Animal body size measurement method and system

Abstract

The present application relates to the technical field of body size measurement, and specifically provides an animal body size measurement method and system, including: when a target animal passes through the limiting fence, obtaining depth image data of the target animal in at least two different directions based on the depth camera; processing the depth image data based on the data processing device to obtain preliminary point cloud data; correcting the point cloud posture of the target animal corresponding to the preliminary point cloud data to obtain corrected point cloud data; reconstructing the point cloud based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal; and obtaining the body size data of the target animal based on the reconstructed three-dimensional model. The animal body size measurement method can improve the accuracy of the body size data obtained based on the reconstructed three-dimensional model by correcting the point cloud posture of the target animal and reconstructing the point cloud based on the corrected point cloud data.

Description

Technical Field

[0001] The present application relates to the technical field of body size measurement, and in particular to a method and system for measuring animal body size. Background Art

[0002] Animal body measurement mainly includes height, length, chest circumference, and tube circumference. These measurements are not only used to understand the growth and development of various parts of the animal and the nutritional status, but are also often used in scientific research, investigation, breeding, and weight estimation.

[0003] The traditional method of measuring animal body size is to manually use a tape measure to fix the animal before measuring. This is not only labor-intensive, but also causes great stress reactions in animals. At the same time, the animal's resistance will also affect the measurement accuracy. Therefore, achieving non-contact and stress-free body size parameter measurement of animals will help improve animal welfare and achieve a sustainable breeding model.

[0004] Non-contact body measurement technology has been well developed in livestock such as cattle, pigs, and horses. Non-contact livestock body measurement is generally achieved through training using image processing, machine learning, and sensor technology. However, in actual applications, image sensors or brackets used to fix image sensors are prone to displacement, resulting in low measurement precision and accuracy. Summary of the invention

[0005] In view of this, an object of the embodiments of the present application is to provide an animal body size measurement method and system to solve the technical problem of low measurement precision and accuracy of existing animal body size data methods.

[0006] In a first aspect, an embodiment of the present application provides an animal body size measurement method, the method being applied to an animal body size measurement system including a data processing device and a plurality of depth cameras; the plurality of depth cameras being configured to be installed at least two different positions of a limiting fence; the method comprising:

[0007] When the target animal passes through the limiting fence, acquiring depth image data of the target animal at at least two different positions based on the depth camera;

[0008] Based on the data processing device, the depth image data is processed to obtain preliminary point cloud data;

[0009] Correcting the point cloud posture of the target animal corresponding to the preliminary point cloud data to obtain corrected point cloud data;

[0010] Performing point cloud reconstruction based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal;

[0011] Based on the reconstructed three-dimensional model, body size data of the target animal is acquired.

[0012] In the above implementation process, the animal body size measurement method obtains depth image data of the target animal in at least two different directions based on the depth camera when the target animal passes through the limiting fence; processes the depth image data based on the data processing device to obtain preliminary point cloud data; corrects the point cloud posture of the target animal corresponding to the preliminary point cloud data to obtain corrected point cloud data; reconstructs the point cloud based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal; and obtains the body size data of the target animal based on the reconstructed three-dimensional model. The animal body size measurement method can improve the accuracy of the body size data obtained based on the reconstructed three-dimensional model by correcting the point cloud posture of the target animal and reconstructing the point cloud based on the corrected point cloud data. And by correcting the point cloud posture of the target animal, it can reduce the "measurement deviation caused by the easy displacement of the image sensor or the bracket used to fix the image sensor", solving the technical problem of low measurement precision and accuracy of the existing animal body size data method.

[0013] Optionally, in an embodiment of the present application, the system further includes: an identification device installed in front and rear positions of the limiting fence, the identification device being configured to identify tags on the target animal passing through the limiting fence; when the target animal passes through the limiting fence, acquiring depth image data of the target animal in at least two different positions based on the depth camera, including: when the identification device recognizes that the target animal passes through the limiting fence, acquiring depth image data of the target animal in at least two different positions based on the depth camera.

[0014] In the above implementation process, since the recognition device can recognize the tags on the target animals passing through the limiting fence, when the target animals pass through the limiting fence, the depth image data of the corresponding target animals can be acquired based on the depth camera, and the acquired depth image data can be associated with the tags on the target animals, thereby achieving accurate correspondence between the depth image data and the target animals, thereby improving the accuracy of the acquired body size data.

[0015] Optionally, in an embodiment of the present application, obtaining the body size data of the target animal based on the reconstructed three-dimensional model includes: slicing the reconstructed three-dimensional model to obtain multiple point cloud slice sections; calculating distance data between body size feature points based on the point cloud slice sections; and obtaining the body size data of the target animal based on the distance data between the body size feature points.

[0016] In the above implementation process, through the slicing processing method, detailed cross-sectional data of different parts of the target animal can be obtained, thereby reducing the error caused by human measurement and improving the measurement accuracy and efficiency; and the point cloud slicing can capture the fine features of the animal's body surface, which can improve the accuracy of the measured body size data. In addition, the slicing processing method is suitable for animals of different sizes, which improves the versatility of the animal body size measurement method.

[0017] Optionally, in an embodiment of the present application, the point cloud posture of the target animal corresponding to the preliminary point cloud data is corrected to obtain the corrected point cloud data, including: determining key data points in the preliminary point cloud data; wherein the key data points are located on the rotation axis of the target animal; based on the key data points, determining the point cloud correction angle of the target animal; and performing rotational normalization processing on the point cloud posture of the target animal based on the point cloud correction angle to obtain the corrected point cloud data.

[0018] In the above implementation process, by performing rotation normalization processing on the point cloud posture of the target animal, the measurement error caused by the displacement of the image sensor can be reduced, and the accuracy of the measured body size data can be improved.

