Binocular camera-based cattle body size performance measurement system

Abstract

The invention relates to the field of livestock breeding, in particular to a cattle body size performance measurement system based on a binocular camera, which comprises a binocular camera acquisition module, a posture recognition and key point extraction module, a point cloud processing module, a posture correction module and a data processing and optimization module. According to the invention, the binocular camera is combined with an advanced algorithm, so that the measurement efficiency in the cattle breeding process is greatly improved; through cooperative work of multiple modules, reliable data support is provided for cattle breeding; manual measurement is not needed in the whole process, manual operation errors are avoided, the labor intensity of workers is reduced, meanwhile, stress reactions of cattle caused by manual fixing and measurement are reduced, and healthy growth of the cattle and scientization and humanization of breeding management are facilitated; the system can adapt to different light conditions, cattle body types and motion states, can stably and accurately complete body size performance measurement tasks no matter in an indoor breeding environment or in an outdoor pasture environment, and has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the field of animal husbandry, and in particular to a system for measuring the body size performance of cattle based on a binocular camera. Background Art

[0002] In the breeding industry, it is very important to regularly and accurately measure the body size performance of calves between three months and two years old. Body size performance covers many key parameters of the cattle body, such as body height, cross height, body oblique length, chest circumference, etc. These parameters directly reflect the growth and development of cattle, and have key guiding significance for the selection and breeding of breeding cattle and the formulation of breeding management strategies.

[0003] The traditional manual measurement method has many disadvantages. First of all, the operation process is cumbersome and requires the cooperation of at least two staff members to forcibly fix the cattle in the restraint frame, trying to keep the cattle upright with their heads straight and stretched forward, and then use special measuring rods and soft tapes to measure various body measurement parameters one by one. In this process, cattle often struggle and butt violently due to discomfort, and staff often need to use stop rods to maintain the position of cattle. This not only consumes a lot of manpower and time, but the measurement time of a cow can easily be more than ten minutes. It is also easy to cause stress reactions in cattle, affecting subsequent growth. At the same time, human operation errors are difficult to avoid, and the consistency of measurement accuracy cannot be ensured.

[0004] With the rapid development of science and technology, especially the increasing maturity of computer vision and neural network technology, there is an urgent need for an automated, high-precision and efficient cattle body size performance measurement solution to replace traditional manual measurement and meet the development needs of the modern cattle breeding industry.

[0005] Therefore, those skilled in the art have provided a cattle body size performance measurement system based on a binocular camera to solve the problems raised in the above background technology. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a cattle body size performance measurement system based on a binocular camera, comprising:

[0007] Binocular camera acquisition module: Two binocular cameras on the side and top are used to collect color images and point cloud data of cattle. The cameras have autofocus and adaptive dimming functions, and adopt wide-angle lens design and image stitching technology;

[0008] Posture recognition and key point extraction module: Use a general posture detection neural network based on color image input to extract key 2D key points and auxiliary key points for body measurement. The key points are used as joint points for posture detection training.

[0009] Point cloud processing module: clustering ground point clouds, calculating ground plane parameters, measuring height accordingly, projecting and calculating body oblique length, estimating chest circumference by combining measured average coefficients, and processing blank areas and noise points without 3D data;

[0010] Posture correction module: A neural network is introduced with the normalized coordinates of the joint points of the limbs facing the camera and the body size data as input, and the corrected body size data is output;

[0011] Data processing and optimization module: Use the RANSAC algorithm to remove outliers from the multi-frame body size data of the cattle throughout the entire journey, and output the final body size data.

[0012] Preferably, during the acquisition process, the binocular camera acquisition module adjusts the focal length and aperture size in real time according to the light changes in the test site and the distance between the cattle and the camera to ensure that the acquired image is clear and complete.

[0013] Preferably, the general posture detection neural network of the posture recognition and key point extraction module is pre-trained on a large number of image data sets containing cattle of different postures and body shapes.

[0014] Preferably, the point cloud processing module adopts an intelligent interpolation algorithm, based on the valid data points around the blank area or noise points, and performs interpolation and filling through weighted averaging, polynomial fitting and other methods to ensure the integrity and continuity of the point cloud data.

