cutting technical field.The 3D vision sensor real-time acquisition cross section profile data with bone meat is combined with conveying displacement reconstruction three-dimensional point cloud model;For rib bending, intercostal boundary is unclear and meat layer thickness is uneven problem, introduce the joint constraint of concave depth, local curvature and cross section continuity, adaptive segmentation independent rib point cloud area.Direct to single rib, extract continuous cross section center point and fit cubic B-spline skeleton curve, determine the number of cutting and cutting position by arc length parameterization adaptive;Again, establish normal section based on skeleton tangent vector, and generate final cutting point by combining curvature and thickness correction.Coordinate conversion and fixed distance compensation output execution parameter, realize accurate length cutting with bone meat, improve cutting precision and product consistency.

Description

Technical Field

[0001] This invention relates to the field of intelligent cutting technology, specifically to an intelligent cutting method for bone-in meat based on 3D point clouds. Background Technology

[0002] Intelligent upgrading of the meat processing industry is a key way to improve production efficiency and product quality. In the process of cutting bone-in meat products, due to the complexity of the raw meat's shape and the invisibility of its internal skeletal structure, obtaining the spatial contour information of the object to guide mechanical equipment for automated cutting has become an important research direction in this field.

[0003] However, existing automated cutting methods still have many technical limitations. On the one hand, current cutting path planning mostly adopts fixed-direction cutting or planning methods based solely on simple contours. This approach is difficult to adapt to raw meat, especially parts with natural curvature and local thickness variations. In actual operation, this insufficient adaptability leads to significant deviations in cutting positions, making it difficult to maintain consistent cutting lengths. When dealing with curved parts, it is easy to over-cut or under-cut, severely affecting the meat yield and the appearance of the final product. On the other hand, in actual production lines, there is usually a certain spatial distance between the vision scanning station and the mechanical cutting station. Existing technologies lack effective spatial coordinate transformation and system error compensation mechanisms, making it impossible to accurately map the positional information acquired by vision to the actuator. This ultimately leads to deviations in the cutting execution position, significantly reducing the actual operational accuracy and reliability of the automated cutting system. Summary of the Invention

[0004] The purpose of this invention is to provide a method for intelligent cutting of bone-in meat based on 3D point cloud, so as to solve the problems mentioned in the background art.

[0005] To address the aforementioned technical problems, this invention provides a method for intelligently cutting bone-in meat based on 3D point clouds, comprising:

[0006] S1. Point cloud acquisition and 3D reconstruction: The raw meat moves along a predetermined direction under the drive of the conveyor mechanism. When the raw meat passes the photoelectric switch, the photoelectric switch outputs a trigger signal, which triggers the laser scanning camera to collect continuous cross-sectional contour data of the raw meat.

[0007] By combining the displacement information, the contour data of each section are stitched together to reconstruct a three-dimensional point cloud model of the raw meat.

[0008] Let the first The set of cross-sectional contour points obtained from frame acquisition is as follows:

[0009] ;

[0010] In the formula, For the first The set of cross-sectional contour points obtained from frame acquisition; For the first The first frame The coordinates of each contour point in the horizontal direction.

[0011] For the first The first frame The coordinates of each contour point in the height direction; The index of the contour point in the current frame; This is the sampling frame number.

[0012] Let the displacement corresponding to a unit pulse of the encoder be... , No. The cumulative pulse count corresponding to the frame is Then the position of the frame in the transport direction is:

[0013] ;

[0014] In the formula, For the first The position coordinates of the frame in the transport direction; For the first The cumulative number of pulses corresponding to the frame; This represents the amount of transport displacement corresponding to a single encoder pulse.

[0015] Then the first Frame contour data can be reconstructed into a 3D point set:

[0016] ;

[0017] In the formula, For the first The three-dimensional point set corresponding to the frame; Point number; This is the sampling frame number.

[0018] By stitching all frames together sequentially according to the conveying direction, a 3D point cloud model of the raw meat is obtained:

[0019] ;

[0020] In the formula, A three-dimensional point cloud model of the raw meat; This represents the total number of sampled frames.

[0021] S2. Point cloud preprocessing: Perform region of interest clipping, statistical outlier filtering, and voxel downsampling on the original point cloud in sequence to remove background points, outlier noise points, and reduce the amount of point cloud data to obtain a preprocessed point cloud.

[0022] Setting points of The number of neighboring points is Its average neighborhood distance is:

[0023] ;

[0024] In the formula, For point The average neighborhood distance; Points to be processed; For point The 1 neighboring point; For point Its first Euclidean distance between neighboring points; The number of neighboring points; This is the index of the neighboring point.

[0025] Because the surface of the raw meat may have local reflections or interference from stray light from the environment, the generated point cloud will contain a small number of suspended isolated points that are far away from the main point cloud.

[0026] Let the mean and standard deviation of the average distance between all points in their neighborhoods be respectively... and Then it satisfies Points were identified as outliers and removed; among them, This is the outlier determination coefficient.

[0027] Statistical outlier filtering can accurately identify and remove floating noise points that deviate from the main distribution by calculating the average distance of the local neighborhood and setting a dynamic threshold using the global mean and standard deviation.

[0028] When downsampling voxels, the mean coordinates of all points within each non-empty voxel are used as the representative point of that voxel:

[0029] ;

[0030] In the formula, For the first Representative point of an individual element unit; For the first The set of points within an individual element unit; For the first The number of points contained within an individual unit.

[0031] Voxel downsampling divides the three-dimensional space into a fixed-size three-dimensional grid and replaces all points within a voxel with the centroid of all points within that voxel.

[0032] Statistical outlier filtering achieves adaptive denoising based on global statistical features, protecting edge features from accidental deletion. Voxel downsampling, while ensuring that the spatial geometry of the point cloud is not distorted, unifies the distribution density of the point cloud and significantly reduces data redundancy.

[0033] S3. Initial segmentation of independent ribs: Extract multiple cross-sectional contours along the transverse direction of the raw meat, and construct a height function on each cross-section. Local concave points are extracted as candidate points for rib boundary.

