A method and system for on-line detection of the space distribution index of the pushed feed by a pushing robot

Abstract

This invention belongs to the field of feeding robot technology, specifically relating to an online detection method and system for the spatial distribution index of feed after feeding by a feeding robot. The method includes: acquiring an image of the actual feed distribution after feeding by the feeding robot in real time; preprocessing the actual feed distribution image to obtain a preprocessed image; generating three-dimensional point cloud data from the preprocessed image; calculating the area of ​​the actual feed distribution region based on the three-dimensional point cloud data; calculating the spatial distribution index of the feed after feeding by the feeding robot based on the preset ideal feed distribution region area and the degree of overlap between the ideal feed distribution region area and the actual feed distribution region area; and judging the feeding quality of the feeding robot based on the difference between the preset ideal feed spatial distribution index and the spatial distribution index of the feed after feeding by the feeding robot. The smaller the difference, the more the distribution pattern of the feed after feeding matches the ideal state, and the higher the feeding quality.

Description

Technical Field

[0001] This invention belongs to the field of feeding robot technology, specifically relating to an online detection method and system for the spatial distribution index of feed after feeding robot push. Background Technology

[0002] During dairy cow feeding, severe rooting of feed is a common occurrence. This not only directly wastes feed but also reduces the actual feed intake and efficiency of the cows. Furthermore, if the rooted feed accumulates outside the feeding trough, it increases the difficulty of cleaning the passageway and may accelerate feed spoilage, breeding bacteria and pathogens, thus posing a potential threat to the health of the herd and the hygiene of the pasture environment.

[0003] To ensure the normal growth and milk production performance of cattle, feeding managers usually need to promptly push scattered feed back to a location easily accessible to the cattle. However, manually pushing feed has problems such as high labor intensity and high labor costs.

[0004] In response to the above situation, feeding robots are already being used in production to replace some manual operations. However, existing feeding robots often cannot guarantee the quality of feeding after operation, potentially resulting in incomplete feeding and low sweeping efficiency. If the feed is not effectively placed, it can lead to prolonged insufficient feed intake in cattle, resulting in a series of health problems such as weight loss, weakened immunity, and insufficient milk production. Therefore, it is necessary to check the feeding quality and optimize the feeding robot's performance based on the inspection results. However, currently, the inspection of feeding quality still mainly relies on manual visual judgment. This method is not only difficult to guarantee in terms of accuracy but is also greatly affected by the working environment and the experience of the personnel. It also faces limitations such as high labor costs, high labor intensity, and high requirements for adaptability to the working environment. Summary of the Invention

[0005] In view of this, the present invention provides an online detection method and system for the spatial distribution index of the feed pushed by a feeding robot, so as to solve the above-mentioned technical problems caused by the manual visual judgment of the feeding quality of the feeding robot in the prior art.

[0006] To achieve the above objectives, this application adopts the following approach:

[0007] A method for online detection of the spatial distribution index of feed after a feeding robot pushes the feed includes the following steps:

[0008] S10. Real-time acquisition of actual feed distribution images after the feeding robot pushes the feed;

[0009] S20. The actual food distribution image is preprocessed to obtain a preprocessed image;

[0010] S30. Generate three-dimensional point cloud data from the preprocessed image;

[0011] S40. Calculate the actual area of ​​the daily food distribution region based on the three-dimensional point cloud data;

[0012] S50. Based on the preset ideal diet distribution area and the degree of overlap between the ideal diet distribution area and the actual diet distribution area, calculate the diet spatial distribution index after the pusher robot pushes the food, and judge the pushing quality of the pusher robot based on the difference between the preset ideal diet spatial distribution index and the diet spatial distribution index after the pusher robot pushes the food.

[0013] Preferably, the actual food distribution image is a depth image, and step S20 includes: converting the actual food distribution image into a grayscale image, performing Gaussian filtering, denoising, image correction, and disparity calculation on the grayscale image to obtain the preprocessed image, wherein the horizontal gradient features of the image are calculated using the horizontal Sobel operator during the Gaussian filtering process.

[0014] Preferably, the SGBM local matching algorithm is used to calculate disparity to obtain the preprocessed image.

[0015] Preferably, in step S40, the step of "calculating the actual daily food distribution area based on the three-dimensional point cloud data" includes: processing the three-dimensional point cloud data, separating the daily food point cloud, and calculating the projected area of ​​the daily food point cloud on a cross-section perpendicular to the direction of travel during the unit travel time of the pushing robot, which is taken as the actual daily food distribution area.

