A multi-camera data fusion-based cow estrus monitoring method

Abstract

The application discloses a kind of based on multi-camera data fusion's cow estrus monitoring method, it is related to cow estrus monitoring technical field, the behavior of cow under complete shielding is visualized by multi-camera dynamic cooperation and skeleton movement deduction;Space-time slice attention mechanism combines joint dynamics calculation, and the target action is deduced from adjacent cow movement, completely solve the problem of misjudgment caused by traditional vision system due to shielding, significantly improve the monitoring reliability in dense cow group;The application establishes the difference filtering model of lameness gait and estrus behavior;Based on acceleration signal time-frequency feature and duration determination, dynamically inhibit false alarm signal, overcome the inherent defects of contact sensor under special physiological state, reduce artificial review cost.

Description

Technical Field

[0001] This invention relates to the field of cattle estrus monitoring technology, and in particular to a method for monitoring cattle estrus based on multi-camera data fusion. Background Technology

[0002] As dairy farming becomes more intensive, large-scale farms with more than 1,000 cows have become the mainstream. These farms generally adopt high-density feeding models, and when the herds gather in the feeding and resting areas, the distance between individuals is often less than 0.5 meters. Mounting behavior of cows in estrus often occurs in these areas. Current mainstream solutions rely on multi-camera vision systems or wearable sensors to replace manual observation with automation technology, thereby improving the efficiency and timeliness of estrus detection.

[0003] Common monitoring solutions include introducing 3D pose estimation and multi-view fusion algorithms, such as bovine keypoint detection based on YOLOv8. However, when a cow in estrus is completely occluded by multiple cows, the visual system loses the target pose information, and the algorithm misjudges the mounting action as standing or squeezing. In addition, existing systems use a strategy of switching views in turn, which takes more than 2 seconds from detecting occlusion to calling the backup view, missing the key frames of the mounting behavior. Furthermore, the accelerometers worn on the neck in some solutions are affected by collisions with neighboring cows in dense groups, generating mounting-like signal noise.

[0004] To address the aforementioned issues, some solutions increase camera density by adding one overhead camera every 50 square meters in the cattle shed, but the roof supports create new blind spots. Other solutions infer the occlusion relationship by the distance between the cattle, but the positioning error is greater than 15cm, which can amplify misjudgments in dense scenes. It is evident that these solutions only partially reduce the false negative rate and have not yet resolved the core contradiction of behavioral inference under complete occlusion. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides a method for monitoring estrus in cattle based on multi-camera data fusion, which solves the problem of high false alarm rates in existing monitoring technologies due to obstruction, interference, and delay in high-density farming scenarios.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides a method for monitoring estrus in cattle based on multi-camera data fusion, comprising:

[0009] Step S1: Deploy at least four sets of wide-angle cameras on the roof of the cowshed. Each set includes a visible light lens and a thermal imaging lens. Each lens is connected to an edge computing node via a servo gimbal.

[0010] Step S2: The edge computing node acquires multiple video streams in real time and constructs a three-dimensional skeletal model database of cattle based on the back pattern and body contour features.

[0011] Step S3: When the visual occlusion area of ​​the target cow exceeds a preset threshold, the gimbals of the adjacent camera groups are scheduled to perform viewpoint compensation and extract the motion vector data of the adjacent cows within the occlusion area.

[0012] Step S4: Based on the motion vector data, deduce the joint displacement dynamics of the occluded cattle and generate a virtual skeleton motion sequence;

[0013] Step S5: If the virtual skeleton motion sequence detects a correlation between hind limb push-off and neck extension, it is marked as a suspected climbing event.

[0014] Step S6: Fuse visible light texture and thermal imaging temperature field change data to perform multispectral verification of suspected climbing events;

[0015] Step S7: Output the estrus behavior determination result to the ranch management system.

[0016] As a preferred embodiment of the cattle estrus monitoring method based on multi-camera data fusion described in this invention, step S3 includes:

[0017] The gimbal is scheduled to perform horizontal rotation compensation and pitch angle adjustment. The horizontal rotation range is ±30°, and the pitch angle adjustment is 10°-15°.