[0019] Optionally, in an embodiment of the present application, the corrected point cloud data includes a target point cloud and at least one group of source point clouds; the point cloud reconstruction based on the corrected point cloud data to obtain the reconstructed three-dimensional model of the target animal includes: coarsely aligning the source point cloud and the target point cloud to obtain coarsely registered point cloud data; and finely aligning the coarsely registered point cloud data to obtain the reconstructed three-dimensional model of the target animal.

[0020] In the above implementation process, the combination of coarse registration and fine registration can effectively avoid falling into local extreme values ​​during the iteration process, thereby improving the reconstruction efficiency of the three-dimensional model while ensuring the accuracy of the three-dimensional model reconstruction.

[0021] Optionally, in an embodiment of the present application, the coarse alignment of the source point cloud and the target point cloud to obtain coarsely aligned point cloud data includes: determining a group of non-collinear corresponding point pairs from the corrected point cloud data; wherein the non-collinear corresponding point pairs include: non-collinear source point pairs and non-collinear target point pairs corresponding to the non-collinear source point pairs; solving the transformation matrix between the non-collinear source point pairs and the non-collinear target point pairs; calculating the mean square error value between the source point cloud and the target point cloud corresponding to the transformation matrix; when the mean square error value does not meet the preset error condition, re-determining a new group of non-collinear corresponding point pairs from the corrected point cloud data; when the mean square error value meets the preset error condition, based on the transformation matrix, coarsely aligning the source point cloud and the target point cloud to obtain the coarsely aligned point cloud data.

[0022] In the above implementation process, through coarse registration, a good initial position estimation can be provided for the subsequent fine registration, so as to establish the correct preliminary point cloud correspondence in the iterative process, avoid subsequent iterations falling into local extreme values, and improve the overall registration accuracy.

[0023] Optionally, in an embodiment of the present application, the coarse registration point cloud data includes a target point cloud and a registration point cloud corresponding to the source point cloud; the source point cloud includes a plurality of the source data points;

[0024] The method of finely registering the coarsely registered point cloud data to obtain the reconstructed three-dimensional model of the target animal includes: determining a target source data point from the source point cloud; determining a corresponding finely registered point cloud based on the registered data point corresponding to the target source data point; wherein the finely registered point cloud includes the registered data point and neighboring data points in the registered point cloud whose distance to the registered data point is less than or equal to a preset distance threshold; solving the initial registration matrix between the target source data point and all data points in the finely registered point cloud; determining an intermediate registration matrix based on the distance value between the registered point cloud and the target point cloud corresponding to each of the initial registration matrices; if the intermediate registration matrix does not meet the preset registration condition, re-determining a new target source data point from the source point cloud; if the intermediate registration matrix meets the preset registration condition, determining the intermediate registration matrix as the fine registration matrix; based on the fine registration matrix, finely registering the source point cloud and the target point cloud to obtain the reconstructed three-dimensional model of the target animal.

[0025] Optionally, in an embodiment of the present application, the depth image data includes multiple pixel coordinates and depth data of each of the pixel coordinates; the processing of the depth image data to obtain preliminary point cloud data includes: based on the intrinsic parameters and extrinsic parameters of the depth camera, determining a first transformation relationship between the camera coordinate system and the depth image coordinate system and a second transformation relationship between the world coordinate system and the camera coordinate system; based on the first transformation relationship and the second transformation relationship, determining the world coordinates corresponding to the pixel coordinates; and obtaining the preliminary point cloud data based on the depth data corresponding to the world coordinates.

[0026] In the above implementation process, through the first transformation relationship between the camera coordinate system and the depth image coordinate system and the second transformation relationship between the world coordinate system and the camera coordinate system, the world coordinates corresponding to the pixel coordinates can be determined, and then the corresponding preliminary point cloud data can be obtained based on the depth data corresponding to the world coordinates, providing reliable data support for the reconstruction of the three-dimensional model of the target animal.

[0027] Optionally, in an embodiment of the present application, point cloud reconstruction is performed based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal, including: preprocessing the corrected point cloud data to obtain preprocessed point cloud data, including: using pass-through filtering and RANSAC algorithm to remove noise data in the corrected point cloud data, using a cubic spline interpolation algorithm to patch missing data in the corrected point cloud data, and using a conditional-voxel filtering box to downsample the corrected point cloud data to obtain the preprocessed point cloud data; wherein the conditional-voxel filtering box includes a conditional filter and a voxel filter; point cloud reconstruction is performed based on the preprocessed point cloud data to obtain a reconstructed three-dimensional model of the target animal.

[0028] In the above implementation process, the noise data in the corrected point cloud data can be removed by adopting the straight-through filtering and RANSAC algorithm, thereby improving the quality of the point cloud data and the accuracy and reliability of the point cloud data. The missing data in the corrected point cloud data can be repaired by using the cubic spline difference algorithm to improve the continuity and integrity of the data; and the isolated points caused by noise can be filtered out by downsampling the corrected point cloud data through the conditional-voxel filter box, further reducing the influence of the noise data on the reconstructed three-dimensional model, so as to improve the accuracy of the measured body size data.

[0029] In a second aspect, an embodiment of the present application provides an animal body measurement system, the system comprising: a data processing device and a plurality of depth cameras; wherein the plurality of depth cameras are configured to be installed at least two different positions of a limiting fence;

[0030] The depth camera is configured to obtain depth image data of the target animal at least two different positions when the target animal passes through the limiting fence;

[0031] The data processing device is configured to process the depth image data to obtain preliminary point cloud data; correct the point cloud posture of the target animal corresponding to the preliminary point cloud data to obtain corrected point cloud data; reconstruct the point cloud based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal; and obtain the body size data of the target animal based on the reconstructed three-dimensional model.

[0032] Optionally, in an embodiment of the present application, the system further includes: an identification device installed in front and rear positions of the limit fence, the identification device being configured to identify tags on the target animal passing through the limit fence; the depth camera being specifically configured to obtain depth image data of the target animal in at least two different positions when the identification device identifies that the target animal has passed through the limit fence.