[0015] Preferably, the neural network of the posture correction module is pre-trained on a large number of data sets containing different movement postures of cattle and corresponding real body size data.

[0016] Preferably, according to the difference in walking speed of cattle, the data processing and optimization module uses the RANSAC algorithm to remove abnormal values ​​from about 20-50 frames of body size data generated by cattle walking through the test field, and outputs the final body size data.

[0017] Technical effects and advantages of the present invention:

[0018] Efficient and fast: Compared with the traditional manual measurement which takes more than ten minutes, the present invention uses binocular cameras combined with advanced algorithms to significantly shorten the cattle body measurement time to less than 2 seconds, greatly improving the measurement efficiency in the breeding process of breeding cattle and making large-scale and high-frequency body measurement monitoring possible.

[0019] High-precision measurement: Through the collaborative work of multiple modules, from accurate key point extraction, fine point cloud processing, posture dynamic correction to data optimization and screening, it effectively overcomes the errors caused by factors such as the non-static state of cattle and complex measurement environment, ensuring that the body measurement accuracy reaches the industry-leading level and provides reliable data support for breeding cattle.

[0020] Automated operation: No manual measurement is required throughout the process, which avoids human operation errors, reduces the labor intensity of staff, and reduces the stress response of cattle caused by manual fixation and measurement, which is beneficial to the healthy growth of cattle and the scientific and humane breeding management.

[0021] Strong adaptability: The system can adapt to different lighting conditions, cattle body shapes and movement states. Whether it is indoor breeding or outdoor pasture environment, it can stably and accurately complete the task of measuring body size performance, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a measurement point diagram of a cattle body size performance measurement system based on a binocular camera provided in an embodiment of the present application;

[0023] Figure 2 This is a real-object picture of the detection of the cattle body size performance measurement system based on a binocular camera provided in an embodiment of the present application;

[0024] Figure 3 It is a 3D point cloud image of a cattle body size performance measurement system based on a binocular camera provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are provided for the purpose of illustration and description, and are not intended to be exhaustive or to limit the present invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention, and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

[0026] Example 1

[0027] See also Figures 1 to 3 In this embodiment, a cattle body size performance measurement system based on a binocular camera is provided, including the present invention aims to provide a cattle body size performance measurement system based on a binocular camera, which effectively solves the problems existing in the traditional manual measurement of cattle body size performance and realizes fast, accurate and automatic measurement.

[0028] The cattle body size performance measurement system based on binocular camera of the present invention mainly includes the following core parts:

[0029] Binocular camera acquisition module

[0030] The system carefully arranges two binocular cameras in the test site, one placed sideways and the other placed on top. The two work together to collect all-round appearance data of cattle. With its unique visual principle, the binocular camera can construct a point cloud with the camera itself as the origin for the scene, and obtain high-resolution color images at the same time. During the acquisition process, the camera has automatic focus and adaptive dimming functions. It can adjust the focal length and aperture size in real time according to the light changes in the test site and the distance between the cattle and the camera, ensuring that the collected images are clear and complete, providing a solid data foundation for subsequent precise measurements.

[0031] In order to expand the monitoring range, the binocular camera adopts a wide-angle lens design, and through image stitching technology, multiple images from different perspectives are seamlessly stitched together to form a complete visual picture covering the entire body of the cow, avoiding missing key information due to limited field of view.

[0032] Posture recognition and key point extraction module

[0033] A deeply trained general posture detection neural network is used to optimize the needs of cattle body measurement. The network uses the collected color image as input, accurately identifies the position of the cattle in the image, and extracts a series of 2D key points that are critical to body measurement, such as the front point of the body oblique length (located at the front edge of the scapula of the cow's front leg), the apex of the cross, etc. These key points are pre-set as posture detection joint points. Through the training of a large number of image samples containing cattle with different postures and body shapes, the network has a strong feature recognition ability and accurately outputs the location information of the required key points.

[0034] In addition to the key points directly used for body size calculation, some auxiliary key points are also extracted for subsequent dynamic correction of the measured values ​​according to the real-time movement posture of the cattle, fully considering the non-static state of the cattle in the actual measurement scene and improving the reliability of the measurement results.