[0018] Simultaneously, the thermal imaging lens is activated to assist in positioning, using temperature field change data to aid in spatial positioning.

[0019] As a preferred embodiment of the cattle estrus monitoring method based on multi-camera data fusion described in this invention, step S4 includes:

[0020] The spatiotemporal slice attention mechanism is used to analyze the movement trajectory of adjacent cattle, and the dynamic displacement of the skeleton of the occluded cattle is inferred based on the spatial topology of the cattle herd.

[0021] The dynamics of joint displacement include the change in angular velocity of the hind limb joints and the direction of the shift of the trunk's center of gravity.

[0022] As a preferred embodiment of the bovine estrus monitoring method based on multi-camera data fusion described in this invention, the deduction of joint displacement dynamics specifically includes:

[0023] a) Extract the displacement acceleration of adjacent cow knee joints;

[0024] b) Calculate the force transmission coefficient of the obscured cattle based on the average spacing of the herd;

[0025] c) Generate the hind limb joint motion trajectory envelope by combining spatial topological constraints.

[0026] As a preferred embodiment of the cattle estrus monitoring method based on multi-camera data fusion described in this invention, in step S4, the calculation of the hind limb joint angular velocity change involves extracting the coordinates of the hip-knee-ankle points of the hind limb from the three-dimensional skeletal model and performing the following calculations sequentially:

[0027] Calculate the joint angle:

[0028] ,

[0029] in, Indicates time The angle between the hind limb joints at all times. The vector representing the distance from the hip joint to the knee joint is... calculate, The vector representing the distance from the knee joint to the ankle joint is... calculate, Indicates the time of the cow's hip joint spatial coordinates, This indicates the knee joint of a cow at a given time. spatial coordinates, Indicates the time of the cow's ankle joint spatial coordinates, The Euclidean norm of a vector;

[0030] Calculate the instantaneous angular velocity.

[0031] ,

[0032] in, Indicates time The instantaneous angular velocity of the hind limb joint at any given moment. Indicates the previous frame time. The joint angle, In radians, The unit is radians per second. This indicates the time interval between two adjacent frame samples, in seconds.

[0033] Calculate the change in angular velocity:

[0034] ,

[0035] in, Indicates time The change in angular velocity of the hind limb joints at any given time. Indicates the instantaneous angular velocity of the previous frame;

[0036] Exponential smoothing filter:

[0037] ,

[0038] in, This represents the change in angular velocity after smoothing. This is an exponential smoothing coefficient, dimensionless, ranging from 0.2 to 0.3, used to balance response speed and noise suppression. It represents the smoothed change in angular velocity after the previous frame.

[0039] Smooth the results The hind limb dynamic features are input into the virtual skeleton motion sequence generation module.

[0040] As a preferred embodiment of the cattle estrus monitoring method based on multi-camera data fusion described in this invention, step S5 includes:

[0041] When the accelerometer returns the climbing characteristic signal, the gait fundamental frequency of the cattle in the previous period is analyzed simultaneously.

[0042] If limping characteristics are present, a dynamic inhibition coefficient is applied to suspected climbing events. The inhibition coefficient increases with the duration of the limping characteristics.

[0043] As a preferred embodiment of the cattle estrus monitoring method based on multi-camera data fusion described in this invention, step S6 includes:

[0044] The fusion weights of visible light and thermal imaging are dynamically allocated based on the ambient light intensity.

[0045] Under strong backlight conditions, thermal imaging temperature gradient maps are used to guide edge restoration of visible light images;

[0046] The edge restoration uses a generative adversarial network, with the thermal imaging temperature field as a conditional input, to generate dust-free texture features.

[0047] The repaired image input is an improved YOLOv8 model, which incorporates a channel attention mechanism in the feature extraction layer.