[0034] Since the boundary between the ribs appears as an inward concave shape on the point cloud cross section, the depth of the concavity reflects the degree of concavity at the boundary.

[0035] For candidate points The depth of the depression is defined as: ;in, For the first The first section The depth of the depression at each candidate boundary point; For the first The height function of a cross section.

[0036] For the first The first section The location of each candidate boundary point; This refers to the position of the peak point adjacent to the left of the candidate point; This represents the position of the peak point adjacent to the right of the candidate point.

[0037] The local curvature term of the candidate boundary point is defined as:

[0038] ;

[0039] In the formula, For the first The first section Local curvature terms at candidate boundary points; The height function with respect to the horizontal coordinate The second derivative of .

[0040] Local curvature reflects the degree of geometric abrupt change in the surface at the boundary; the greater the curvature, the more obvious the junction between bone and muscle.

[0041] To ensure the continuity of the boundary between adjacent sections, a continuity constraint term is introduced:

[0042] ;

[0043] In the formula, For the first The continuity constraint term for the c-th candidate boundary point in the cross section; This is the location of the boundary point already determined in the previous section; This is a continuously adjustable parameter.

[0044] The continuity constraint utilizes the characteristic that the position of the rib dividing point on adjacent sections has a smooth and gradual change in space to prevent the dividing point position from jumping due to local meat damage or noise.

[0045] Constructing the boundary point scoring function:

[0046] ;

[0047] In the formula, For the first The first section The comprehensive score of each candidate boundary point; For the depth of the depression; For local curvature terms; This is a continuity constraint term; , , These are the weighting coefficients.

[0048] By weighting and integrating the depth term reflecting the concave morphology, the curvature term reflecting geometric abrupt changes, and the constraint term reflecting spatial continuity, a multi-dimensional comprehensive evaluation mechanism is constructed.

[0049] This allows for robust and accurate tracking and determination of rib boundaries even when there are local defects or deformations on the surface of the raw meat, greatly improving the accuracy of segmenting independent rib regions under complex phenotypes.

[0050] In each cross section, the candidate point with the highest score is selected as the rib boundary position of the current cross section, and the corresponding boundary points are connected between consecutive cross sections, thereby dividing the overall raw meat point cloud into multiple independent rib point cloud regions:

[0051] ;

[0052] In the formula, For the first The point cloud region corresponding to each independent rib; The total number of independent rib point cloud regions; This is the serial number of an independent rib.

[0053] S4. Single Rib Skeleton Curve Extraction and Fitting: For each independent rib point cloud region... The local geometric skeleton is extracted and the skeleton curve is fitted, specifically including the following steps:

[0054] S4-1. Establish local reference directions for ribs: For the first... Independent rib point cloud region Principal component analysis is performed on all points to obtain their principal extension direction vectors. The main extension direction vector is used as the local length direction of the rib.

[0055] Let the first The rib-strip point cloud region is:

[0056] ;

[0057] Calculate its centroid:

[0058] ;

[0059] Because ribs in their natural state are long and thin in space, the variance of their data distribution is greatest in this direction.

[0060] Calculate the covariance matrix of all points in the independent rib point cloud region, and perform eigenvalue decomposition on the covariance matrix. The eigenvector corresponding to the largest eigenvalue is the main extension direction vector of the rib point cloud.

[0061] Construct the covariance matrix:

[0062] ;

[0063] Take the covariance matrix The eigenvector corresponding to the largest eigenvalue is used as the main extension direction vector. .

[0064] Principal component analysis is used to adaptively extract the natural extension direction of each rib without the need for a fixed cutting axis, so that subsequent slicing and extraction operations can always be performed perpendicular to the actual direction of the rib itself.

[0065] To avoid distortion of the slice cross-section caused by rib bending or misalignment, ensure that the extracted center trajectory point can accurately reflect the central axis shape of the rib.

[0066] In the formula, For the first The centroid of the rib-striped point cloud region; Point cloud region The number of points in the middle; For the first Covariance matrix of the rib-strip point cloud region; For the first The main extension direction vector of the rib.

[0067] S4-2. Extracting local sections along the main extension direction: along the main extension direction According to the preset step size For rib point cloud regions By performing layered slicing, multiple local cross-sectional point sets are obtained. .

[0068] Let the first The cross-sectional range of each local section is:

[0069] ;

[0070] Then the first The set of local cross-section points is represented as follows:

[0071] ;

[0072] In the formula, Extract step size for local sections; For point Projection value in the main extension direction; For the first Ribs in the first A set of points within a local section.

[0073] S4-3. Extracting the center points of local sections: For each set of local section points... Calculate the geometric center, which serves as the center point of this local section:

[0074] ;

[0075] In the formula, For the first Ribs in the first The center point within a local cross section; For the first The number of points within a local cross section.

[0076] Sort all local cross-section center points in ascending order of their projection values ​​along the main extension direction to form a set of center trajectory points:

[0077] ;

[0078] In the formula, For the first The set of center trajectory points of the ribs; This represents the number of local cross sections corresponding to the rib.

[0079] S4-4, Center Trajectory Smoothing and Outlier Removal: To reduce the impact of local missing points, noise points, and irregular abrupt changes on skeleton fitting, the center trajectory point set is... The system performs smoothing and removes center points whose distance between adjacent trajectory points exceeds a preset distance threshold.

[0080] S4-5, Skeleton Curve Fitting and Arc Length Parameterization: The smoothed center trajectory point set is fitted using a cubic B-spline to obtain the first... Rib skeleton curve:

[0081] ;

[0082] In the formula, For the first The skeletal curves of the ribs; These are the parameters for the skeleton curve. ; The basis functions are cubic B-spline functions. For the first One control point; The total number of control points is reduced by one.

[0083] Cubic B-spline curves possess local support and second derivative continuity, enabling them to maintain a smooth curve when passing through control points, preventing the entire curve from oscillating violently due to minor fluctuations in local center points.

[0084] To achieve subsequent fixed-length cutting, the skeleton curve is parameterized by arc length, and the arc length function is defined as follows:

[0085] ;

[0086] In the formula, For the first Rib skeleton curve from the starting point to the parameter position The arc length; For the skeleton curve with respect to parameters The first derivative; It is the integral variable.