[0033] The beneficial effects of the present application include at least: the animal body size measurement method can improve the accuracy of the body size data obtained based on the reconstructed three-dimensional model by correcting the point cloud position of the target animal and reconstructing the point cloud based on the corrected point cloud data. And by correcting the point cloud position of the target animal, it can reduce the "measurement deviation caused by the easy displacement of the image sensor or the bracket used to fix the image sensor", solving the technical problem of low measurement precision and accuracy of the existing animal body size data method. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A schematic diagram of a flow chart of a method for measuring animal body size provided in an embodiment of the present application;

[0036] Figure 2 A schematic diagram of the structure of an animal body size measurement system provided in an embodiment of the present application;

[0037] Figure 3 A depth image and point cloud outline schematic diagram of a target animal provided in an embodiment of the present application;

[0038] Figure 4 A schematic diagram of the structure of another animal body size measurement system provided in an embodiment of the present application;

[0039] Figure 5 A schematic diagram of slicing a reconstructed three-dimensional model provided in an embodiment of the present application;

[0040] Figure 6 A schematic diagram of the point cloud posture correction effect provided in an embodiment of the present application;

[0041] Figure 7 A schematic diagram of a flow chart of a rough registration method provided in an embodiment of the present application;

[0042] Figure 8 A schematic diagram of a flow chart of a precise registration method provided in an embodiment of the present application;

[0043] Fig. 9 A schematic diagram of the three-dimensional model reconstruction effect provided in an embodiment of the present application;

[0044] Fig.10 A schematic diagram of a distortion-corrected image provided in an embodiment of the present application;

[0045] Fig.11 A schematic diagram of the preprocessing effect of correcting point cloud data provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] The following embodiments of the technical solution of the present application are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present application, and are therefore only used as examples, and cannot be used to limit the scope of protection of the present application.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0048] In the description of the embodiments of the present application, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0049] See also Figure 1 The flowchart of an animal body measurement method provided by the embodiment of the present application is shown. Figure 2 , Figure 2 The present invention provides a schematic diagram of the structure of an animal body measurement system 20. The animal body measurement system 20 includes a data processing device 201 and a plurality of depth cameras 202, wherein the plurality of depth cameras 202 are configured to be installed at least at two different positions of the limiting fence. Figure 2 It shows the situation that the depth camera 202 is installed in the upper and lower positions and the left and right positions of the limit bar respectively.

[0050] The data processing device 201 may be implemented by a microcontroller unit (MCU), a single-chip microcomputer or a field-programmable gate array (FPGA). The depth camera may be implemented by a TOF (Time of Flight) camera (e.g., Kinect v2 depth camera), a binocular camera or a structured light camera. The shape of the limit bar may be a trapezoidal limit bar, or a rectangle or other reasonable shapes.

[0051] The animal body size measurement method is applied to the animal body size measurement system 20; the animal body size measurement method may include the following steps:

[0052] S101, when a target animal passes through the limiting fence, acquiring depth image data of the target animal in at least two different directions based on the depth camera;

[0053] Based on the data processing device:

[0054] S102, processing the depth image data to obtain preliminary point cloud data;

[0055] S103, correcting the point cloud posture of the target animal corresponding to the preliminary point cloud data to obtain corrected point cloud data;

[0056] S104, performing point cloud reconstruction based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal;

[0057] S105. Acquire body size data of the target animal based on the reconstructed three-dimensional model.

[0058] Wherein, in step 101, the target animal may be a cow, a sheep or a pig, etc. The two different orientations may be an up-down orientation and a left-right orientation, or an up-down orientation and a front-back orientation. Depth image data in two different orientations may be obtained based on the depth camera, or depth image data in three different orientations may be obtained (that is, depth image data in the up-down orientation, the left-right orientation, and the front-back orientation are simultaneously obtained). The number of depth cameras may be two, three or other reasonable values. When the number of depth cameras is three, the three depth cameras may be installed at three different orientations of the limit fence respectively; two of the depth cameras may also be installed at the same orientation, and the other depth camera may be installed at another orientation; this application does not make specific limitations on this.

[0059] In step 102, the depth image data may include a depth image and a pixel value of each pixel in the depth image; or may include a plurality of pixel coordinates and depth data of each of the pixel coordinates. In the case where the depth image data includes a plurality of pixel coordinates and depth data of each of the pixel coordinates, the world coordinate corresponding to each pixel coordinate may be determined based on the intrinsic and extrinsic parameters of the depth camera; and then the preliminary point cloud data may be obtained based on the depth data corresponding to the world coordinate.

[0060] Please refer to Figure 3 , Figure 3 A depth image and point cloud outline diagram of a target animal provided in an embodiment of the present application. Figure 3 Taking the target animal as a sheep as an example, the depth image data at three different directions obtained by the depth camera and the corresponding point cloud outline schematic diagram are shown. Specifically, Figure 3Shown on the left are depth image data based on three different orientations obtained by the depth camera (the top is the depth image obtained by the depth camera based on the left and right orientation, the middle is the depth image obtained by the depth camera based on the up and down orientation, and the bottom is the depth image obtained by the depth camera based on the front and back orientation), and shown on the right are schematic diagrams of point cloud contours corresponding to the depth images in the corresponding orientations.

[0061] Among them, in step 103, the point cloud posture of the target animal can be corrected by using coordinate transformation or point cloud registration method to obtain corrected point cloud data; so as to reduce "the measurement deviation caused by the displacement of the image sensor or the bracket used to fix the image sensor". For example, the key data points in the preliminary point cloud data can be determined first; then based on the key data points, the point cloud correction angle of the target animal is determined; based on the point cloud correction angle, the point cloud posture of the target animal is rotated to obtain corrected point cloud data.

[0062] Among them, in step 104, before performing point cloud reconstruction, the corrected point cloud data can be preprocessed by using methods such as filtering, standardization, and normalization, and then the preprocessed point cloud data is reconstructed to obtain a reconstructed three-dimensional model.

