Single-visual-angle goat body size measuring method, system, equipment and medium

Abstract

The invention discloses a single-visual-angle goat body size measuring method, system, device and medium, and relates to the technical field of computer three-dimensional visual point clouds, and the method comprises the steps: obtaining goat point cloud data; dividing the goat point cloud data into a series of point cloud blocks, covering part of the point cloud blocks, and obtaining single-view point cloud blocks to be input; inputting the single-view point cloud blocks and the position features into a feature encoder of a deep neural network model, learning attention weights of the features, obtaining coarse point features and key point features through a feed-forward network, modeling the three-dimensional point cloud through global features of the single-view point cloud blocks, obtaining three-dimensional point cloud coordinates, and obtaining the three-dimensional point cloud coordinates. And performing iterative optimization by using the global features, the key point features and the coarse point features, and predicting the body size parameters of the goat. According to the method, collection of single-view point clouds in a real scene is simulated, global features, coarse features of part of point clouds and key point features are optimized, and possible damage to goats caused by a traditional method is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of computer three-dimensional visual point cloud technology, and in particular to a method, system, equipment and medium for measuring the body size of a goat from a single perspective. Background Art

[0002] Accurately acquiring livestock body measurements is the basis of precision animal husbandry. The traditional method of measuring goat body measurements is mainly manual measurement with the help of tools such as tape measures and calipers. However, such manual methods are easily affected by the subjectivity of the operator and may cause stress or harm to the goat. Therefore, in order to reduce the adverse effects on goats, modern goat body measurement research is increasingly turning to non-contact, automated measurement methods. With the rapid development of computer vision technology and the progress of three-dimensional vision on point cloud data, more and more body measurement work has begun to use point cloud data for research, and significant progress has also been made in livestock body measurement.

[0003] However, the mainstream point cloud acquisition method based on multi-view cameras can obtain complete point clouds, but the system has many deployment steps, including camera alignment and synchronization. These requirements hinder the promotion and application of methods based on multi-view camera systems to obtain livestock body size parameters. In contrast, point cloud data obtained from a single perspective does not require the integration of multiple camera systems, and is simple to operate and portable. However, single-view data lacks the completeness of data obtained by multi-view systems. Point cloud completion technology based on deep learning provides new possibilities for sheep body size measurement.

[0004] When completing the goat point cloud based on a single-view point cloud, the completion effect may be distorted, resulting in the offset of the positioning of the goat's body measurement points, which in turn causes body measurement errors. Due to the changes in the posture of different body parts, the key points may be incorrectly positioned, resulting in inaccurate body measurement. Summary of the invention

[0005] The purpose of the present invention is to provide a single-view goat body size measurement method, system, device and medium to solve the problems in the prior art in view of the above-mentioned deficiencies in the prior art.

[0006] The present invention specifically provides the following technical solution: a method for measuring the body size of a goat from a single perspective, comprising the following steps:

[0007] Collecting depth images of goats at various viewing angles, and processing the depth images to obtain point cloud data of goats;

[0008] Dividing the goat point cloud data into a series of point cloud blocks, and masking some of the point cloud blocks by online reverse projection, simulating single-view shooting, and obtaining single-view point cloud blocks to be input;

[0009] A deep neural network model constructed by the AdaPoinTr backbone network is obtained, and the single-view point cloud block is input into the deep neural network model to obtain the body size parameters of the goat, including:

[0010] Obtaining the position feature of the single-view point cloud block, and inputting the single-view point cloud block and the position feature into a feature encoder of a deep neural network model, learning the attention weight of the feature, and obtaining the coarse point feature and the key point feature through a feedforward network;

[0011] Decoding the attention weights, coarse point features, and key point features through a decoder to generate global features of the single-view point cloud block;

[0012] The 3D point cloud is modeled through the global features of the single-view point cloud block to obtain the 3D point cloud coordinates;

[0013] The 3D point cloud modeling process is iteratively optimized using global features, key point features and coarse point features. The complete 3D structure of the goat point cloud is predicted through the optimized 3D point cloud coordinates, and the body size key points are detected. The body size parameters of the goat are obtained through the complete 3D structure and body size key points.