[0048] As a preferred embodiment of the cattle estrus monitoring method based on multi-camera data fusion described in this invention, wherein: in step S6, at time... After measuring the ambient light intensity, perform the fusion weight calculation, including:

[0049] Measure and normalize ambient light, and assume:

[0050] ,

[0051] in, Indicates time Ambient light intensity, unit: lux This represents the raw measurement value from the camera's built-in light sensor;

[0052] Visible light weight mapping:

[0053] ,

[0054] in, This represents the unpruned visible light fusion weights. This represents the steepness coefficient of the illumination response, in units of lux⁻¹. Neutral light threshold, unit: lux.

[0055] In the formula, the steepness coefficient With neutral threshold Determined by the preset critical illumination range:

[0056] ,

[0057] in, Indicates the lowest visible light weight. Indicates the highest visible light weight. This represents the low light threshold, in lux. This represents the critical value for high light intensity, in lux.

[0058] ;

[0059] Crop visible light weights:

[0060] ,

[0061] in, This represents the visible light fusion weights after clipping. and These are operations for calculating the maximum and minimum values ​​of each element, respectively.

[0062] Calculate thermal imaging weights:

[0063] ,

[0064] in, Indicates the thermal imaging fusion weights.

[0065] Pixel-level image fusion:

[0066] ,

[0067] in, Indicates the pixel intensity of the fused image. Indicates the pixel intensity of a visible light image. This represents the normalized pixel value of a thermal imaging image. Represents pixel coordinates.

[0068] As a preferred embodiment of the cattle estrus monitoring method based on multi-camera data fusion described in this invention, the edge computing node adopts a regional processing architecture, with target detection performed by the FPGA built into the camera and the edge server running a cross-lens spatiotemporal alignment algorithm.

[0069] The spatiotemporal alignment algorithm uses a 3D skeletal model database as a reference to map multiple video streams to a unified coordinate system.

[0070] As a preferred embodiment of the cattle estrus monitoring method based on multi-camera data fusion described in this invention, the cross-camera spatiotemporal alignment algorithm includes a skeletal key point matching module, which realizes multi-view data association through the spatial position of the cattle's scapula and hip joint.

[0071] The beneficial effects of this invention are as follows: This invention visualizes the behavior of cattle under complete occlusion by using multi-camera dynamic collaboration and skeletal motion inference; the spatiotemporal slice attention mechanism combined with joint dynamics calculation infers the target action from the movement of adjacent cattle, completely solving the problem of misjudgment of climbing and straddling caused by occlusion in traditional vision systems, and significantly improving the monitoring reliability in dense cattle herds.

[0072] This invention innovatively establishes a differential filtering model for limping gait and estrus behavior; based on the time-frequency characteristics and duration determination of acceleration signals, it dynamically suppresses false alarm signals, overcoming the inherent defects of contact sensors under special physiological conditions and reducing the cost of manual verification; it introduces an illumination weighting function and thermal imaging-guided edge restoration technology to form a double insurance for enhanced environmental robustness; through logistic mapping and normalization fusion, it automatically enhances effective information sources under strong backlight or dust conditions, ensuring that multispectral data is always in the optimal resolution state, breaking through the failure bottleneck of existing systems in extreme environments.

[0073] The regional processing architecture of this invention distributes the computational load, and the spatiotemporal alignment algorithm reduces the amount of data transmission through skeletal key point matching; the 3D model-driven processing flow avoids redundant calculations, enabling complex analysis to be completed locally in real time on the ranch, meeting the needs of large-scale applications. Attached Figure Description

[0074] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0075] Figure 1This is a flowchart illustrating the cattle estrus monitoring method based on multi-camera data fusion in Example 1. Detailed Implementation

[0076] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0077] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0078] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0079] Example 1, referring to Figure 1 This embodiment provides a method for monitoring estrus in cattle based on multi-camera data fusion, including the following steps:

[0080] Step S1: Deploy at least four sets of wide-angle cameras on the roof of the cowshed. Each set includes a visible light lens and a thermal imaging lens. Each lens is connected to an edge computing node via a servo gimbal.

[0081] Step S2: The edge computing node acquires multiple video streams in real time and constructs a three-dimensional skeletal model database of cattle based on the back pattern and body contour features.