[0016] Preferably, the ground point cloud and the food point cloud are separated using the cloth simulation filtering (CSF) algorithm, and the point cloud plane is fitted using the random sampling consensus (RANSAC) algorithm. The segmented food point cloud is then projected onto the plane to calculate the projected area.

[0017] Preferably, the calculation of the projected area specifically involves: extracting the boundary of the projected two-dimensional point set using the convex hull algorithm, and calculating the area enclosed by the convex hull.

[0018] Preferably, in step S50, the spatial distribution index of the feed after the pushing robot is pushed is calculated using the following formula:

[0019] σ= ;

[0020] In the formula, σ is the spatial distribution index of the feed after the pusher robot pushes the feed, S1 is the area of ​​the ideal feed distribution area, and S2 is the area of ​​the current actual feed distribution area.

[0021] An online detection system for the spatial distribution index of feed after a feeding robot pushes the feed is provided to implement the aforementioned online detection method for the spatial distribution index of feed after a feeding robot pushes the feed. The system includes an image acquisition unit, a processing unit, and a remote monitoring terminal. The image acquisition unit is installed behind the feeding robot and is used to capture real-time images of the feed pile after the robot pushes the feed to obtain an image of the actual feed distribution. The processing unit is communicatively connected to the image acquisition unit and is configured to:

[0022] a. Receive and preprocess the actual food distribution image and perform disparity calculation to generate three-dimensional point cloud data;

[0023] b. Calculate the actual area of ​​the daily food distribution region based on the aforementioned three-dimensional point cloud data;

[0024] c. Retrieve the area of ​​the ideal daily food distribution region that has been pre-stored;

[0025] d. Calculate the spatial distribution index of the pushed-out diet by the pushing robot based on the area of ​​the ideal diet distribution area and the area of ​​the actual diet distribution area;

[0026] The remote monitoring terminal is communicatively connected to the processing unit, and the remote monitoring terminal is used to receive and display the spatial distribution index of the daily feed after the pusher robot pushes.

[0027] Preferably, the image acquisition unit is a binocular camera.

[0028] Preferably, the processing unit is an industrial control computer.

[0029] The aforementioned online detection method and system for the spatial distribution index of feed after being pushed by the feeding robot acquires the actual feed distribution image in real time, preprocesses it, and then generates three-dimensional point cloud data based on the preprocessed image. The area of ​​the actual feed distribution region is calculated based on the three-dimensional point cloud data. The spatial distribution index of the feed after being pushed by the feeding robot is calculated based on the area of ​​the ideal feed distribution region and the degree of overlap between the ideal and actual feed distribution regions. The smaller the difference between the ideal feed spatial distribution index and the spatial distribution index of the feed after being pushed by the feeding robot, the closer the distribution pattern of the feed after being pushed is to the ideal state, and the higher the quality of the pushing. This application proposes for the first time the accurate and calculable quantitative indicator of "spatial distribution index of feed after being pushed by the feeding robot," fundamentally changing the traditional evaluation method that relies on subjective visual inspection and filling a technological gap in this field. It realizes online, real-time, and automatic detection of pushing quality. The detection process is carried out synchronously with the pushing operation, overcoming the lag of manual detection and enabling immediate detection of operational quality problems. By using the feed-pushing robot to push the feed spatial distribution index, the quality of the feed-pushing robot can be optimized, ensuring that the feed is pushed in place, improving the sweeping rate, thereby reducing the loss caused by feed being pushed out of the feeding area, and ensuring the feeding quality of dairy cows. Attached Figure Description

[0030] Figure 1 This is a flowchart of the online detection method for the spatial distribution index of the pushed-up food ration by the pusher robot in this application.

[0031] Figure 2 This is a schematic diagram of the feed distribution area behind the feeding robot when it is working in this application.

[0032] Figure 3 This is a scene depicting a dairy farm feed pile, simulating the movement of a feeding robot, as described in this application.

[0033] Figure 4 This is a hardware communication diagram of the online detection system for the spatial distribution index of the feed pusher robot in this application. Detailed Implementation

[0034] To facilitate understanding of this application, a more comprehensive description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are also given. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of this application.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0036] Please refer to Figures 1 to 3 In one specific embodiment, a method for online detection of the spatial distribution index of feed after a feeding robot pushes the feed includes the following steps:

[0037] S10. Real-time acquisition of actual feed distribution images after the feeding robot pushes the feed;

[0038] S20. The actual food distribution image is preprocessed to obtain a preprocessed image;

[0039] S30. Generate three-dimensional point cloud data from the preprocessed image;

[0040] S40. Calculate the actual area of ​​the daily food distribution region based on the three-dimensional point cloud data;

[0041] S50. Based on the preset ideal diet distribution area and the degree of overlap between the ideal diet distribution area and the actual diet distribution area, calculate the diet spatial distribution index after the pusher robot pushes the food, and judge the pushing quality of the pusher robot based on the difference between the preset ideal diet spatial distribution index and the diet spatial distribution index after the pusher robot pushes the food.