[0014] Preferably, dividing the goat point cloud data into a series of point cloud blocks comprises the following steps:

[0015] The farthest point sampling method is used to downsample the goat point cloud data to obtain a uniformly distributed downsampled point cloud;

[0016] Each point in the uniformly distributed downsampled point cloud is regarded as the center point of each point cloud block.

[0017] Preferably, the method of masking part of the point cloud blocks by online reverse projection to simulate single-view shooting includes the following steps:

[0018] Use online back projection to project the input point cloud and discretize it into a two-dimensional grid map, and group all points in the point cloud into grid units;

[0019] The point cloud is sorted according to the distance from the point cloud origin to the projection plane, and the foresight point cloud is retained after the points to be retained are mapped back to the three-dimensional space.

[0020] Preferably, when the single-view point cloud block is input into the deep neural network model, the following steps are also included:

[0021] Different single-view point cloud blocks are used independently as input to the deep neural network model;

[0022] The local features of each single-view point cloud block are captured by the feature encoder, and the features of all single-view point cloud blocks are obtained through all local features.

[0023] Preferably, the three-dimensional point cloud is modeled by the global features of the single-view point cloud block to obtain the three-dimensional point cloud coordinates, comprising the following steps:

[0024] Determine the specific position of each single-view point cloud block through the center point of the single-view point cloud block, and determine the spatial structural relationship between each single-view point cloud block;

[0025] Based on the folding operation, the coarse point features and the features of the two-dimensional grid point set are taken as input, and the shape of the point cloud block is reconstructed by deforming the two-dimensional grid to obtain the three-dimensional point cloud block;

[0026] The key point features are used to detect the body size key points of the 3D point cloud blocks, obtain the 3D point cloud coordinates, and then all the 3D point cloud blocks with 3D point cloud coordinates are spliced ​​into a complete goat point cloud.

[0027] Preferably, predicting the complete three-dimensional structure of the goat point cloud through the optimized three-dimensional point cloud coordinates, detecting the body size key points, and obtaining the body size parameters of the goat through the complete three-dimensional structure and the body size key points, comprises the following steps:

[0028] After the key point features and coarse point features are normalized, activated and processed by the fully connected layer, the positions of n body measurement points and the three-dimensional point cloud coordinates of the body measurement points are obtained;

[0029] The body size parameters of the goat, including body length, body height, chest circumference, hip height and hip width, are obtained through the position of each body size measurement point and the three-dimensional point cloud coordinates of the body size measurement point.

[0030] Preferably, when processing the depth image, point cloud processing is performed on the depth image through point cloud registration and segmentation to obtain the required goat point cloud data.

[0031] The present invention provides a single-view goat body measurement system, comprising:

[0032] A data acquisition module is used to collect depth images of goats at various viewing angles, and process the depth images to obtain point cloud data of goats;

[0033] A simulation module, used for dividing the goat point cloud data into a series of point cloud blocks, and masking some of the point cloud blocks by online reverse projection, simulating single-view shooting, and obtaining single-view point cloud blocks to be input;

[0034] The measurement module is used to obtain a deep neural network model constructed by the AdaPoinTr backbone network, and input the single-view point cloud block into the deep neural network model to obtain the body size parameters of the goat, including:

[0035] A preliminary feature generation module, used to obtain the position features of the single-view point cloud block, and input the single-view point cloud block and the position features into the feature encoder of the deep neural network model, learn the attention weight of the features, and obtain the coarse point features and key point features through a feedforward network;

[0036] A global feature generation module, used for decoding the attention weight, coarse point feature and key point feature through a decoder to generate a global feature of the single-view point cloud block;

[0037] A modeling module is used to model the three-dimensional point cloud through the global features of the single-view point cloud block to obtain the three-dimensional point cloud coordinates;

[0038] The prediction module is used to iteratively optimize the 3D point cloud modeling process using global features, key point features and coarse point features, predict the complete 3D structure of the goat point cloud through the optimized 3D point cloud coordinates, detect body size key points, and obtain the body size parameters of the goat through the complete 3D structure and body size key points.