[0063] In step 105, the body size data of the target animal may include body height (the vertical distance from the ground to a specific body part of the animal, usually used to evaluate the height of the animal), body length (the straight-line distance from a starting part to another ending part of the animal's body, usually excluding the length of the head, tail and limbs), chest circumference (the circumference around the widest part of the animal's chest, often used to measure the size of the animal), neck circumference (the circumference around the thinnest part of the animal's neck, used to evaluate the size of the neck), leg distance (the distance between the two hind legs, used to evaluate the leg structure and space occupancy of the animal), sleeve length (for certain specific animals, such as pet dogs, the length from the shoulder to the leg, used to evaluate the animal's body shape and proportion), chest width, and rump height and other body size parameters. Reconstructing a three-dimensional model can be used to measure the body size data of the target animal. Specifically, body size data can be measured through interactive measurement methods (i.e., the user manually marks the key parts of the animal); or through automatic recognition methods, using machine learning or deep learning algorithms to automatically identify feature points in the reconstructed three-dimensional model and calculate the corresponding body size data to ensure the accuracy and efficiency of the measurement.

[0064] It can be seen that the animal body size measurement method provided in the embodiment of the present application can improve the accuracy of the body size data obtained based on the reconstructed three-dimensional model by correcting the point cloud position and posture of the target animal and reconstructing the point cloud based on the corrected point cloud data. And by correcting the point cloud position and posture of the target animal, it is possible to reduce "the measurement deviation caused by the easy displacement of the image sensor or the bracket used to fix the image sensor", solving the technical problem of low measurement precision and accuracy of the existing animal body size data method.

[0065] Please refer to Figure 4 , Figure 4 A schematic structural diagram of another animal body size measurement system 20 provided in an embodiment of the present application (the data processing device 201 is not shown). Figure 4 It shows that the target animal is a sheep, and the depth cameras 202 are respectively installed in the upper and lower positions, left and right positions, and front and back positions of the limiting fence.

[0066] In some optional embodiments, the animal body size measurement system 20 also includes: an identification device 203 installed in the front and rear positions of the limiting fence, and the identification device 203 is configured to identify the tag on the target animal passing through the limiting fence; S101, when the target animal passes through the limiting fence, obtaining depth image data of the target animal in at least two different positions based on the depth camera, including: when the identification device recognizes that the target animal passes through the limiting fence, obtaining depth image data of the target animal in at least two different positions based on the depth camera.

[0067] Among them, the identification device 203 can be implemented by a wireless electronic identifier, an ear tag reader / writer, an RFID electronic reader (for example, the JY-L81 series, which uses a low-frequency 134.2kHz operating frequency, has a synchronization function, strong penetration, high stability, and supports long-distance reading of multiple electronic tags). The specific installation location and implementation method of the identification device 203 can be adjusted according to the type of tag and the tag position set on the target animal. The number of identification devices 203 can be 2, or 4 or other reasonable values. In the case where the number of identification devices 203 is 2, the two identification devices 203 can be installed in front of, behind, and behind the limit fence, respectively. Figure 4As shown, when the number of identification devices 203 is 4, two identification devices 203 can be installed in front of and behind the limiting fence respectively (and the two identification devices 203 in front or behind can be installed on both sides of the limiting fence respectively). When the identification device does not identify that the target animal has passed through the limiting fence, the depth camera can enter a standby state; when the identification device identifies that the target animal has passed through the limiting fence, the depth camera can enter a working state and obtain depth image data of the target animal in at least two different directions. Since the identification device can identify the tag on the target animal passing through the limiting fence, the depth image data of the corresponding target animal can be obtained based on the depth camera when the target animal passes through the limiting fence, and the obtained depth image data is associated with the tag on the target animal, thereby achieving an accurate correspondence between the depth image data and the target animal, thereby improving the accuracy of the obtained body size data.

[0068] In some optional embodiments, S105, based on the reconstructed three-dimensional model, obtaining the body size data of the target animal, includes: slicing the reconstructed three-dimensional model to obtain multiple point cloud slice sections; based on the point cloud slice sections, calculating the distance data between body size feature points; and obtaining the body size data of the target animal according to the distance data between the body size feature points.

[0069] Among them, based on the point cloud slicing algorithm in Open3D, the appropriate slicing direction and position can be selected to cut the point cloud data in the reconstructed 3D model into multiple sections (the thickness and spacing of the slices can be adjusted according to the actual measurement requirements), and in each slice, the body size feature points of the target animal are identified, and the body size measurement is achieved by calculating the distance between the body size feature points. Body size feature points can include specific bones (skull, coccyx position) or joints and other feature points related to body size data. Through the slicing processing method, detailed cross-sectional data of different parts of the target animal can be obtained, thereby reducing the errors caused by human measurement and improving the measurement accuracy and efficiency; and the point cloud slicing can capture the fine features of the animal's body surface, which can improve the accuracy of the measured body size data. In addition, the slicing processing method is suitable for animals of different sizes, which improves the versatility of the animal body size measurement method.

[0070] Please refer to Figure 5 , Figure 5 A schematic diagram of slicing a reconstructed three-dimensional model provided in an embodiment of the present application. Figure 5 Taking sheep as the target animal as an example, a schematic diagram of the slice processing results of the sheep's body length data, hind leg height data, chest circumference data, front leg height data, chest width data and neck circumference data is specifically shown; among which, the first row from left to right is the body length data, hind leg height data and chest circumference data, and the second row from left to right is the front leg height data, chest width data and neck circumference data.

[0071] In some optional embodiments, S103, correcting the point cloud posture of the target animal corresponding to the preliminary point cloud data to obtain corrected point cloud data, including: determining key data points in the preliminary point cloud data; wherein the key data points are located on the rotation axis of the target animal; based on the key data points, determining the point cloud correction angle of the target animal; performing rotation normalization processing on the point cloud posture of the target animal based on the point cloud correction angle to obtain the corrected point cloud data.