[0035] Point cloud processing module

[0036] Responsible for deep processing of the point cloud data generated by the binocular camera. First, accurately cluster the ground point cloud from the complex point cloud data, and calculate the precise parameters of the ground plane, such as the plane equation, normal vector, etc., through advanced plane fitting algorithms. Based on this, in response to the measurement requirements of body height and cross height, according to the ground plane parameters, accurately calculate the vertical distance from the key point to the ground to obtain accurate height data.

[0037] For the measurement of the body oblique length, the relevant key points are projected onto the xy plane of the camera coordinate system, and the distance between the two projections is calculated using geometric calculation methods to restore the true value of the body oblique length. For chest circumference measurement, the camera's field of view is limited. For the parts below the belly that are difficult to directly photograph, a highly reliable average coefficient is fitted based on a large amount of data accumulated from multiple field measurements. The distance integral operation of a series of points in the photographable area from the back to the belly is combined to accurately estimate the chest circumference.

[0038] In the face of blank areas with no 3D data due to the limited field of view of the binocular camera and noise points caused by external fence imaging, an intelligent interpolation algorithm is adopted. Based on the valid data points around the blank areas or noise points, reasonable weighted averaging, polynomial fitting and other methods are used to interpolate and fill in, ensuring the integrity and continuity of the point cloud data, providing reliable data support for subsequent precise measurement.

[0039] Posture correction module

[0040] Considering that cattle rarely present a standard standing posture during the test and are mostly in motion, while the body measurement standard requires cattle to be in a static state, a lightweight neural network is introduced for posture correction. The network takes the normalized coordinates of the joint points of the limbs of the cattle facing the camera and the body size data obtained by preliminary measurement as input, making full use of the symmetry of the cattle's limbs and reducing data redundancy. By training on a large number of samples containing different movement postures of cattle and corresponding real body size data, the network can accurately learn the inherent laws between movement posture and body size data deviation, output corrected high-precision body size data, and effectively eliminate the measurement errors caused by cattle movement.

[0041] Data processing and optimization module

[0042] To further improve the measurement accuracy, the system continuously measures the entire process of cattle walking through the test site. Depending on the difference in the walking speed of the cattle, about 20-50 frames of data are usually generated. For the body size data obtained in each frame of data, the robust RANSAC algorithm is used to remove outliers. The algorithm fits the optimal body size data model through multiple random samplings, excludes outliers that deviate too much from the model, and finally selects a set of the most representative and reliable body size data from multiple frames of data as the final output result, ensuring the high accuracy and stability of the measurement results.

[0043] Working principle:

[0044] During the system initialization phase, two binocular cameras are first firmly installed on a suitable support structure of the test site in a predetermined position, one sideways and one on top, to ensure that their field of view can fully cover the activity area where cattle may appear. The binocular cameras are accurately calibrated using a standard calibration plate to calibrate their internal parameters (such as focal length, optical center position, etc.) and external parameters (such as the rotation and translation relationship of the camera) to ensure the accuracy of point cloud generation and image acquisition.

[0045] When the cattle enter the test site, the binocular camera acquisition module is immediately started, and the side camera and the top camera synchronously collect the cattle's color image and point cloud data at a rate of 30 frames per second (adjustable according to actual needs). If the light in the test site suddenly dims, the camera's adaptive dimming function responds quickly, increasing the aperture and sensitivity to ensure that the image remains clear and bright; if the cattle approach the camera, the autofocus function accurately adjusts the focal length to keep the cattle within the clear imaging range. The collected image data is transmitted to the posture recognition and key point extraction module in real time.

[0046] After receiving the image, the posture recognition and key point extraction module inputs it into the pre-trained general posture detection neural network. Taking the measurement of body oblique length as an example, the network accurately locates the 2D coordinates of the front point of the body oblique length based on the front edge features of the scapula of the cow's front legs learned from massive image samples, and outputs the coordinates of other key measurement points including the vertex of the cross, while extracting the coordinates of some auxiliary key points for subsequent correction. These key point coordinate information is then transmitted to the point cloud processing module and the posture correction module.