[0042] When using the method provided in this application to detect the spatial distribution index of feed after a feeding robot pushes feed online, an ideal feed distribution area, typically a rectangular cross-section, close to the cow pen and meeting the optimal feeding needs of dairy cows, is pre-measured or defined, and its area is calculated as the area of ​​the ideal feed distribution area, which can be stored in the system as a benchmark for comparison. During the operation of the feeding robot, its onboard vision sensor (such as a binocular camera) can be used to capture real-time images of the feed pile after the robot pushes feed, obtaining an image of the actual feed distribution. The acquired raw images are preprocessed to improve image quality and reduce environmental interference, resulting in a preprocessed image. Subsequently, the three-dimensional coordinates corresponding to each two-dimensional coordinate are calculated using the preprocessed image and the Q matrix, and color point clouds are generated simultaneously. The intrinsic and extrinsic parameters obtained through camera calibration are used to convert the two-dimensional feed pile image into point cloud data in three-dimensional space based on the principle of triangulation. The three-dimensional point cloud data of the feed pile is generated and processed using the OpenCV image algorithm library, the PCL point cloud algorithm library, and Cloudcompare software. After processing the 3D point cloud data, the projected area of ​​the feed pile point cloud on a cross-section perpendicular to the direction of travel of the pushing robot is calculated. This area is the actual feed distribution area. Next, based on the ideal feed distribution area and the degree of overlap between the ideal and actual feed distribution areas, the spatial distribution index of the feed after the pushing robot is calculated. The predetermined ideal feed spatial distribution index is 1, while the spatial distribution index of the feed after the pushing robot is a value between 0 and 1. The smaller the difference between the ideal feed spatial distribution index and the spatial distribution index of the feed after the pushing robot, the closer the distribution pattern of the feed after pushing is to the ideal state, and the higher the pushing quality.

[0043] Based on this, in an optional embodiment, the calculated distribution index can be fed back to the central control system of the pushing robot in real time. The control system can dynamically adjust the parameters of the pushing robot in the next work cycle, such as the downward pressure of the pushing robot's rollers, the traveling speed, and the lateral swing amplitude, according to the magnitude of the spatial distribution index value of the pushing robot and the preset strategy (e.g., if the spatial distribution index of the pushing robot is continuously lower than the threshold of 0.85), so as to achieve work optimization.

[0044] The online detection method for the spatial distribution index of feed pushed by the feeding robot provided in this application introduces for the first time an accurate and calculable quantitative indicator, the "spatial distribution index of feed pushed by the feeding robot," fundamentally changing the traditional assessment method that relies on subjective visual inspection and filling a technological gap in this field. It achieves online, real-time, and automatic detection of feeding quality. The detection process is synchronized with the feeding operation, overcoming the lag of manual detection and enabling immediate detection of operational quality problems. Furthermore, the spatial distribution index of feed pushed by the feeding robot can optimize the feeding quality of the robot, ensuring that feed is pushed in place, improving the sweeping rate, thereby reducing feed loss caused by feed being pushed out of the feeding area and ensuring the feeding quality of dairy cows.

[0045] In a preferred embodiment, the actual food distribution image is a depth image (a color image with depth information acquired by a binocular camera). Step S20 includes: converting the actual food distribution image into a grayscale image, performing Gaussian filtering, denoising, image correction, and disparity calculation on the grayscale image to obtain the preprocessed image. During the Gaussian filtering, a horizontal Sobel operator is used to calculate the horizontal gradient features of the image. The Sobel operator calculation formula is as follows:

[0046] ;

[0047] In the formula, Sobel(x,y) represents the gradient value at (x,y), and P(x,y) represents the pixel gray value at (x,y).

[0048] Then, a mapping function is used to map each pixel in the image processed by the horizontal Sobel operator to a new image. The formula for the mapping function is as follows:

[0049] ;

[0050] In the formula, P represents the gradient value after processing by the horizontal Sobel operator. This represents the grayscale value of the new image pixels after mapping. This represents a constant parameter that controls the threshold of pixel grayscale values. The default value is 15. The resulting new image will be used for subsequent cost calculations.