[0087] Arc length parameterization establishes a one-to-one mapping relationship between the parameters of a curve and its actual physical length, so that equal intervals in the parameter space correspond to equal arc lengths in the physical space.

[0088] S5. Determining the cutting position based on the skeleton arc length: Let the target cutting length be... Then, the segmentation parameters satisfying the following formula are determined on the skeleton curve:

[0089] ;

[0090] ;

[0091] In the formula, For the first The first rib Skeleton curve parameters corresponding to each target cutting position; For the first The total arc length of the ribs; The preset target cutting length; The target cutting position number; This is for floor function.

[0092] When cutting curved ribs, if a fixed cutting plane parallel to the coordinate axis is used for cutting, the cutting plane will not be perpendicular to the extension direction of the rib, resulting in a slanted cutting plane and an inclined end face of the cut meat piece.

[0093] S6. Adaptive Normal Cut Construction: At each target cutting position At this point, find the unit tangent vector of the skeleton curve:

[0094] ;

[0095] In the formula, For the first Ribs in the first The unit tangent vector at each target cutting position.

[0096] Using the points corresponding to the skeleton curve as the passing points and the unit tangent vector as the normal vector, an adaptive normal tangent plane is established:

[0097] ;

[0098] In the formula, For the first Ribs in the first Adaptive normal tangent at each target cutting position; Let be the coordinate vector of any point on the cross surface; unit tangent vector Transpose of; For the skeleton curve in parameters The spatial coordinates of the location.

[0099] The adaptive normal tangent uses the tangent vector at the target position on the skeleton curve as the normal vector, so that the tangent rotates and adjusts in real time with the bending posture of the skeleton curve, and always remains strictly orthogonal to the extension direction of the rib at that point.

[0100] When the rib bends extremely sharply or its thickness changes abruptly at the target cutting location, a simple tangential normal cut may cut too much into the meat on one side.

[0101] At this point, a section correction amount needs to be introduced. The greater the local curvature, the more severe the bending, and the section angle needs to be deflected accordingly to conform to the bending trend.

[0102] The greater the thickness deviation, the more uneven the meat distribution on both sides. It is necessary to make a slight adjustment to the cut surface in the opposite direction to make it lean towards the thinner side and avoid the bone interference on the thicker side.

[0103] To accommodate local bending and thickness variations in the ribs, a further correction is introduced:

[0104] ;

[0105] In the formula, For the first Ribs in the first The amount of cut surface correction at each target cutting position; This represents the local curvature of the skeleton curve at that location; This refers to the local thickness deviation at that location; and This is a correction factor.

[0106] The corrected normal tangent is:

[0107] ;

[0108] In the formula, For the first The rib skeleton curve in the first The normal vector at each target cutting position.

[0109] Introducing corrections based on curvature and thickness deviations allows for secondary fine compensation of the cut surface posture under extreme conditions, further preventing situations where the tool is damaged or the cut surface becomes severely irregular due to cutting into the bone in sharp bends or areas with thickness distortion.

[0110] S7. Cutting point calculation: Take the distance normal tangent plane. Less than the threshold The points form a local cutting region:

[0111] ;

[0112] In the formula, For the first Ribs in the first The local cutting area corresponding to each target cutting position; This represents the distance threshold for the local cut-off region.

[0113] Use the geometric center of the locally cut region as the cutting point:

[0114] ;

[0115] In the formula, For the first Ribs in the first Cutting points at each target cutting location; This represents the number of points within the local cutting region.

[0116] Therefore, we obtain the first... Set of cutting points for the ribs:

[0117] ;

[0118] In the formula, For the first The set of cutting points corresponding to each rib; For the first The number of cutting points corresponding to each rib.

[0119] S8, Coordinate Transformation, Fixed Distance Compensation and Cutting Execution: Transforms the cutting point from the point cloud coordinate system to the machine coordinate system.

[0120] Let the coordinates of the cutting point in the point cloud coordinate system be... Its coordinates in the machine coordinate system are The homogeneous transformation matrix from the point cloud coordinate system to the machine coordinate system is: Then we have:

[0121] ;

[0122] In the formula, This is the homogeneous transformation matrix from the point cloud coordinate system to the machine coordinate system; The coordinates of the cutting point in the point cloud coordinate system; The coordinates of the cutting point in the machine coordinate system.

[0123] In actual production lines, there is a fixed physical distance between the scanning station where the line laser scanning camera is located and the cutting station where the cutting saw is located.

[0124] After the raw meat completes 3D reconstruction and cutting point calculation at the scanning station, it needs to continue moving forward by the conveyor belt for a fixed distance to reach the cutting station.

[0125] Therefore, the calculated coordinates of the cutting point in the conveying direction need to be superimposed with this fixed spatial distance offset to serve as the final target position for mechanical execution.

[0126] Since the scanning station is located upstream of the cutting station, and there is a fixed distance between them along the conveying direction, it is necessary to compensate for the fixed distance between the scanning station and the cutting station before sending the cutting point coordinates to the lower-level machine.

[0127] Let the fixed distance along the conveying direction between the scanning station and the cutting station be... Then, the final execution position of the cutting point along the conveying direction in the mechanical coordinate system satisfies:

[0128] ;

[0129] In the formula, The coordinates of the cutting point along the conveying direction in the point cloud coordinate system; This represents the final execution position of the cutting point in the machine coordinate system.

[0130] These are the scaling coefficients; This is the zero-point offset coefficient; This refers to the fixed distance between the scanning station and the cutting station along the conveying direction.

[0131] If the conveying mechanism uses an encoder for displacement detection, and the conveying displacement corresponding to a unit encoder pulse is... The cumulative number of pulses corresponding to the cutting execution is:

[0132] ;

[0133] In the formula, The cumulative number of pulses corresponding to the cutting execution.

[0134] The lower-level machine receives the cutting point information after coordinate transformation and fixed distance compensation. Based on the final execution position or the cumulative number of pulses, it controls the conveying mechanism and the cutting execution unit to work together so that the raw meat moves to the cutting station and completes the corresponding cutting.