[0039] The present invention also provides a computer device, characterized in that it includes a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of the above-mentioned single-view goat body size measurement method.

[0040] The present invention also provides a storage medium on which a computer program is stored, characterized in that when the computer program is executed by a processor, the steps of the above-mentioned method for measuring the body size of a goat in a single-viewing angle are implemented.

[0041] Beneficial effects:

[0042] The present invention simulates the collection of single-view point clouds in real scenes, and can also be used to flexibly and efficiently obtain goat point clouds from various viewpoints, generate and optimize modeling of global features, coarse features of some point clouds and key point features, obtain coordinates of three-dimensional point clouds, and construct a complete goat point cloud through multiple three-dimensional point clouds with coordinates, thereby reducing the problem of distortion of the completion effect when completing the goat point cloud, thereby causing the positioning deviation of the goat body measurement points, and accurately positioning the key points when the postures of different body parts change, thereby greatly improving the accuracy of body measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of a method for estimating body size of a goat from a single perspective based on point cloud completion according to the present invention;

[0044] Figure 2 The improved AdaPoinTr point cloud completion and key point detection model proposed by the present invention;

[0045] Figure 3The following is a comparison diagram of the experimental effects provided by the embodiments of the present invention; wherein: Figure 3 (a1), (a2), (a3) ​​and (a4) are respectively the sheep 1-left picture, sheep 1-right picture, sheep 2-left picture and sheep 2-right picture in the input data. Figure 3 (b1), (b2), (b3) and (b4) are respectively sheep 1-left picture, sheep 1-right picture, sheep 2-left picture and sheep 2-right picture in FoldingNet. Figure 3 (c1), (c2), (c3) and (c4) are respectively the sheep 1-left picture, sheep 1-right picture, sheep 2-left picture and sheep 2-right picture in PCN. Figure 3 (d1), (d2), (d3) and (d4) are respectively sheep 1-left picture, sheep 1-right picture, sheep 2-left picture and sheep 2-right picture in TopNet. Figure 3 (e1), (e2), (e3) and (e4) are sheep 1-left picture, sheep 1-right picture, sheep 2-left picture and sheep 2-right picture in GRNet respectively. Figure 3 (f1), (f2), (f3) and (f4) are respectively sheep 1-left picture, sheep 1-right picture, sheep 2-left picture and sheep 2-right picture in PoinTr. Figure 3 (g1), (g2), (g3) and (g4) are respectively Sheep 1-left picture, Sheep 1-right picture, Sheep 2-left picture and Sheep 2-right picture in AdaPoinTr;

[0046] Figure 4 A supplementary diagram for comparing experimental results provided by an embodiment of the present invention; wherein: Figure 4 (a1), (a2), (a3) ​​and (a4) are respectively sheep 1-left picture, sheep 1-right picture, sheep 2-left picture and sheep 2-right picture in the method of the present invention, Figure 4 (b1), (b2), (b3) and (b4) are the clinic values ​​of Sheep 1-left picture, Sheep 1-right picture, Sheep 2-left picture and Sheep 2-right picture respectively. DETAILED DESCRIPTION

[0047] The following is a clear and complete description of the technical solutions of the embodiments of the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0048] like Figure 1 As shown, the present invention proposes a single-view goat body size measurement method, which specifically includes the following steps:

[0049] Step S1: collecting depth images of goats at various viewing angles, and processing the depth images to obtain point cloud data of goats.