[0047] The point cloud processing module quickly locates the relevant point cloud in the point cloud data generated by the binocular camera based on the received key point coordinates. For body height measurement, the ground point cloud is accurately clustered, the ground plane equation is fitted, and the vertical distance from the body height key point to the ground is calculated. If there is a blank area without 3D data due to camera field of view occlusion, the intelligent interpolation algorithm uses polynomial fitting based on the surrounding valid point cloud data to interpolate and fill in, ensuring that the point cloud data is complete for chest circumference, body oblique length and other measurement calculations, and finally obtains preliminary body size data and transmits it to the posture correction module.

[0048] The posture correction module inputs the received preliminary body size data and the normalized coordinates of the limb joints on the side of the cow facing the camera into the correction neural network. Assuming that the body height data of the cow fluctuates in a wave-like manner due to movement during walking, the network outputs the corrected accurate body height value based on the movement posture and body height deviation law learned through training. Similarly, other body size data are corrected and the corrected data is transmitted to the data processing and optimization module.

[0049] The data processing and optimization module collects 20-50 frames (depending on the speed) of body size data generated by the cattle walking through the test site, and uses the RANSAC algorithm to eliminate outliers. Through multiple random sampling and fitting of the body size data model, abnormal body size values ​​caused by the cattle's sudden and large shaking are eliminated, and finally a set of stable and accurate body size data is output to achieve rapid and accurate measurement of the cattle's body size performance.

[0050] The electrical components appearing in this article are all electrically connected to an external main controller and 220V AC power, and the main controller can be a conventional known device that controls a computer, etc. The specific implementation method of the present disclosure omits the detailed description of known functions and known components. To ensure the compatibility of the equipment, the operating methods used are consistent with the parameters of marketed equipment.

[0051] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. A cattle body size performance measurement system based on a binocular camera, characterized in that: Binocular camera acquisition module: Two binocular cameras on the side and top are used to collect color images and point cloud data of cattle. The cameras have autofocus and adaptive dimming functions, and adopt wide-angle lens design and image stitching technology; Posture recognition and key point extraction module: Use a general posture detection neural network based on color image input to extract key 2D key points and auxiliary key points for body measurement. The key points are used as joint points for posture detection training. Point cloud processing module: cluster ground point clouds, calculate ground plane parameters, measure height accordingly, calculate body oblique length by projection, estimate chest circumference by combining measured average coefficient, and process blank areas and noise points without 3D data; Posture correction module: A neural network is introduced with the normalized coordinates of the joint points of the limbs facing the camera and the body size data as input, and the corrected body size data is output; Data processing and optimization module: Use the RANSAC algorithm to remove outliers from the multi-frame body size data of the cattle throughout the entire journey, and output the final body size data.

2. The system for measuring cattle body size and performance based on binocular cameras according to claim 1, characterized in that: During the acquisition process, the binocular camera acquisition module adjusts the focal length and aperture size in real time according to the light changes in the test site and the distance between the cattle and the camera to ensure that the acquired image is clear and complete.

3. The cattle body size performance measurement system based on binocular camera according to claim 1 is characterized in that: The general posture detection neural network of the posture recognition and key point extraction module is pre-trained on a large number of image data sets containing cattle with different postures and body shapes.

4. The system for measuring cattle body size and performance based on binocular cameras according to claim 1, characterized in that: The point cloud processing module adopts an intelligent interpolation algorithm, based on the valid data points around the blank area or noise points, and performs interpolation and filling through weighted averaging, polynomial fitting and other methods to ensure the integrity and continuity of the point cloud data.

5. The system for measuring cattle body size and performance based on binocular cameras according to claim 1, characterized in that: The neural network of the posture correction module is pre-trained on a large number of data sets containing different movement postures of cattle and corresponding real body size data.

6. The system for measuring cattle body size and performance based on binocular cameras according to claim 1, characterized in that: According to the difference in walking speed of cattle, the data processing and optimization module uses the RANSAC algorithm to remove abnormal values ​​from about 20-50 frames of body size data generated by cattle walking through the test site, and outputs the final body size data.

Images