[0135] The converted cutting point information is sent to the lower-level machine, which controls the cutting actuator to move to the corresponding position in sequence according to the cutting point information to complete the fixed-length cutting of meat with bone.

[0136] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0137] Existing technologies often employ fixed-direction cutting or path planning strategies relying solely on simple contours, making it difficult to adapt to the natural curvature and local thickness variations of raw meat, especially encountering obstacles in complex anatomical regions where bones and muscles intersect. This solution proposes a path generation method based on geometric skeleton extraction and skeleton curve fitting, combined with an adaptive normal slicing surface construction mechanism along the skeleton tangent vector, and introduces curvature and thickness correction factors. This allows the cutting surface to dynamically conform to complex curved surface shapes, thereby avoiding the overcutting or undercutting problems caused by traditional rigid cutting methods.

[0138] Traditional manual or semi-mechanized cutting methods struggle to ensure uniform cut sizes, especially when processing large volumes of raw meat and performing multiple cuts, often resulting in inconsistent piece sizes. This solution implements fixed-length cutting based on the actual arc length of the skeletal curve, overcoming the limitations of traditional straight-line-based cutting methods in curved conditions, significantly improving the consistency of cut segment lengths and the standardization of the final product.

[0139] When dealing with areas where bone structure is not visible and muscles and connective tissue are tightly interwoven, existing automated deboning techniques often struggle to accurately position and smoothly advance the cutting tool, leading to material waste and reduced product quality. This solution constructs a scoring function that integrates local depression depth, curvature features, and continuity constraints to determine the optimal boundary position. This enables accurate and coherent segmentation of independent regions from the overall 3D point cloud, enhancing the system's robustness in positioning on complex surface topography and irregular structures.

[0140] Existing meat cutting equipment often suffers from deviations in the final cutting position due to spatial distance between the visual scanning station and the mechanical execution station, and the lack of effective coordinate transformation and error compensation mechanisms. This solution establishes a homogeneous transformation matrix from the point cloud coordinate system to the machine coordinate system, and combines it with a fixed distance compensation strategy between the scanning station and the cutting station. This effectively eliminates systematic errors caused by spatial displacement, ensuring high-precision reproduction of the vision-planned path in actual cutting execution. Attached Figure Description

[0141] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0142] Figure 1 This is a flowchart illustrating an intelligent bone-in meat cutting method based on 3D point clouds according to the present invention.

[0143] Figure 2 This is a schematic diagram illustrating the evolution of the effect from the original scanned point cloud to the cutting position generated by the present invention;

[0144] Figure 3 This is a schematic diagram of the adaptive normal cutting surface effect of the present invention;

[0145] Figure 4 This is a diagram of the intelligent pork chop cutting device according to Embodiment 2 of the present invention;

[0146] Reference numerals: 1-Scanner; 2-Conveyor belt; 3-Clamping device; 4-Cutter. Detailed Implementation

[0147] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0148] Example 1:

[0149] Please see Figure 1 This embodiment provides a method for intelligent cutting of bone-in meat based on 3D point clouds, including:

[0150] S1. Point cloud acquisition and 3D reconstruction: The raw meat moves along a predetermined direction under the drive of the conveyor mechanism. When the raw meat passes the photoelectric switch, the photoelectric switch outputs a trigger signal, which triggers the laser scanning camera to collect continuous cross-sectional contour data of the raw meat.

[0151] By combining the displacement information, the contour data of each section are stitched together to reconstruct a three-dimensional point cloud model of the raw meat.

[0152] Let the first The set of cross-sectional contour points obtained from frame acquisition is as follows:

[0153] ;

[0154] In the formula, For the first The set of cross-sectional contour points obtained from frame acquisition; For the first The first frame The coordinates of each contour point in the horizontal direction.

[0155] For the first The first frame The coordinates of each contour point in the height direction; The index of the contour point in the current frame; This is the sampling frame number.

[0156] Let the displacement corresponding to a unit pulse of the encoder be... , No. The cumulative pulse count corresponding to the frame is Then the position of the frame in the transport direction is:

[0157] ;

[0158] In the formula, For the first The position coordinates of the frame in the transport direction; For the first The cumulative number of pulses corresponding to the frame; This represents the amount of transport displacement corresponding to a single encoder pulse.

[0159] Then the first Frame contour data can be reconstructed into a 3D point set:

[0160] ;

[0161] In the formula, For the first The three-dimensional point set corresponding to the frame; Point number; This is the sampling frame number.

[0162] By stitching all frames together sequentially according to the conveying direction, a 3D point cloud model of the raw meat is obtained:

[0163] ;

[0164] In the formula, A three-dimensional point cloud model of the raw meat; This represents the total number of sampled frames.

[0165] S2. Point cloud preprocessing: Perform region of interest clipping, statistical outlier filtering, and voxel downsampling on the original point cloud in sequence to remove background points, outlier noise points, and reduce the amount of point cloud data to obtain a preprocessed point cloud.

[0166] Setting points of The number of neighboring points is Its average neighborhood distance is:

[0167] ;

[0168] In the formula, For point The average neighborhood distance; Points to be processed; For point The 1 neighboring point; For point Its first Euclidean distance between neighboring points; The number of neighboring points; This is the index of the neighboring point.

[0169] In the actual implementation process, due to the possibility of local reflections or interference from stray light on the surface of the raw meat, a small number of suspended isolated points far away from the main point cloud will be mixed in the generated point cloud.

[0170] Let the mean and standard deviation of the average distance between all points in their neighborhoods be respectively... and Then it satisfies Points were identified as outliers and removed; among them, This is the outlier determination coefficient.

[0171] Statistical outlier filtering can accurately identify and remove floating noise points that deviate from the main distribution by calculating the average distance of the local neighborhood and setting a dynamic threshold using the global mean and standard deviation.

[0172] When downsampling voxels, the mean coordinates of all points within each non-empty voxel are used as the representative point of that voxel:

[0173] ;

[0174] In the formula, For the first Representative point of an individual element unit; For the first The set of points within an individual element unit; For the first The number of points contained within an individual unit.