[0050] The depth image data of the goat from each perspective is collected, and the collected data is preprocessed, including point cloud registration and point cloud segmentation steps to obtain a complete goat point cloud.

[0051] In order to obtain a more complete point cloud of goats in different growth states and sizes, a channel through which a single goat can pass is set up, and a depth camera is arranged on one side of the channel. The camera is about 0.8 meters away from the channel and about 0.5 meters high to ensure that the depth image data of the goat map is collected; in order to reduce environmental and light noise, the collection scene and channel are deployed in a scene without direct sunlight as much as possible, and the channel allows a single goat to pass through, ensuring that the depth camera only captures the data of a single goat; the data collection of the goat depth map ensures that goats of different sizes and moving states can be captured, so as to ensure the capture rate of the passing goat data and maximize the acquisition of the goat's depth image data. After conventional point cloud registration, segmentation and other operations, a complete goat point cloud is obtained.

[0052] Step S2: Divide the goat point cloud data into a series of point cloud blocks, and mask some of the point cloud blocks through online reverse projection to simulate single-view shooting and obtain the point cloud blocks to be input.

[0053] Specifically:

[0054] Step S21: dividing the point cloud data into a series of point cloud blocks, including the following steps:

[0055] The farthest point sampling method is used to downsample the goat point cloud data to obtain a uniformly distributed downsampled point cloud.

[0056] Each point in the uniformly distributed downsampled point cloud is regarded as the center point of each point cloud block.

[0057] Step S22: masking part of the point cloud blocks by online reverse projection to simulate single-view shooting, including the following steps:

[0058] Using online back projection, the input point cloud is projected and discretized into a two-dimensional grid map, and all points in the point cloud are grouped in grid units.

[0059] The point cloud is sorted according to the distance from the point cloud origin to the projection plane, and the foresight point cloud is retained after the points to be retained are mapped back to the three-dimensional space.

[0060] The farthest point sampling (FPS) method is used to obtain the downsampled point cloud of the complete goat point cloud. The online back projection proposed in the present invention is used to project and discretize the input point cloud into a two-dimensional grid map. All points in the point cloud are grouped in grid units and sorted according to the distance from the origin to the projection plane. The retained points are mapped back to the three-dimensional space to retain the "forward-looking" goat point cloud. Subsequently, different point cloud blocks are independently used as inputs to the network model, and the local features of each point cloud block are captured by the feature encoder to obtain the features of all visible point cloud blocks.

[0061] Build the AdaPoinTr backbone network, and build the point cloud completion network head and key point detection head by utilizing the Transformer encoder, decoder, multi-head attention mechanism module, and point cloud reconstruction module of the AdaPoinTr backbone network.

[0062] like Figure 2 As shown in the figure, the improved AdaPoinTr deep neural network model is trained using the single-view goat point cloud data obtained by back-projection to obtain a point cloud completion and key point detection pre-training model.

[0063] Step S3: Obtain a deep neural network model constructed by the AdaPoinTr backbone network, and input the single-view point cloud block into the deep neural network model to obtain the body size parameters of the goat. Including:

[0064] Step S31: Obtain the position features of the single-view point cloud block, and input the single-view point cloud block and the position features into the feature encoder of the deep neural network model, learn the attention weight of the features, and obtain the coarse point features and key point features through the feedforward network.

[0065] When inputting a single-view point cloud block into a deep neural network model, the following steps are also included:

[0066] Different single-view point cloud blocks are independently used as input to the deep neural network model.

[0067] The local features of each single-view point cloud block are captured by the feature encoder, and the features of all single-view point cloud blocks are obtained through all local features.

[0068] Step S32: Decode the attention weights, coarse point features, and key point features through a decoder to generate global features of a single-view point cloud block.

[0069] Step S33: Model the three-dimensional point cloud through the global features of the single-view point cloud block to obtain the three-dimensional point cloud coordinates.