[0082] Step S3: When the visual occlusion area of ​​the target cow exceeds a preset threshold, the gimbals of the adjacent camera groups are scheduled to perform viewpoint compensation and extract the motion vector data of the adjacent cows within the occlusion area.

[0083] Step S3 includes:

[0084] The gimbal is scheduled to perform horizontal rotation compensation and pitch angle adjustment. The horizontal rotation range is ±30°, and the pitch angle adjustment is 10°-15°.

[0085] Simultaneously activate the thermal imaging lens to assist in positioning, using temperature field change data to aid spatial positioning;

[0086] Step S4: Based on the motion vector data, deduce the joint displacement dynamics of the occluded cattle and generate a virtual skeleton motion sequence;

[0087] Step S4 includes:

[0088] The spatiotemporal slice attention mechanism is used to analyze the movement trajectory of adjacent cattle, and the dynamic displacement of the skeleton of the occluded cattle is inferred based on the spatial topology of the cattle herd.

[0089] The dynamics of joint displacement include the change in angular velocity of the hind limb joints and the direction of the shift of the trunk's center of gravity.

[0090] The dynamic deduction of joint displacement specifically includes:

[0091] a) Extract the displacement acceleration of adjacent cow knee joints;

[0092] b) Calculate the force transmission coefficient of the obscured cattle based on the average spacing of the herd;

[0093] c) Generate the hindlimb joint motion trajectory envelope by combining spatial topological constraints;

[0094] The direction of the trunk center of gravity shift is deduced through the following steps: obtain the coordinates of the scapula and hip joint from the 3D skeletal model of the obscured cattle, and calculate the initial center of gravity position; estimate the direction of the resultant force on the obscured cattle based on the force transmission model according to the motion vector data of adjacent cattle; calculate the center of gravity acceleration by combining the mass distribution of the cattle, and obtain the shift direction by time integration.

[0095] In step S4, which calculates the change in hindlimb joint angular velocity, the coordinates of the hip, knee, and ankle points of the hindlimb are extracted from the three-dimensional skeletal model, and the following calculations are performed sequentially:

[0096] Calculate the joint angle:

[0097] ,

[0098] in, Indicates time The angle between the hind limb joints at all times. The vector representing the distance from the hip joint to the knee joint is... calculate, The vector representing the distance from the knee joint to the ankle joint is... calculate, Indicates the time of the cow's hip joint spatial coordinates, This indicates the knee joint of a cow at a given time. spatial coordinates, Indicates the time of the cow's ankle joint spatial coordinates, The Euclidean norm of a vector;

[0099] Calculate the instantaneous angular velocity.

[0100] ,

[0101] in, Indicates time The instantaneous angular velocity of the hind limb joint at any given moment. Indicates the previous frame time. The joint angle, In radians, The unit is radians per second. This indicates the time interval between two adjacent frame samples, in seconds.

[0102] Calculate the change in angular velocity:

[0103] ,

[0104] in, Indicates time The change in angular velocity of the hind limb joints at any given time. Indicates the instantaneous angular velocity of the previous frame;

[0105] Exponential smoothing filter:

[0106] ,

[0107] in, This represents the change in angular velocity after smoothing. This is an exponential smoothing coefficient, dimensionless, ranging from 0.2 to 0.3, used to balance response speed and noise suppression. It represents the smoothed change in angular velocity after the previous frame.

[0108] Smooth the results The hindlimb dynamic features are input into the virtual skeleton motion sequence generation module;

[0109] Specifically, this method closely integrates with the 3D skeleton model, accurately extracts joint motion changes through vector angle and temporal difference, reduces the impact of video noise by using exponential smoothing, ensures the geometric accuracy of joint angle measurements by vector dot product and norm, highlights the acceleration and deceleration trends of motion by instantaneous difference, and effectively suppresses abrupt errors by filtering with adjustable smoothing coefficients while maintaining motion details. The overall computational load is moderate, making it suitable for real-time operation at edge nodes. It can provide stable and reliable dynamic angular velocity features without the need for additional sensors.