[0072] Please refer to Figure 6 , Figure 6 A schematic diagram of the point cloud posture correction effect provided in an embodiment of the present application. Figure 6 This is a schematic diagram of the effect, taking the target animal as a sheep. Figure 6 As shown, the rotation axis of the target animal can be an axial line from the head to the tail of the target animal. Accordingly, the key data points can include the head data point and tail data points The point cloud correction angle of the target animal can be calculated based on the head data points and tail data points , and use the point cloud to correct the angle , construct the rotation matrix ; and based on the constructed rotation matrix The point cloud pose of the target animal is rotated and normalized to obtain the corrected point cloud data. For example, based on Calculate the corrected point cloud data points after rotation normalization The corrected point cloud data includes multiple corrected point cloud data points. After obtaining the corrected point cloud data, it can also be compared with the standard point cloud data to verify the correction effect of the rotation normalization operation. By performing rotation normalization on the point cloud posture of the target animal, the measurement error caused by the displacement of the image sensor can be reduced, and the accuracy of the measured body size data can be improved.

[0073] In some optional embodiments, the corrected point cloud data includes a target point cloud and at least one group of source point clouds; S104, performing point cloud reconstruction based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal, including: performing coarse alignment of the source point cloud and the target point cloud to obtain coarsely aligned point cloud data; performing fine alignment of the coarsely aligned point cloud data to obtain a reconstructed three-dimensional model of the target animal.

[0074] Among them, in the process of coarse registration, the main direction fitting algorithm or the method based on the fast point feature histogram can be used for preliminary alignment to quickly reduce the difference between the source point cloud and the target point cloud. ‌ Based on the method of the fast point feature histogram, the fast point feature histogram features of each point in the point cloud are calculated, and these features are used to perform coarse registration of the source point cloud and the target point cloud. In the process of fine registration, the iterative closest point algorithm can be used to minimize the distance error between the source point cloud and the target point cloud to further optimize the alignment accuracy. The basic principle of the iterative closest point algorithm is to find the optimal rigid body transformation between two point clouds (which can be the source point cloud and the target point cloud after coarse registration) so that the two point clouds overlap as much as possible. By combining coarse registration and fine registration, it is possible to effectively avoid falling into local extreme values ​​during the iteration process, while ensuring the accuracy of 3D model reconstruction, and improving the reconstruction efficiency of the 3D model.

[0075] Please refer to Figure 7 , Figure 7 A schematic diagram of a flow chart of a coarse registration method provided in an embodiment of the present application.

[0076] In some optional embodiments, the coarse registration of the source point cloud and the target point cloud to obtain coarsely registered point cloud data includes: S301, determining a group of non-collinear corresponding point pairs from the corrected point cloud data; wherein the non-collinear corresponding point pairs include: non-collinear source point pairs and non-collinear target point pairs corresponding to the non-collinear source point pairs; S302, solving the transformation matrix between the non-collinear source point pairs and the non-collinear target point pairs; S303, calculating the mean square error value between the source point cloud and the target point cloud corresponding to the transformation matrix; S304, when the mean square error value does not meet the preset error condition, re-determining a new group of non-collinear corresponding point pairs from the corrected point cloud data; S305, when the mean square error value meets the preset error condition, based on the transformation matrix, coarsely aligning the source point cloud and the target point cloud to obtain the coarsely registered point cloud data.

[0077] The non-collinear corresponding point pairs may include 3 corresponding data, or may include 5 or other reasonable number of corresponding data. ,and , , As an example, correspondingly, the non-collinear source point pair can be expressed as , the non-collinear target point pair can be expressed as The transformation matrix between the non-collinear source point pair and the non-collinear target point pair can be calculated, and when the mean square error value corresponding to the transformation matrix satisfies the preset error condition, the source point cloud and the target point cloud can be roughly aligned based on the transformation matrix to obtain the roughly aligned point cloud data. For example, the singular value transformation matrix solution method can be used to solve the non-collinear source point pair. With non-collinear target point pairs The transformation matrix between the source point cloud and the target point cloud corresponding to the transformation matrix is ​​calculated, and then the mean square error value between the source point cloud and the target point cloud corresponding to the transformation matrix is ​​calculated. The preset error condition includes a preset error threshold and an iteration number threshold. When the mean square error value is less than or equal to the preset error threshold, or when the number of iterations corresponding to the mean square error value is greater than or equal to the iteration number threshold, the mean square error value is considered to meet the preset error condition. Through coarse registration, a good initial position estimate can be provided for the subsequent fine registration, so as to establish the correct preliminary point cloud correspondence in the iterative process, avoid the subsequent iterations from falling into local extreme values, and improve the overall registration accuracy.

[0078] Please refer to Figure 8 , Figure 8 A flowchart of a precise registration method provided in an embodiment of the present application.

[0079] In some optional embodiments, the coarse registration point cloud data includes a target point cloud and a registration point cloud corresponding to the source point cloud; the source point cloud includes a plurality of source data points; the fine registration of the coarse registration point cloud data to obtain a reconstructed three-dimensional model of the target animal includes: S401, determining a target source data point from the source point cloud; S402, determining a corresponding fine registration point cloud based on the registration data point corresponding to the target source data point; wherein the fine registration point cloud includes the registration data point and neighboring data points in the registration point cloud whose distance to the registration data point is less than or equal to a preset distance threshold; S403, solving the target An initial registration matrix between the source data points and all the data points in the precise registration point cloud; S404, determining an intermediate registration matrix based on the distance value between the registration point cloud and the target point cloud corresponding to each of the initial registration matrices; S405, when the intermediate registration matrix does not meet the preset registration conditions, re-determining a new target source data point from the source point cloud; S406, when the intermediate registration matrix meets the preset registration conditions, determining the intermediate registration matrix as the precise registration matrix; S407, based on the precise registration matrix, precisely registering the source point cloud and the target point cloud to obtain a reconstructed three-dimensional model of the target animal.