[0070] Among them, the three-dimensional point cloud is modeled through the global features of the single-view point cloud block to obtain the three-dimensional point cloud coordinates; specifically:

[0071] The center point of each single-view point cloud block is used to represent the position of the current single-view point cloud block, and a trainable position embedding model is used to obtain the position features of the single-view point cloud block.

[0072] The features of the visible point cloud block and its position features are used as input. The multi-head Outlook attention module learns the features to obtain the attention weights of the features. The feedforward network is used to obtain the features of the input point cloud block encoded by the Transformer module, including coarse point features and key point features.

[0073] The Outlook attention module and the feedforward network module output the features of each point cloud block of the entire point cloud.

[0074] Point cloud position embedding uses the center point of each point cloud block to represent the position of the current point cloud block, and uses a trainable position embedding model to obtain the position features of the point cloud block; the AdaPoinTr encoder uses the features of the visible point cloud block and its position features as input, and the multi-head Outlook attention module learns the features to obtain the attention weights of the features, and uses the feedforward network to obtain the features of the input point cloud block after being encoded by the Transformer module, including coarse point features and key point features; the AdaPoinTr decoder will output the features of each point cloud block of the entire point cloud by another series of Outlook attention modules and feedforward network modules.

[0075] The point cloud center point prediction further determines the specific position of each single-view point cloud block and the spatial structural relationship between each single-view point cloud block through the center point of the single-view point cloud block; the point cloud block deformation based on the folding operation will use the rough point cloud features (coarse point features) containing only the center point and the features of the two-dimensional grid point set as input, and reconstruct the shape of the point cloud block by deforming the two-dimensional grid to obtain a three-dimensional point cloud block; the key point features are used to detect the volume key points of the three-dimensional point cloud block, obtain the three-dimensional point cloud coordinates, and splice all the three-dimensional point cloud blocks with three-dimensional point cloud coordinates into a complete point cloud.

[0076] Step S34: Iteratively optimize the three-dimensional point cloud modeling process using global features, key point features, and coarse point features, predict the complete three-dimensional structure of the goat point cloud through the optimized three-dimensional point cloud coordinates, detect body size key points, and obtain the body size parameters of the goat through the complete three-dimensional structure and body size key points.

[0077] Specifically:

[0078] After the key point features and coarse point features of the encoded input point cloud block are normalized, activated and processed by the fully connected layer, the positions of 10 body measurement points and the three-dimensional point cloud coordinates of the body measurement points are obtained.

[0079] The goat point cloud completion model was constructed to obtain the complete point cloud of the goat and the corresponding body measurement point coordinates. The body size parameters of the goat, such as body length, body height, chest circumference, hip height and hip width, were obtained through the position of each body measurement point and the three-dimensional point cloud coordinates of the body measurement point.

[0080] That is, the positions of 10 body measurement points are obtained for further body measurement calculation. The coarse point features can provide relatively more specific guidance for the positioning of key measurement points, so that the detection of key measurement points and the completion of point clouds can interact better and improve the accuracy of key point detection. At the same time, the detected key points will be fed back to the update of coarse points, so that the two tasks have better interaction and influence. Through the constructed goat point cloud completion model and measurement point detection method, the complete point cloud of the goat and the corresponding body measurement point coordinates can be obtained. Based on this result, the body size parameters of the goat, such as body length, body height, chest circumference, hip height and hip width, can be calculated.

[0081] like Figure 3 and Figure 4 As shown in the figure, it is a comparison chart of the experimental effects of the present invention. It can be seen that compared with the traditional method, the present invention selects a "non-contact, non-harmful, high-yield, and high-efficiency" deep learning method, which is more efficient than the traditional goat body size estimation method; the present invention not only realizes a method for quickly obtaining the goat body size, but also provides a direction for the estimation of the body size of related poultry and livestock, while also saving the time cost of related personnel and improving animal welfare, which is worthy of promotion.