[0110] Step S5: If the virtual skeleton motion sequence detects a correlation between hind limb push-off and neck extension, it is marked as a suspected climbing event.

[0111] Step S5 includes:

[0112] When the accelerometer returns the climbing characteristic signal, the gait fundamental frequency of the cattle in the previous period is analyzed simultaneously.

[0113] If limping characteristics are present, a dynamic suppression coefficient is applied to suspected climbing events; the dynamic suppression coefficient increases with the duration of the limping characteristics, and the specific generation rules are as follows:

[0114] When the duration of the limpness feature is ≤ 5 minutes, the inhibition coefficient is taken as 0.1-0.3;

[0115] When the duration of the limp feature is > 5 minutes, the inhibition coefficient is based on 0.3, and increases by 0.05 per minute for each duration exceeding 5 minutes, with an upper limit of 0.6.

[0116] Step S6: Fuse visible light texture and thermal imaging temperature field change data to perform multispectral verification of suspected climbing events;

[0117] Step S6 includes:

[0118] The fusion weights of visible light and thermal imaging are dynamically allocated based on the ambient light intensity.

[0119] Under strong backlight conditions, thermal imaging temperature gradient maps are used to guide edge restoration of visible light images;

[0120] Edge restoration employs an adversarial generative network, using the thermal imaging temperature field as a conditional input to generate dust-free texture features;

[0121] The repaired image is input into an improved YOLOv8 model, which incorporates a channel attention mechanism in the feature extraction layer;

[0122] In step S6, at time... After measuring the ambient light intensity, perform the fusion weight calculation, including:

[0123] Measure and normalize ambient light, and assume:

[0124] ,

[0125] in, Indicates time Ambient light intensity, unit: lux This represents the raw measurement value from the camera's built-in light sensor;

[0126] Visible light weight mapping:

[0127] ,

[0128] in, This represents the unpruned visible light fusion weights. This represents the steepness coefficient of the illumination response, in units of lux⁻¹. Neutral light threshold, unit: lux.

[0129] In the formula, the steepness coefficient With neutral threshold Determined by the preset critical illumination range:

[0130] ,

[0131] in, Indicates the lowest visible light weight. Indicates the highest visible light weight. This represents the low light threshold, in lux. This represents the critical value for high light intensity, in lux.

[0132] Before system deployment, ambient light intensity data was collected at different locations and time periods within the cattle shed. Low and high light thresholds were determined based on the data distribution. These two thresholds served as dynamic distinguishing points; when the real-time measured light intensity fell within these ranges, it was proportionally mapped to the visible light fusion weight range. To retain sufficient visible light detail under low-light conditions, the lower limit of the visible light weight was fixed at 0.2; to avoid completely discarding thermal imaging information under extremely strong light conditions, the upper limit of the visible light weight was fixed at 0.8. During actual operation, the fusion ratio of visible light and thermal imaging was dynamically adjusted based on the relative position of real-time light intensity and these two thresholds, ensuring that both weights remained within the preset range.

[0133] ;

[0134] Crop visible light weights:

[0135] ,

[0136] in, This represents the visible light fusion weights after clipping. and These are operations for calculating the maximum and minimum values ​​of each element, respectively.

[0137] Calculate thermal imaging weights:

[0138] ,

[0139] in, Indicates the thermal imaging fusion weights.

[0140] Pixel-level image fusion:

[0141] ,

[0142] in, Indicates the pixel intensity of the fused image. Indicates the pixel intensity of a visible light image. This represents the normalized pixel value of a thermal imaging image. Represents pixel coordinates;

[0143] The specific method of thermal imaging pixel normalization is as follows: based on the range calibration of the thermal imaging camera, the original temperature value of each pixel in each frame of the image is subtracted from the lower limit of the sensor range, and then divided by the difference between the upper and lower limits of the sensor range. The result is mapped to the range of 0 to 1. This ensures that all thermal imaging pixels are located on the same normalization scale, which facilitates pixel-level fusion with visible light images with the same weight.