[0175] In practice, voxel downsampling involves dividing the three-dimensional space into a fixed-size three-dimensional grid (voxels), and replacing all points within that voxel with the centroid of all points within that voxel.

[0176] Statistical outlier filtering achieves adaptive denoising based on global statistical features, protecting edge features from accidental deletion. Voxel downsampling, while ensuring that the spatial geometry of the point cloud is not distorted, unifies the distribution density of the point cloud and significantly reduces data redundancy.

[0177] S3. Initial segmentation of independent ribs: Extract multiple cross-sectional contours along the transverse direction of the raw meat, and construct a height function on each cross-section. Local concave points are extracted as candidate points for rib boundary.

[0178] In the specific implementation process, since the boundary between the ribs appears as an inward concave shape on the point cloud cross section, the depth of the concavity reflects the degree of concavity at the boundary.

[0179] For candidate points The depth of the depression is defined as: ;in, For the first The first section The depth of the depression at each candidate boundary point; For the first The height function of each cross section.

[0180] For the first The first section The location of each candidate boundary point; This refers to the position of the peak point adjacent to the left of the candidate point; This represents the position of the peak point adjacent to the right of the candidate point.

[0181] The local curvature term of the candidate boundary point is defined as:

[0182] ;

[0183] In the formula, For the first The first section Local curvature terms at candidate boundary points; The height function with respect to the horizontal coordinate The second derivative of .

[0184] Local curvature reflects the degree of geometric abrupt change in the surface at the boundary; the greater the curvature, the more obvious the junction between bone and muscle.

[0185] To ensure the continuity of the boundary between adjacent sections, a continuity constraint term is introduced:

[0186] ;

[0187] In the formula, For the first The continuity constraint term for the c-th candidate boundary point in the cross section; This is the location of the boundary point already determined in the previous section; This is a continuously adjustable parameter.

[0188] In the specific implementation process, the continuity constraint utilizes the characteristic that the position of the rib dividing point on adjacent sections has a smooth and gradual change in space to prevent the dividing point position from jumping due to local meat damage or noise.

[0189] Construct the boundary point scoring function:

[0190] ;

[0191] In the formula, For the first The first section The comprehensive score value of each candidate boundary point; For the depth of the depression; For local curvature terms; This is a continuity constraint term; , , These are the weighting coefficients.

[0192] In the specific implementation process, a multi-dimensional comprehensive evaluation mechanism is constructed by weighting and integrating the depth term reflecting the concave morphology, the curvature term reflecting geometric abrupt changes, and the constraint term reflecting spatial continuity.

[0193] This allows for robust and accurate tracking and determination of rib boundaries even when there are local defects or deformations on the surface of the raw meat, greatly improving the accuracy of segmenting independent rib regions under complex phenotypes.

[0194] In each cross section, the candidate point with the highest score is selected as the rib boundary position of the current cross section, and the corresponding boundary points are connected between consecutive cross sections, thereby dividing the overall raw meat point cloud into multiple independent rib point cloud regions:

[0195] ;

[0196] In the formula, For the first The point cloud region corresponding to each independent rib; The total number of independent rib point cloud regions; This is the serial number of an independent rib.

[0197] Please see Figure 2 The figure shows the evolution of the effect generated from the original scanned point cloud to the cutting position in this invention.

[0198] First, the raw point cloud obtained from the scan is preprocessed to remove the background and extract the target region.

[0199] Secondly, the preserved target area is reconstructed in three dimensions to obtain a complete point cloud model of the bone and flesh.

[0200] Then, the effective regions that actually participate in the segmentation are further screened from the point cloud model, and the normalized target point cloud is obtained.

[0201] Finally, multiple cutting position lines are generated on the effective area according to a preset cutting interval.

[0202] S4. Single Rib Skeleton Curve Extraction and Fitting: For each independent rib point cloud region... The local geometric skeleton is extracted and the skeleton curve is fitted. Specifically, the following steps are included:

[0203] S4-1. Establish local reference directions for ribs: For the first... Independent rib point cloud region Principal component analysis is performed on all points to obtain their principal extension direction vectors. The main extension direction vector is used as the local length direction of the rib.

[0204] Let the first The rib-strip point cloud region is:

[0205] ;

[0206] Calculate its centroid:

[0207] ;

[0208] Because ribs in their natural state are long and thin in space, the variance of their data distribution is greatest in this direction.

[0209] Calculate the covariance matrix of all points in the independent rib point cloud region, and perform eigenvalue decomposition on the covariance matrix. The eigenvector corresponding to the largest eigenvalue is the main extension direction vector of the rib point cloud.

[0210] Construct the covariance matrix:

[0211] ;

[0212] Take the covariance matrix The eigenvector corresponding to the largest eigenvalue is used as the main extension direction vector. .

[0213] In the specific implementation process, the principal component analysis method is used to adaptively extract the natural extension direction of each rib, without the need to preset a fixed cutting axis, so that the subsequent slicing and extraction operations can always be carried out perpendicular to the actual direction of the rib itself.

[0214] To avoid distortion of the slice cross-section caused by rib bending or misalignment, ensure that the extracted center trajectory point can accurately reflect the central axis shape of the rib.

[0215] In the formula, For the first The centroid of the rib-striped point cloud region; Point cloud region The number of points in the middle; For the first Covariance matrix of the rib-strip point cloud region; For the first The main extension direction vector of the rib.

[0216] S4-2. Extracting local sections along the main extension direction: along the main extension direction According to the preset step size For rib point cloud regions By performing layered slicing, multiple local cross-sectional point sets are obtained. .

[0217] Let the first The cross-sectional range of each local section is:

[0218] ;

[0219] Then the first The set of local cross-section points is represented as follows:

[0220] ;

[0221] In the formula, Extract step size for local sections; For point Projection value in the main extension direction; For the first Ribs in the first A set of points within a local section.

[0222] S4-3. Extracting the center points of local sections: For each set of local section points... Calculate the geometric center, which serves as the center point of this local section:

[0223] ;

[0224] In the formula, For the first Ribs in the first The center point within a local cross section; For the first The number of points within a local cross section.