[0080] Among them, taking the target source data point P as an example, the registration data point corresponding to the target source data point can be determined based on the transformation matrix obtained in the rough registration process , and the neighboring data points in the registered point cloud whose distance to the registered data point is less than or equal to the preset distance threshold , neighboring data points The number may be 1, 5, 10 or other reasonable values. Solve the initial registration matrix between the target source data points and all data points in the fine registration point cloud , ;in, represents the i-th target source data point, Represents the i-th neighboring data point .based on Determine the registration point cloud corresponding to each initial registration matrix ; and calculate the registration point cloud The distance value between the point cloud and the target point cloud is the smallest. The corresponding registration matrix is ​​determined as the intermediate registration matrix. The preset registration condition can be a preset number of iterations or a preset distance threshold. When the distance value between the target point cloud and the target point cloud is less than or equal to the preset distance threshold, the registration point cloud corresponding to the intermediate registration matrix is , determined as the precise registration matrix; or when the number of iterations is greater than or equal to the preset number of iterations, the registration point cloud with the smallest distance value is The corresponding intermediate registration matrix is ​​determined as the precise registration matrix.

[0081] Please refer to Fig. 9 , Fig. 9 This is a schematic diagram of the three-dimensional model reconstruction effect provided in the embodiment of the present application. It can be understood that: Fig. 9 The following is a schematic diagram of the 3D model reconstruction effect, taking the target animal as a sheep as an example. Specifically, Fig. 9 (a), (b) and (c) in the figure respectively show the outline diagrams of point cloud data at different perspectives before the coarse registration and fine registration operations. Fig. 9 (d) in the figure shows a schematic diagram of the three-dimensional model reconstruction effect after coarse alignment and fine alignment operations.

[0082] In some optional embodiments, the depth image data includes multiple pixel coordinates and depth data of each pixel coordinate; S102, processing the depth image data to obtain preliminary point cloud data, including: based on the intrinsic parameters and extrinsic parameters of the depth camera, respectively determining a first transformation relationship between the camera coordinate system and the depth image coordinate system and a second transformation relationship between the world coordinate system and the camera coordinate system; based on the first transformation relationship and the second transformation relationship, determining the world coordinates corresponding to the pixel coordinates; and obtaining the preliminary point cloud data according to the depth data corresponding to the world coordinates.

[0083] Among them, the intrinsic parameter K of the depth camera can be expressed as ;in, , Respectively represent the focal length of the depth camera, , Respectively represent the optical center of the depth camera; the external parameters of the depth camera can be expressed as , . Define the position and direction of the camera coordinate system relative to the world coordinate system, and use the formula , from the position point in the world coordinate system Transform to the position point in the camera coordinate system Then through , from the position point in the camera coordinate system Transform to the position point in the image coordinate system . Considering the pixel size In this case, then through , from the position point in the image coordinates Transform to pixel coordinates ;in, is the position of the origin in the image coordinate system in the pixel coordinate system. , the depth data obtained by the depth camera Convert the depth data to the world coordinate system The least squares method can be used to fine-tune the extrinsic parameters of the depth camera, and the accuracy of the unified coordinate system can be verified by comparing the observations of different cameras on the same point to ensure the accuracy of the coordinate transformation between all depth cameras. , calculate the position point at each pixel coordinate and the corresponding depth value , the corresponding three-dimensional coordinates in the camera coordinate system Through the first transformation relationship between the camera coordinate system and the depth image coordinate system and the second transformation relationship between the world coordinate system and the camera coordinate system, the world coordinates corresponding to the pixel coordinates can be determined, and then the corresponding preliminary point cloud data can be obtained according to the depth data corresponding to the world coordinates, providing reliable data support for the reconstruction of the three-dimensional model of the target animal.

[0084] In some optional embodiments, determining the world coordinates corresponding to the pixel coordinates based on the first transformation relationship and the second transformation relationship includes: determining the initial image coordinates corresponding to the pixel coordinates based on the first transformation relationship; performing distortion correction on the initial image coordinates to obtain corrected image coordinates; and determining the world coordinates corresponding to the corrected image coordinates based on the second transformation relationship.

[0085] The type of distortion correction can be adjusted according to the actual image acquisition environment and the cause of distortion. Taking the distortion correction including radial distortion correction as an example, the radial distance r when not distorted is first calculated (for example, it can be the radial distance between the head and the tail of the target animal), and then based on as well as Correct radial distortion; where: represents the undistorted image coordinates, represents the distorted image coordinates, , , is the radial distortion coefficient. According to the distortion correction result, the corrected coordinates can be mapped back to the image using the remapping technique to obtain the corrected image coordinates after dedistortion.

[0086] Please refer to Fig.10 , Fig.10 A schematic diagram of a distortion-corrected image provided in an embodiment of the present application. Fig.10 The left side shows an image before radial distortion correction, and the right side shows an image after radial distortion correction.

[0087] In some optional embodiments, S104, performing point cloud reconstruction based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal, including: preprocessing the corrected point cloud data to obtain preprocessed point cloud data, including: using pass-through filtering and RANSAC algorithm to remove noise data in the corrected point cloud data, using a cubic spline interpolation algorithm to patch missing data in the corrected point cloud data, and using a conditional-voxel filtering box to downsample the corrected point cloud data to obtain the preprocessed point cloud data; wherein the conditional-voxel filtering box includes a conditional filter and a voxel filter; performing point cloud reconstruction based on the preprocessed point cloud data to obtain a reconstructed three-dimensional model of the target animal.