[0144] Specifically, this method is based on real-time illumination measurement. It uses a logistic mapping function with a threshold range to dynamically convert ambient light into fusion weights for visible light and thermal imaging. The steepness of the mapping function curve is automatically determined by the parameters within the critical illumination range, eliminating the need for manual parameter tuning. The cropping step ensures that the weights always fall within the preset minimum and maximum ranges, avoiding excessive weight bias towards a single mode due to extreme illumination. The final pixel-level fusion can not only fully utilize the advantages of thermal imaging in strong backlight or dim environments, but also leverage the high resolution advantage of visible light when there is sufficient illumination. This scheme has a simple algorithm, moderate computational load, is suitable for real-time deployment at edge nodes, and can effectively improve the quality of multispectral images.

[0145] Step S7: Output the estrus behavior determination result to the ranch management system;

[0146] The edge computing nodes adopt a regional processing architecture, with the camera's built-in FPGA performing target detection and the edge server running a cross-lens spatiotemporal alignment algorithm.

[0147] The spatiotemporal alignment algorithm uses a 3D skeleton model database as a reference to map multiple video streams to a unified coordinate system;

[0148] The cross-camera spatiotemporal alignment algorithm includes a skeletal keypoint matching module, which uses the spatial position of the cow's scapula and hip joint to achieve multi-view data association.

[0149] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring estrus in cattle based on multi-camera data fusion, characterized in that, include, Step S1: Deploy at least four sets of wide-angle cameras on the roof of the cowshed. Each set includes a visible light lens and a thermal imaging lens. Each lens is connected to an edge computing node via a servo gimbal. Step S2: The edge computing node acquires multiple video streams in real time and constructs a three-dimensional skeletal model database of cattle based on the back pattern and body contour features. Step S3: When the visual occlusion area of ​​the target cow exceeds a preset threshold, the gimbals of the adjacent camera groups are scheduled to perform viewpoint compensation and extract the motion vector data of the adjacent cows within the occlusion area. Step S4: Based on the motion vector data, deduce the joint displacement dynamics of the occluded cattle and generate a virtual skeleton motion sequence; Step S5: If the virtual skeleton motion sequence detects a correlation between hind limb push-off and neck extension, it is marked as a suspected climbing event. Step S6: Fuse visible light texture and thermal imaging temperature field change data to perform multispectral verification of suspected climbing events; Step S7: Output the estrus behavior determination result to the ranch management system.

2. The method for monitoring estrus in cattle based on multi-camera data fusion as described in claim 1, characterized in that, Step S3 includes: The gimbal is scheduled to perform horizontal rotation compensation and pitch angle adjustment. The horizontal rotation range is ±30°, and the pitch angle adjustment is 10°-15°. Simultaneously, the thermal imaging lens is activated to assist in positioning, using temperature field change data to aid in spatial positioning.

3. The method for monitoring estrus in cattle based on multi-camera data fusion as described in claim 1, characterized in that, Step S4 includes: The spatiotemporal slice attention mechanism is used to analyze the movement trajectory of adjacent cattle, and the dynamic displacement of the skeleton of the occluded cattle is inferred based on the spatial topology of the cattle herd. The dynamics of joint displacement include the change in angular velocity of the hind limb joints and the direction of the shift of the trunk's center of gravity.

4. The method for monitoring estrus in cattle based on multi-camera data fusion as described in claim 3, characterized in that, The deduction of the joint displacement dynamics specifically includes: a) Extract the displacement acceleration of adjacent cow knee joints; b) Calculate the force transmission coefficient of the obscured cattle based on the average spacing of the herd; c) Generate the hind limb joint motion trajectory envelope by combining spatial topological constraints.