[0082] Based on the above method, the present invention provides a single-view goat body measurement system, comprising: a data acquisition module, a simulation module and a measurement module, wherein the measurement module comprises a preliminary feature generation module, a global feature generation module, a modeling module and a prediction module.

[0083] The data acquisition module is used to collect the depth images of goats from different perspectives, and process the depth images to obtain the goat point cloud data; the simulation module is used to divide the point cloud data into a series of point cloud blocks, and mask some point cloud blocks through online reverse projection, simulate single-view shooting, and obtain the single-view point cloud blocks to be input; the measurement module is used to obtain the deep neural network model constructed by the AdaPoinTr backbone network, and input the single-view point cloud blocks into the deep neural network model to obtain the body size parameters of the goat, including:

[0084] The preliminary feature generation module is used to obtain the position features of the single-view point cloud block, and input the single-view point cloud block and the position features into the feature encoder of the deep neural network model, learn the attention weights of the features, and obtain the coarse point features and key point features through the feedforward network; the global feature generation module is used to decode the attention weights, coarse point features and key point features through the decoder to generate the global features of the single-view point cloud block; the modeling module is used to model the three-dimensional point cloud through the global features of the single-view point cloud block to obtain the three-dimensional point cloud coordinates; the prediction module is used to iteratively optimize the three-dimensional point cloud modeling process using the global features, key point features and coarse point features, predict the complete three-dimensional structure of the goat point cloud through the optimized three-dimensional point cloud coordinates, detect the body size key points, and obtain the body size parameters of the goat through the complete three-dimensional structure and body size key points.

[0085] The present invention also provides a computer device, comprising a memory and a processor. The memory stores a program. When the program is executed by the processor, the processor executes the steps of a single-view goat body size measurement method.

[0086] In accordance with the disclosed embodiments, a computing device may communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth communications, etc.), or with any device (e.g., routers, modems, etc.) that enables a computing device to communicate with one or more other computing devices.

[0087] The present invention also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, a method for predicting the remaining time of an aircraft is implemented.

[0088] According to the disclosed embodiments, the storage medium may be a non-volatile computer-readable storage medium, such as but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0089] The various embodiments of the present invention are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to in detail.

[0090] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A single-view goat body measurement method, characterized in that: include: Collecting depth images of goats at various viewing angles, and processing the depth images to obtain point cloud data of goats; Dividing the goat point cloud data into a series of point cloud blocks, and masking some of the point cloud blocks by online reverse projection, simulating single-view shooting, and obtaining single-view point cloud blocks to be input; A deep neural network model constructed by the AdaPoinTr backbone network is obtained, and the single-view point cloud block is input into the deep neural network model to obtain the body size parameters of the goat, including: Obtaining the position feature of the single-view point cloud block, and inputting the single-view point cloud block and the position feature into a feature encoder of a deep neural network model, learning the attention weight of the feature, and obtaining the coarse point feature and the key point feature through a feedforward network; Decoding the attention weights, coarse point features, and key point features through a decoder to generate global features of the single-view point cloud block; The 3D point cloud is modeled through the global features of the single-view point cloud block to obtain the 3D point cloud coordinates; The 3D point cloud modeling process is iteratively optimized using global features, key point features and coarse point features. The complete 3D structure of the goat point cloud is predicted through the optimized 3D point cloud coordinates, and the body size key points are detected. The body size parameters of the goat are obtained through the complete 3D structure and body size key points.

2. A single-view goat body measurement method as claimed in claim 1, characterized in that: Dividing the goat point cloud data into a series of point cloud blocks comprises the following steps: The farthest point sampling method is used to downsample the goat point cloud data to obtain a uniformly distributed downsampled point cloud; Each point in the uniformly distributed downsampled point cloud is regarded as the center point of each point cloud block.