[0088] Among them, the straight-through filter can remove the main background noise. You can first determine the dimension to be filtered (can be x, y or z) and the corresponding minimum value and maximum value , and define the area of ​​interest in the point cloud data, and then traverse each data point in the area of ​​interest to check whether the data point is within the set range (i.e., whether it satisfies ), if the coordinate value of the data point is within the dimension range, the data point is retained, otherwise the data point is removed. The RANSAC algorithm can remove noise points such as the limit fence that affect the target animal. For example, for the selected model plane, the RANSAC algorithm randomly extracts a minimum sample set to fit the plane, and sets the point , and The equation of the plane can be found: ; Among them, the plane normal vector The formula can be Determine; then find out based on the normal vector . Use the fitted model parameters to calculate the distance from all points to the model and determine whether they are internal points; this can be done based on the formula Calculate the given point Distance to plane ; If the distance d is less than the set threshold, the given point is considered For each iteration, the number of interior points that meet the conditions is recorded , select the model with the largest number as the best model, and output the corresponding interior point set model and parameters.

[0089] Among them, multiple cubic polynomial functions can be used for piecewise interpolation to achieve the patching of missing data. Specifically, for the i-th interval ,exist ;in, is a multiple trinomial in the ith interval, , , , is the unknown coefficient (which can be solved by the continuity and derivative continuity of boundary conditions and smoothness conditions). The condition-voxel filter box is used to downsample the corrected point cloud data, which may include: determining the filtering range and voxel size; traversing all point clouds, dividing the point clouds into different voxel spaces, calculating the center point of the inner point, and checking whether each space meets the retention condition; retaining the point cloud that meets the condition and removing the point cloud that does not meet the condition; retaining the point cloud after double filtering, and using the voxel center point to generate a new downsampled point cloud, and using the generated new downsampled point cloud as the preprocessed point cloud data.

[0090] Please refer to Fig.11 , Fig.11 A schematic diagram of the preprocessing effect of correcting point cloud data provided in an embodiment of the present application. Fig.11 This is a schematic diagram of the pretreatment effect, taking sheep as an example. Fig.11 (a) shows a schematic diagram of the effect comparison before and after the straight-through filtering. Fig.11 (b) shows a schematic diagram of the effect comparison before and after the RANSAC algorithm is processed. Fig.11 (c) shows a schematic diagram of the effect comparison before and after the repair process using the cubic spline interpolation algorithm. Fig.11 (d) in the figure shows a schematic diagram comparing the effects before and after downsampling based on the conditional-voxel filter box.

[0091] The embodiment of the present application also provides an animal body measurement system 20, comprising: a data processing device 201 and a plurality of depth cameras 202; wherein the plurality of depth cameras 202 are configured to be installed at least two different positions of a limiting fence;

[0092] The depth camera 202 is configured to obtain depth image data of the target animal at least two different positions when the target animal passes through the limiting fence;

[0093] The data processing device 201 is configured to process the depth image data to obtain preliminary point cloud data; correct the point cloud posture of the target animal corresponding to the preliminary point cloud data to obtain corrected point cloud data; reconstruct the point cloud based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal; and obtain the body size data of the target animal based on the reconstructed three-dimensional model.

[0094] In an optional embodiment, the animal body size measurement system 20 may further include a limiting fence 204 and a limiting channel 205; the limiting channel 205 is arranged on one side of the limiting fence 204 in the direction in which the target animal enters. The limiting channel 205 can increase the probability of the target animal entering the limiting fence 204, thereby improving the efficiency of measuring the body size of the target animal.

[0095] It should be understood that the animal body measurement system 20 corresponds to the above-mentioned animal body measurement method embodiment and can execute the various steps involved in the above-mentioned method embodiment. The specific functions of the animal body measurement system 20 can be found in the above description. To avoid repetition, the detailed description is appropriately omitted here. The data processing device 201 in the animal body measurement system 20 may include at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the operating system (OS) of the device.

[0096] In several embodiments provided by the embodiments of the present application, it should be understood that the disclosed device / system and method can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a part of the module, program segment or code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0097] In addition, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part.

[0098] The above description is only an optional implementation manner of the embodiments of the present application, but the protection scope of the embodiments of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in the embodiments of the present application, which should be covered within the protection scope of the embodiments of the present application.

Claims

1. A method for measuring animal body size, characterized in that: The method is applied to an animal body measurement system including a data processing device and a plurality of depth cameras; the plurality of depth cameras are configured to be installed at at least two different positions of a limiting fence; the method comprises: When the target animal passes through the limiting fence, acquiring depth image data of the target animal at at least two different positions based on the depth camera; Based on the data processing device, the depth image data is processed to obtain preliminary point cloud data; Correcting the point cloud posture of the target animal corresponding to the preliminary point cloud data to obtain corrected point cloud data; Performing point cloud reconstruction based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal; Based on the reconstructed three-dimensional model, obtaining body size data of the target animal; Wherein, the step of obtaining the body size data of the target animal based on the reconstructed three-dimensional model comprises: slicing the reconstructed three-dimensional model to obtain a plurality of point cloud slice sections; calculating the distance data between body size feature points based on the point cloud slice sections; and obtaining the body size data of the target animal according to the distance data between the body size feature points; Wherein, the point cloud posture of the target animal corresponding to the preliminary point cloud data is corrected to obtain the corrected point cloud data, including: determining key data points in the preliminary point cloud data; wherein the key data points are located on the rotation axis of the target animal; based on the key data points, determining the point cloud correction angle of the target animal; performing rotation normalization processing on the point cloud posture of the target animal based on the point cloud correction angle to obtain the corrected point cloud data; The depth image data includes a plurality of pixel coordinates and depth data of each pixel coordinate; the processing of the depth image data to obtain preliminary point cloud data includes: determining a first transformation relationship between a camera coordinate system and a depth image coordinate system and a second transformation relationship between a world coordinate system and a camera coordinate system based on intrinsic parameters and extrinsic parameters of the depth camera; determining a world coordinate corresponding to the pixel coordinate based on the first transformation relationship and the second transformation relationship; and obtaining the preliminary point cloud data according to the depth data corresponding to the world coordinate. Among them, determining the world coordinates corresponding to the pixel coordinates based on the first transformation relationship and the second transformation relationship includes: determining the initial image coordinates corresponding to the pixel coordinates based on the first transformation relationship; performing distortion correction on the initial image coordinates to obtain corrected image coordinates; and determining the world coordinates corresponding to the corrected image coordinates based on the second transformation relationship.