5. The method for monitoring estrus in cattle based on multi-camera data fusion as described in claim 4, characterized in that, In step S4, which calculates the change in hindlimb joint angular velocity, the coordinates of the hip, knee, and ankle points of the hindlimb are extracted from the three-dimensional skeletal model, and the following calculations are performed sequentially: Calculate the joint angle: ， in, Indicates time The angle between the hind limb joints at all times. The vector representing the distance from the hip joint to the knee joint is... calculate, The vector representing the distance from the knee joint to the ankle joint is... calculate, Indicates the time of the cow's hip joint spatial coordinates, This indicates the knee joint of a cow at a given time. spatial coordinates, Indicates the time of the cow's ankle joint spatial coordinates, The Euclidean norm of a vector; Calculate the instantaneous angular velocity. ， in, Indicates time The instantaneous angular velocity of the hind limb joint at any given moment. Indicates the previous frame time. The joint angle, In radians, The unit is radians per second. This indicates the time interval between two adjacent frame samples, in seconds. Calculate the change in angular velocity: ， in, Indicates time The change in angular velocity of the hind limb joints at any given time. Indicates the instantaneous angular velocity of the previous frame; Exponential smoothing filter: ， in, This represents the change in angular velocity after smoothing. This is an exponential smoothing coefficient, dimensionless, ranging from 0.2 to 0.3, used to balance response speed and noise suppression. It represents the smoothed change in angular velocity after the previous frame. Smooth the results The hind limb dynamic features are input into the virtual skeleton motion sequence generation module.

6. The method for monitoring estrus in cattle based on multi-camera data fusion as described in claim 1, characterized in that, Step S5 includes: When the accelerometer returns the climbing characteristic signal, the gait fundamental frequency of the cattle in the previous period is analyzed simultaneously. If limping characteristics are present, a dynamic inhibition coefficient is applied to suspected climbing events. The inhibition coefficient increases with the duration of the limping characteristics.

7. The method for monitoring estrus in cattle based on multi-camera data fusion as described in claim 1, characterized in that, Step S6 includes: The fusion weights of visible light and thermal imaging are dynamically allocated based on the ambient light intensity. Under strong backlight conditions, thermal imaging temperature gradient maps are used to guide edge restoration of visible light images; The edge restoration uses a generative adversarial network, with the thermal imaging temperature field as a conditional input, to generate dust-free texture features. The repaired image input is an improved YOLOv8 model, which incorporates a channel attention mechanism in the feature extraction layer.

8. The method for monitoring estrus in cattle based on multi-camera data fusion as described in claim 7, characterized in that, In step S6, at time... After measuring the ambient light intensity, perform the fusion weight calculation, including: Measure and normalize ambient light, and assume: ， in, Indicates time Ambient light intensity, unit: lux This represents the raw measurement value from the camera's built-in light sensor; Visible light weight mapping: ， in, This represents the unpruned visible light fusion weights. This represents the steepness coefficient of the illumination response, in units of lux⁻¹. Neutral light threshold, unit: lux. In the formula, the steepness coefficient With neutral threshold Determined by the preset critical illumination range: ， in, Indicates the lowest visible light weight. Indicates the highest visible light weight. This represents the low light threshold, in lux. This represents the critical value for high light intensity, in lux. ； Crop visible light weights: ， in, This represents the visible light fusion weights after clipping. and These are operations for calculating the maximum and minimum values ​​of each element, respectively. Calculate thermal imaging weights: ， in, Indicates the thermal imaging fusion weights. Pixel-level image fusion: ， in, Indicates the pixel intensity of the fused image. Indicates the pixel intensity of a visible light image. This represents the normalized pixel value of a thermal imaging image. Represents pixel coordinates.

9. The method for monitoring estrus in cattle based on multi-camera data fusion as described in claim 1, characterized in that, The edge computing node adopts a regional processing architecture, with the camera's built-in FPGA performing target detection and the edge server running a cross-lens spatiotemporal alignment algorithm. The spatiotemporal alignment algorithm uses a 3D skeletal model database as a reference to map multiple video streams to a unified coordinate system.

10. The method for monitoring estrus in cattle based on multi-camera data fusion as described in claim 9, characterized in that, The cross-camera spatiotemporal alignment algorithm includes a skeletal keypoint matching module, which uses the spatial position of the cow's scapula and hip joint to achieve multi-view data association.

Images