[0051] Disparity calculation includes generating a depth map from the corrected image and visualizing it. The depth calculation formula is as follows:

[0052] ;

[0053] In the formula, z is the depth, b is the optical center distance (baseline length) between the binocular cameras, f is the camera focal length, and d is the parallax (the difference in coordinates of corresponding feature pixels between the left and right cameras).

[0054] In one specific embodiment, the SGBM local matching algorithm is used to calculate disparity to obtain the preprocessed image. The specific process is as follows:

[0055] First, the disparity map is initialized on the corrected image. Then, based on the actual characteristics of the acquired food distribution image and the calibration parameters of the binocular camera (including baseline distance, focal length, etc.), a reasonable disparity search range is preset. For a fixed-mounted camera and a known shooting distance, this range can be estimated in advance through geometric relationships. During initialization, each pixel in the disparity map is assigned an initial disparity value, which can be uniformly set to 0, or a minimum estimated value can be set based on scene priors. Subsequently, semi-global matching and cost calculation based on SGBM are performed. For corresponding pixel pairs in the left and right views, their brightness differences are calculated, usually using the sum of absolute differences or the sum of squared differences as the initial matching cost. After the pixel-level cost calculation is completed, cost aggregation is performed along multiple directions (including horizontal and vertical directions), that is, the matching costs of adjacent pixels are accumulated through dynamic programming to form an aggregated cost path. During the aggregation process, the path costs need to be normalized to ensure the comparability between different paths. Finally, the minimum cost among all aggregated paths is selected for each pixel, and its optimal disparity value is determined accordingly, thus obtaining the initial disparity map. To improve the accuracy and robustness of the disparity map, disparity optimization and post-processing are required for the initial disparity map. This process mainly includes the following steps:

[0056] Cost aggregation optimization: Utilizing the semi-global optimization strategy in the SGBM algorithm, cost aggregation is performed in multiple directions to enhance matching consistency;

[0057] Filtering and smoothing: Weighted least squares filtering (e.g., implemented through the SFilter function) is used to smooth discontinuous regions in the disparity map caused by noise or occlusion.

[0058] Hole filling: For invalid regions (holes) detected in the disparity map, fill them with the mean or interpolation result of the surrounding reliable disparity values ​​to maintain the integrity of the disparity map.

[0059] Through the above steps, a high-quality, continuous, and accurate disparity map is finally obtained, providing a reliable depth information foundation for subsequent 3D point cloud reconstruction.

[0060] Since the three-dimensional point cloud data includes the ground, feed piles, possible railings or other background objects, in an optional embodiment, step S40, "calculating the actual daily feed distribution area based on the three-dimensional point cloud data", includes: processing the three-dimensional point cloud data, separating the daily feed point cloud, and calculating the projected area of ​​the daily feed point cloud on a cross-section perpendicular to the direction of travel during the unit travel time of the pushing robot, as the actual daily feed distribution area.

[0061] First, the acquired 3D point cloud data is processed. Algorithms (e.g., based on height, color, or texture features) are used to separate the point cloud data belonging to the diet (livestock feed) from the point cloud data of the background (such as the ground, railings, etc.), resulting in a pure diet point cloud. Then, the unit travel time of the feeding robot (e.g., every second or every 0.5 seconds) is used as the analysis time unit. This ensures that the area calculation is synchronized with the robot's movement, facilitating continuous online analysis. Next, within each unit travel time, a profile perpendicular to the robot's direction of travel is determined. This profile can be understood as a vertical "slice" plane horizontally in front of the robot. The diet point cloud separated within this time period is vertically projected onto the aforementioned profile. Then, the area of ​​the region formed by these projected points on the profile is calculated; this area is defined as the actual diet distribution area within that time period.

[0062] More specifically, the Cloth Simulation Filter (CSF) algorithm is used to separate the ground point cloud and the food point cloud, and the Random Sample Consensus (RANSAC) algorithm is used to fit the point cloud plane. The segmented food point cloud is then projected onto this plane to calculate the projected area. This embodiment combines two classic computer vision algorithms, and the area calculation is completed through the following steps:

[0063] Step 1: Segmenting the Ground and Food Contexts Using the CSF Algorithm: A cloth-based filtering algorithm is used for point cloud segmentation to effectively separate ground and non-ground point clouds. Points with a distance less than a threshold are considered ground points (i.e., background to be removed), while points with a distance greater than the threshold are considered clean food point clouds.