[0225] Sort all local cross-section center points in ascending order of their projection values ​​along the main extension direction to form a set of center trajectory points:

[0226] ;

[0227] In the formula, For the first The set of center trajectory points of the ribs; This represents the number of local cross sections corresponding to the rib.

[0228] S4-4, Center Trajectory Smoothing and Outlier Removal: To reduce the impact of local missing points, noise points, and irregular abrupt changes on skeleton fitting, the center trajectory point set is... The system performs smoothing and removes center points whose distance between adjacent trajectory points exceeds a preset distance threshold.

[0229] S4-5, Skeleton Curve Fitting and Arc Length Parameterization: The smoothed center trajectory point set is fitted using a cubic B-spline to obtain the first... Rib skeleton curve:

[0230] ;

[0231] In the formula, For the first The skeletal curves of the ribs; These are the parameters for the skeleton curve. ; The basis functions are cubic B-spline functions. For the first One control point; The total number of control points is reduced by one.

[0232] Cubic B-spline curves possess local support and second derivative continuity, enabling them to maintain a smooth curve when passing through control points, preventing the entire curve from oscillating violently due to minor fluctuations in local center points.

[0233] To achieve subsequent fixed-length cutting, the skeleton curve is parameterized by arc length, and the arc length function is defined as follows:

[0234] ;

[0235] In the formula, For the first Rib skeleton curve from the starting point to the parameter position The arc length; For the skeleton curve with respect to parameters The first derivative; It is the integral variable.

[0236] In practice, arc length parameterization establishes a one-to-one mapping relationship between the curve parameters and their actual physical space length, so that equal intervals in the parameter space correspond to equal arc lengths in the physical space.

[0237] S5. Determining the cutting position based on the skeleton arc length: Let the target cutting length be... Then, the segmentation parameters satisfying the following formula are determined on the skeleton curve:

[0238] ;

[0239] ;

[0240] In the formula, For the first The first rib Skeleton curve parameters corresponding to each target cutting position; For the first The total arc length of the ribs; The preset target cutting length; The target cutting position number; This is for floor operations.

[0241] In practice, when cutting curved ribs, if a fixed cutting surface parallel to the coordinate axis is used for cutting, the cutting surface will not be perpendicular to the extension direction of the rib, resulting in a slanted cutting surface and an inclined end face of the cut meat piece.

[0242] S6. Adaptive Normal Cut Construction: At each target cutting position At this point, find the unit tangent vector of the skeleton curve:

[0243] ;

[0244] In the formula, For the first Ribs in the first The unit tangent vector at each target cutting position.

[0245] Using the points corresponding to the skeleton curve as the passing points and the unit tangent vector as the normal vector, an adaptive normal tangent plane is established:

[0246] ;

[0247] In the formula, For the first Ribs in the first Adaptive normal tangent at each target cutting position; Let be the coordinate vector of any point on the cross surface; unit tangent vector transpose; For the skeleton curve in parameters The spatial coordinates of the location.

[0248] The adaptive normal tangent uses the tangent vector at the target position on the skeleton curve as the normal vector, so that the tangent rotates and adjusts in real time with the bending posture of the skeleton curve, and always remains strictly orthogonal to the extension direction of the rib at that point.

[0249] In practice, when the rib bends extremely sharply or its thickness changes abruptly at the target cutting position, a simple tangential normal cut may cut too much into the meat on one side.

[0250] At this point, a section correction amount needs to be introduced. The greater the local curvature, the more severe the bending, and the section angle needs to be deflected accordingly to conform to the bending trend.

[0251] The greater the thickness deviation, the more uneven the meat distribution on both sides. It is necessary to make a slight adjustment to the cut surface in the opposite direction to make it lean towards the thinner side and avoid the bone interference on the thicker side.

[0252] To accommodate local bending and thickness variations in the ribs, a further correction is introduced:

[0253] ;

[0254] In the formula, For the first Ribs in the first The amount of cut surface correction at each target cutting position; This represents the local curvature of the skeleton curve at that location; This refers to the local thickness deviation at that location; and This is a correction factor.

[0255] The corrected normal tangent is:

[0256] ;

[0257] In the formula, For the first The rib skeleton curve in the first The normal vector at each target cutting position.

[0258] Introducing corrections based on curvature and thickness deviations allows for secondary fine compensation of the cut surface posture under extreme conditions, further preventing situations where the tool is damaged or the cut surface becomes severely irregular due to cutting into the bone in sharp bends or areas with thickness distortion.

[0259] S7. Cutting point calculation: Take the distance normal tangent plane. Less than the threshold The points form a local cutting region:

[0260] ;

[0261] In the formula, For the first Ribs in the first The local cutting area corresponding to each target cutting position; This represents the distance threshold for the local cut-off region.

[0262] Use the geometric center of the locally cut region as the cutting point:

[0263] ;

[0264] In the formula, For the first Ribs in the first Cutting points at each target cutting location; This represents the number of points within the local cutting region.

[0265] Therefore, we obtain the first... Set of cutting points for the ribs:

[0266] ;

[0267] In the formula, For the first The set of cutting points corresponding to each rib; For the first The number of cutting points corresponding to each rib.

[0268] Please see Figure 3 The figure shows the adaptive normal cutting surface of the present invention, which performs further geometric analysis on the point cloud of a single independent rib to achieve accurate determination of the fixed-length cutting position.

[0269] First, extract the three-dimensional point cloud data of the individual ribs to obtain the complete spatial morphology of a single rib.

[0270] Subsequently, the point cloud of the rib was segmented and analyzed along its extension direction, and a skeleton curve representing the overall bending trend of the rib was obtained by fitting.

[0271] Based on the skeleton curve, the arc length is fixed-length segmented according to the preset cutting length to determine the positions of multiple target segments.

[0272] Finally, an adaptive normal cutting surface is constructed at each target segment location based on the local tangential direction of the skeleton curve.

[0273] In this way, the cutting surface can be dynamically adjusted according to the actual bending shape of the rib, thereby avoiding the positional deviation caused by traditional fixed-direction cutting on the bent rib and improving the accuracy and consistency of fixed-length cutting.