2. The method according to claim 1, characterized in that in, The system further comprises: an identification device installed in front and rear positions of the limiting fence, the identification device being configured to identify a tag on the target animal passing through the limiting fence; When the target animal passes through the limiting fence, acquiring depth image data of the target animal in at least two different directions based on the depth camera includes: When the recognition device recognizes that the target animal passes through the limiting fence, depth image data of the target animal in at least two different directions is acquired based on the depth camera.

3. The method according to claim 1, characterized in that in, The modified point cloud data includes a target point cloud and at least one set of source point clouds; performing point cloud reconstruction based on the modified point cloud data to obtain a reconstructed three-dimensional model of the target animal includes: Performing coarse registration on the source point cloud and the target point cloud to obtain coarse registration point cloud data; The coarsely registered point cloud data is finely registered to obtain a reconstructed three-dimensional model of the target animal.

4. The method according to claim 3, characterized in that in, The coarsely registering the source point cloud and the target point cloud to obtain coarsely registered point cloud data includes: Determine a set of non-collinear corresponding point pairs from the corrected point cloud data; wherein the non-collinear corresponding point pairs include: a non-collinear source point pair and a non-collinear target point pair corresponding to the non-collinear source point pair; Solving the transformation matrix between the non-collinear source point pair and the non-collinear target point pair; Calculating a mean square error value between the source point cloud and the target point cloud corresponding to the transformation matrix; When the mean square error value does not satisfy a preset error condition, re-determining a new set of non-collinear corresponding point pairs from the corrected point cloud data; When the mean square error value satisfies a preset error condition, the source point cloud and the target point cloud are roughly aligned based on the transformation matrix to obtain the roughly aligned point cloud data.

5. The method according to claim 4, characterized in that in, The coarse registration point cloud data includes a target point cloud and a registration point cloud corresponding to the source point cloud; the source point cloud includes a plurality of source data points; The step of finely registering the roughly registered point cloud data to obtain a reconstructed three-dimensional model of the target animal comprises: determining a target source data point from the source point cloud; Based on the registration data points corresponding to the target source data points, a corresponding fine registration point cloud is determined; wherein the fine registration point cloud includes the registration data points and neighboring data points in the registration point cloud whose distance to the registration data points is less than or equal to a preset distance threshold; Solving the initial registration matrix between the target source data point and all data points in the fine registration point cloud; Determining an intermediate registration matrix based on the distance value between the registration point cloud and the target point cloud corresponding to each of the initial registration matrices; In the case where the intermediate registration matrix does not satisfy a preset registration condition, re-determining a new target source data point from the source point cloud; In the case where the intermediate registration matrix satisfies a preset registration condition, determining the intermediate registration matrix as a precise registration matrix; Based on the precise registration matrix, the source point cloud and the target point cloud are precisely registered to obtain a reconstructed three-dimensional model of the target animal.

6. The method according to claim 1, characterized in that in, The step of performing point cloud reconstruction based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal comprises: Preprocessing the corrected point cloud data to obtain preprocessed point cloud data, including: using a straight-through filter and a RANSAC algorithm to remove noise data in the corrected point cloud data, using a cubic spline interpolation algorithm to repair missing data in the corrected point cloud data, and using a conditional-voxel filter box to downsample the corrected point cloud data to obtain the preprocessed point cloud data; wherein the conditional-voxel filter box includes a conditional filter and a voxel filter; Point cloud reconstruction is performed based on the preprocessed point cloud data to obtain a reconstructed three-dimensional model of the target animal.

7. An animal body measurement system, characterized in that: The system comprises: a data processing device and a plurality of depth cameras; wherein the plurality of depth cameras are configured to be installed at at least two different positions of the limiting fence; The depth camera is configured to obtain depth image data of the target animal at least two different positions when the target animal passes through the limiting fence; The data processing device is configured to process the depth image data to obtain preliminary point cloud data; correct the point cloud posture of the target animal corresponding to the preliminary point cloud data to obtain corrected point cloud data; reconstruct the point cloud based on the corrected point cloud data to obtain a reconstructed three-dimensional model of the target animal; and obtain the body size data of the target animal based on the reconstructed three-dimensional model; Wherein, the step of obtaining the body size data of the target animal based on the reconstructed three-dimensional model comprises: slicing the reconstructed three-dimensional model to obtain a plurality of point cloud slice sections; calculating the distance data between body size feature points based on the point cloud slice sections; and obtaining the body size data of the target animal according to the distance data between the body size feature points; Wherein, the point cloud posture of the target animal corresponding to the preliminary point cloud data is corrected to obtain the corrected point cloud data, including: determining key data points in the preliminary point cloud data; wherein the key data points are located on the rotation axis of the target animal; based on the key data points, determining the point cloud correction angle of the target animal; performing rotation normalization processing on the point cloud posture of the target animal based on the point cloud correction angle to obtain the corrected point cloud data; The depth image data includes a plurality of pixel coordinates and depth data of each pixel coordinate; the processing of the depth image data to obtain preliminary point cloud data includes: determining a first transformation relationship between a camera coordinate system and a depth image coordinate system and a second transformation relationship between a world coordinate system and a camera coordinate system based on intrinsic parameters and extrinsic parameters of the depth camera; determining a world coordinate corresponding to the pixel coordinate based on the first transformation relationship and the second transformation relationship; and obtaining the preliminary point cloud data according to the depth data corresponding to the world coordinate. Among them, determining the world coordinates corresponding to the pixel coordinates based on the first transformation relationship and the second transformation relationship includes: determining the initial image coordinates corresponding to the pixel coordinates based on the first transformation relationship; performing distortion correction on the initial image coordinates to obtain corrected image coordinates; and determining the world coordinates corresponding to the corrected image coordinates based on the second transformation relationship.

Images