3. A single-view goat body measurement method as claimed in claim 1, characterized in that: The method of masking part of the point cloud blocks by online reverse projection to simulate single-view shooting includes the following steps: Use online back projection to project the input point cloud and discretize it into a two-dimensional grid map, and group all points in the point cloud into grid units; The point cloud is sorted according to the distance from the point cloud origin to the projection plane, and the foresight point cloud is retained after the points to be retained are mapped back to the three-dimensional space.

4. A single-view goat body measurement method as claimed in claim 1, characterized in that: When the single-view point cloud block is input into the deep neural network model, the following steps are also included: Different single-view point cloud blocks are used independently as input to the deep neural network model; The local features of each single-view point cloud block are captured by the feature encoder, and the features of all single-view point cloud blocks are obtained through all local features.

5. The single-view goat body measurement method according to claim 1, characterized in that: The method of modeling the three-dimensional point cloud by using the global features of the single-view point cloud block to obtain the three-dimensional point cloud coordinates includes the following steps: Determine the specific position of each single-view point cloud block through the center point of the single-view point cloud block, and determine the spatial structural relationship between each single-view point cloud block; Based on the folding operation, the coarse point features and the features of the two-dimensional grid point set are taken as input, and the shape of the point cloud block is reconstructed by deforming the two-dimensional grid to obtain the three-dimensional point cloud block; The key point features are used to detect the body size key points of the 3D point cloud blocks, obtain the 3D point cloud coordinates, and then all the 3D point cloud blocks with 3D point cloud coordinates are spliced ​​into a complete goat point cloud.

6. A single-view goat body measurement method as claimed in claim 5, characterized in that: The method of predicting the complete three-dimensional structure of the goat point cloud through the optimized three-dimensional point cloud coordinates, detecting the body size key points, and obtaining the body size parameters of the goat through the complete three-dimensional structure and the body size key points includes the following steps: After the key point features and coarse point features are normalized, activated and processed by the fully connected layer, the positions of n body measurement points and the three-dimensional point cloud coordinates of the body measurement points are obtained; The body size parameters of the goat, including body length, body height, chest circumference, hip height and hip width, are obtained through the position of each body size measurement point and the three-dimensional point cloud coordinates of the body size measurement point.

7. A single-view goat body measurement method as claimed in claim 1, characterized in that: When processing the depth image, point cloud processing is performed on the depth image through point cloud registration and segmentation to obtain the required goat point cloud data.

8. A single-view goat body measurement system, characterized in that: include: A data acquisition module is used to collect depth images of goats at various viewing angles and process the depth images to obtain point cloud data of goats; A simulation module, used for dividing the goat point cloud data into a series of point cloud blocks, and masking some of the point cloud blocks by online reverse projection, simulating single-view shooting, and obtaining single-view point cloud blocks to be input; The measurement module is used to obtain a deep neural network model constructed by the AdaPoinTr backbone network, and input the single-view point cloud block into the deep neural network model to obtain the body size parameters of the goat, including: A preliminary feature generation module, used to obtain the position features of the single-view point cloud block, and input the single-view point cloud block and the position features into the feature encoder of the deep neural network model, learn the attention weight of the features, and obtain the coarse point features and key point features through a feedforward network; A global feature generation module, used for decoding the attention weight, coarse point feature and key point feature through a decoder to generate a global feature of the single-view point cloud block; A modeling module is used to model the three-dimensional point cloud through the global features of the single-view point cloud block to obtain the three-dimensional point cloud coordinates; The prediction module is used to iteratively optimize the 3D point cloud modeling process using global features, key point features and coarse point features, predict the complete 3D structure of the goat point cloud through the optimized 3D point cloud coordinates, detect body size key points, and obtain the body size parameters of the goat through the complete 3D structure and body size key points.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of a method for measuring body dimensions of a goat from a single perspective as claimed in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for measuring body dimensions of a goat from a single perspective as claimed in any one of claims 1 to 7 are implemented.

Images