[0274] S8, Coordinate Transformation, Fixed Distance Compensation and Cutting Execution: Transforms the cutting point from the point cloud coordinate system to the machine coordinate system.

[0275] Let the coordinates of the cutting point in the point cloud coordinate system be... Its coordinates in the machine coordinate system are The homogeneous transformation matrix from the point cloud coordinate system to the machine coordinate system is: Then we have:

[0276] ;

[0277] In the formula, This is the homogeneous transformation matrix from the point cloud coordinate system to the machine coordinate system; The coordinates of the cutting point in the point cloud coordinate system; The coordinates of the cutting point in the machine coordinate system.

[0278] In actual production lines, there is a fixed physical distance between the scanning station where the line laser scanning camera is located and the cutting station where the cutting saw is located.

[0279] In the actual implementation process, after the raw meat completes the three-dimensional reconstruction and cutting point calculation at the scanning station, it needs to continue moving forward by the conveyor belt for a fixed distance to reach the cutting station.

[0280] Therefore, the calculated coordinates of the cutting point in the conveying direction need to be superimposed with this fixed spatial distance offset to serve as the final target position for mechanical execution.

[0281] Since the scanning station is located upstream of the cutting station, and there is a fixed distance between them along the conveying direction, it is necessary to compensate for the fixed distance between the scanning station and the cutting station before sending the cutting point coordinates to the lower-level machine.

[0282] Let the fixed distance along the conveying direction between the scanning station and the cutting station be... Then, the final execution position of the cutting point along the conveying direction in the mechanical coordinate system satisfies:

[0283] ;

[0284] In the formula, The coordinates of the cutting point along the conveying direction in the point cloud coordinate system; This represents the final execution position of the cutting point in the machine coordinate system.

[0285] These are the scaling coefficients; This is the zero-point offset coefficient; This refers to the fixed distance between the scanning station and the cutting station along the conveying direction.

[0286] If the conveying mechanism uses an encoder for displacement detection, and the conveying displacement corresponding to a unit encoder pulse is... The cumulative number of pulses corresponding to the cutting execution is:

[0287] ;

[0288] In the formula, The cumulative number of pulses corresponding to the cutting execution.

[0289] The lower-level machine receives the cutting point information after coordinate transformation and fixed distance compensation. Based on the final execution position or the cumulative number of pulses, it controls the conveying mechanism and the cutting execution unit to work together so that the raw meat moves to the cutting station and completes the corresponding cutting.

[0290] In the specific implementation process, the converted cutting point information is sent to the lower-level machine. The lower-level machine controls the cutting execution mechanism to move to the corresponding position in sequence according to the cutting point information to complete the fixed-length cutting of meat with bone.

[0291] Example 2:

[0292] Please see Figure 4 This embodiment discloses an intelligent pork chop cutting device.

[0293] The intelligent bone-in meat cutting method described in Embodiment 1 is adopted. The cutting equipment includes a scanner 1 located above the conveyor belt 2, and a clamping device 3 is provided at the end of the conveyor belt 2. The clamping device 3 is used to press and tighten the bone-in meat being passed through so that the cutting blade 4 can cut it. The cutting blade 4 includes a longitudinal cutting blade (the blade edge is along the direction of transmission of the conveyor belt 2) and a transverse cutting blade. Figure 4 The red arrow in the diagram indicates the conveying direction of conveyor belt 2.

[0294] First, the host computer is started and the initialization settings of each actuator and sensor are completed. The raw meat is placed on the conveyor belt 2, which transports the raw meat to the scanning station. When the photoelectric sensor detects that the raw meat has reached the predetermined position, the scanner 1 is triggered to collect the point cloud data on the surface of the raw meat, thus completing step S1 in Example 1. The host computer reconstructs, preprocesses, extracts the target area, fits the skeleton, and calculates the cutting position for the point cloud data. If the cutting position is successfully obtained, the corresponding cutting control parameters are generated and sent to the slave computer. The slave computer controls the conveying mechanism to transport the raw meat to the cutting station, and simultaneously controls the clamping device 3 and the cutting actuator (cutting blade 4) to complete the automatic cutting. After the cutting is completed, the conveyor belt 2 continues to discharge material, the system is reset, and the next cycle begins. If the cutting position is not successfully identified, the system alarms and returns to rescan or adjust the position of the raw meat.

[0295] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0296] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent cutting of bone-in meat based on 3D point clouds, characterized in that: The method includes: S1. Collect the cross-sectional contour data of the raw meat, combine it with the conveying displacement information to splice and reconstruct a three-dimensional point cloud model of the raw meat. S2. Perform set region clipping, statistical outlier filtering, and voxel downsampling on the 3D point cloud model in sequence to obtain the preprocessed point cloud; S3. Extract multiple cross-sectional contours along the transverse direction of the raw meat. Extract indentation points on each cross-section as candidate rib boundary points. Construct a scoring function based on indentation depth, local curvature, and continuity constraints to determine the boundary positions, thus segmenting the overall raw meat point cloud into multiple independent rib point cloud regions; specifically including: Different cross-sectional profiles are extracted along the transverse direction of the raw meat, and a height function is constructed on each cross-section. The indentation points were extracted as candidate boundary points between the ribs. ; Calculate the indentation depth of the candidate boundary point ;in, ; and These represent the positions of the adjacent peak points to the left and right of the candidate dividing point, respectively. The local curvature term of the candidate boundary point is defined as: ； In the formula, For the first The first section Local curvature terms at candidate boundary points; The height function with respect to the horizontal coordinate The second derivative; Calculate the continuity constraint term for the candidate boundary point: ； In the formula, For the first The first section Continuity constraint terms for each candidate boundary point; This is the location of the boundary point already determined in the previous section; For continuous adjustment parameters; Preset weighting coefficients respectively , and Construct the boundary point scoring function: ； In the formula, For the first The first section The comprehensive score value of each candidate boundary point; In each section, the candidate boundary point with the highest comprehensive score is taken as the rib boundary position of the current section, and the corresponding candidate boundary points are connected between continuous sections to divide the overall raw meat point cloud into multiple independent rib point cloud regions. S4. For each independent rib point cloud region, extract the geometric skeleton and perform skeleton curve fitting to obtain the skeleton curve. S5. Determine the target cutting position based on the preset target cutting length and the arc length of the skeleton curve; S6. At each target cutting position, calculate the unit tangent vector of the skeleton curve and establish an adaptive normal tangent plane; S7. Determine the cutting region based on the adaptive normal tangent and calculate the cutting point; S8. Convert the cutting point to the mechanical coordinate system and perform fixed distance compensation, then control the actuator to complete the cutting.

2. The intelligent bone-in meat cutting method based on 3D point cloud according to claim 1, characterized in that: S1 specifically includes: When the raw meat passes through the photoelectric switch, the trigger line laser scanning camera collects continuous cross-sectional contour data of the raw meat; Let the first The cross-sectional profile point set obtained from frame acquisition ;in, ; For the first The first frame The coordinates of each contour point in the horizontal direction. For the first The first frame The coordinates of each contour point in the height direction; Let the displacement corresponding to a unit pulse of the encoder be... , No. The cumulative number of pulses corresponding to the frame is The position of the frame in the transport direction is obtained. ;in, ; The first Frame contour data reconstructed into a 3D point set All frames are stitched together sequentially according to the conveying direction to obtain a 3D point cloud model of the raw meat. ;in, ; ; This represents the total number of sampled frames.

3. The intelligent bone-in meat cutting method based on 3D point cloud according to claim 1, characterized in that: The statistical outlier filtering in S2 specifically includes: Calculation points average neighborhood distance ;in, ; For point The Neighboring points, The number of neighboring points. The distance is Euclidean. Calculate the mean of the neighborhood average distances of all points. and standard deviation ; will satisfy Points were identified as outliers and removed; among them, The outlier determination coefficient; During voxel downsampling, the mean coordinates of all points within each non-empty voxel are used as the representative point of that voxel. ;in, ; For the first The set of points within an individual element unit. This represents the number of points contained within the voxel unit.

4. The intelligent bone-in meat cutting method based on 3D point cloud according to claim 1, characterized in that: S4 includes: S4-1, regarding the first Principal component analysis was performed on the point cloud regions of each independent rib to obtain their principal extension direction vectors. And the main extension direction vector is taken as the length direction of the rib; S4-2, along the main extension direction According to the preset step size The rib point cloud region is sliced ​​into layers to obtain multiple cross-sectional point sets; S4-3. Calculate the geometric center for each cross-section point set, and use it as the center point of that cross-section; sort all the cross-section center points according to the projection values ​​of the main extension direction to form the center trajectory point set.

5. The intelligent bone-in meat cutting method based on 3D point cloud according to claim 4, characterized in that: S4 also includes: S4-4, Set of Center Trajectory Points Perform smoothing processing and remove center points whose adjacent spacing is greater than a preset distance threshold; S4-5. Fit the smoothed center trajectory point set using cubic B-splines to obtain the skeleton curve of the rib: ； In the formula, For the first The skeletal curves of the ribs; These are the parameters for the skeleton curve. ; For cubic B-spline basis functions, For the first One control point, Subtract 1 from the total number of control points; Define arc length function The arc length of the skeleton curve is parameterized; in, ; For the first Rib skeleton curve from the starting point to the parameter position The arc length; For the skeleton curve with respect to parameters The first derivative; It is the integral variable.

6. The intelligent bone-in meat cutting method based on 3D point cloud according to claim 5, characterized in that: S5 specifically includes: The preset target slicing length is Determine the segmentation parameters on the skeleton curve that satisfy the following formula: ； ； In the formula, For the first The first rib Skeleton curve parameters corresponding to each target cutting position; For the first The total arc length of the ribs; This is a floor function; Step S6 specifically includes: At each target cutting location At this point, find the unit tangent vector of the skeleton curve. Using the points corresponding to the skeleton curve as the passing points and the unit tangent vector as the normal vector, an adaptive normal tangent plane is established: ； In the formula, For the first Ribs in the first Adaptive normal tangent at each target cutting position; unit tangent vector Transpose of; Let be the coordinate vector of any point on the cross surface; For the skeleton curve in parameters Spatial coordinates of the location; For the first Ribs in the first The amount of cut surface correction at each target cutting position; For the first The rib skeleton curve in the first The normal vector at each target cutting position.

7. The intelligent bone-in meat cutting method based on 3D point cloud according to claim 6, characterized in that: S6 mid-section correction amount The calculation formula is as follows: ； In the formula, and These are the preset correction coefficients; This represents the local curvature of the skeleton curve at that location. This represents the thickness deviation at that location.

8. The intelligent bone-in meat cutting method based on 3D point cloud according to claim 6, characterized in that: S7 specifically includes: Take the distance-adaptive normal tangent Less than the threshold point This constitutes the cutting area: ； In the formula, For the first Ribs in the first The cutting area corresponding to each target cutting position; For the first The point cloud region corresponding to each independent rib; Use the geometric center of the cutting region as the cutting point: ； In the formula, For the first Ribs in the first Cutting points at each target cutting location; This represents the number of points within the cut area. Therefore, we obtain the first... Set of cutting points for the ribs: ； In the formula, for The set of cutting points corresponding to each rib; For the first The number of cutting points corresponding to each rib.

9. The intelligent bone-in meat cutting method based on 3D point cloud according to claim 2, characterized in that: S8 specifically includes: Based on the homogeneous transformation matrix from the point cloud coordinate system to the machine coordinate system The cutting point is transformed from the point cloud coordinate system to the machine coordinate system; Based on the fixed distance along the conveying direction between the scanning station and the cutting station Calculate the final execution position of the cutting point along the conveying direction in the machine coordinate system. ; in, ; These are the scaling coefficients; This is the zero-point offset coefficient; The coordinates of the cutting point along the conveying direction in the point cloud coordinate system; If the conveying mechanism uses an encoder for displacement detection, the cumulative number of pulses corresponding to the cutting execution is calculated. : ； The lower-level machine controls the conveying mechanism and the cutting execution unit to work together to complete the cutting based on the final execution position or the cumulative number of pulses.

Images