[0064] The second step is to determine the projection plane using the RANSAC algorithm: The food point cloud separated by the CSF is used as input, and a random sampling consensus algorithm is used to fit a plane. RANSAC calculates a plane model by repeatedly sampling a small number of points randomly, and counts how many points conform to the model (i.e., are located within a certain threshold near the plane). After multiple iterations, it finds the "optimal plane" containing the most interior points, thus eliminating outliers.

[0065] The third step is to project the point cloud onto the plane and calculate the area: Project all the segmented food point clouds perpendicularly onto the plane along the direction perpendicular to the RANSAC fitting plane (i.e., the direction of the normal vector). The projected points form a dense two-dimensional point set on the plane.

[0066] The CSF algorithm is specifically designed for terrain separation, accurately distinguishing the ground from the accumulated material (feed), and is more adaptable to uneven terrain than the simple height threshold method. The RANSAC algorithm is insensitive to outliers (such as floating dust cloud points or scattered ground points not fully filtered by CSF), robustly fitting the correct projection plane and ensuring the stability of area calculations. This combined algorithm does not require prior knowledge of the precise ground model or the precise shape of the feed, and can adaptively handle different scenarios, making it ideal for online, automated detection in variable real-world farming environments. The accurate point cloud processing and geometric calculations yield reliable projected area data, providing a solid foundation for subsequent calculations of the spatial distribution index of the feed-pushing robot.

[0067] In a specific embodiment, calculating the projected area involves: extracting the boundary of the projected two-dimensional point set using the convex hull algorithm, and calculating the area enclosed by the convex hull.

[0068] In this embodiment, all the food point clouds projected onto the plane are first considered as a discrete set of points on a two-dimensional plane. A convex hull algorithm (such as Graham's scan method or Jarvis's step method) is used to process this point set. This algorithm finds the outermost subset of these points and encloses them all with a convex polygon. This convex polygon is the smallest convex region containing all points, and its vertices are a subset of points from the origin set. After obtaining the vertex sequence determined by the convex hull algorithm, the area of ​​this convex polygon is calculated using geometric methods. The calculated area of ​​the convex polygon is the final determined area of ​​the actual food distribution region.

[0069] Compared to other calculation methods, convex hull calculations offer better computational complexity and stability, meeting the real-time requirements of online detection. The surface of a grain pile may contain localized depressions or internal cavities (such as traces of livestock grazing). These indentations may be irrelevant information for evaluating the overall distribution breadth and coverage. The convex hull algorithm ignores these internal concave features, focusing only on the outermost contour, resulting in an area that more accurately reflects the overall spatial extent occupied by the feed.

[0070] Furthermore, in step S50, the spatial distribution index of the feed after the pushing robot is pushed is calculated using the following formula:

[0071] σ= ;

[0072] In the formula, σ is the spatial distribution index of the feed after the pusher robot pushes the feed, S1 is the area of ​​the ideal feed distribution area, and S2 is the area of ​​the current actual feed distribution area. It should be noted that the overlap between the ideal feed distribution area S1 and the actual feed distribution area S2 is the overlapping part of the ideal feed distribution area S1 and the actual feed distribution area S2.

[0073] Based on this, in an optional embodiment, the actual food distribution image is further compensated by real-time acquisition of motion data (e.g., angular velocity and acceleration) of the pushing robot, which effectively eliminates image blurring and geometric deformation caused by vibration, turning or road bumps of the pushing robot, thereby obtaining a clearer and more stable actual food distribution image.

[0074] Please refer to Figures 2 to 4 In one specific embodiment, this application also provides an online detection system for the spatial distribution index of feed after a feeding robot pushes the feed, used to implement the above-mentioned online detection method for the spatial distribution index of feed after a feeding robot pushes the feed. The system includes an image acquisition unit, a processing unit, and a remote monitoring terminal. The image acquisition unit is installed behind the feeding robot and is used to take real-time pictures of the feed pile after the feeding robot pushes the feed to obtain an image of the actual feed distribution. The processing unit is communicatively connected to the image acquisition unit and is configured to:

[0075] a. Receive and preprocess the actual food distribution image to obtain a preprocessed image, and generate three-dimensional point cloud data from the preprocessed image;

[0076] b. Calculate the actual area of ​​the daily food distribution region based on the aforementioned three-dimensional point cloud data;

[0077] c. Retrieve the area of ​​the ideal daily food distribution region that has been pre-stored;

[0078] d. Calculate the spatial distribution index of the feed pushed by the feeding robot based on the preset ideal feed distribution area and the degree of overlap between the ideal feed distribution area and the actual feed distribution area;

[0079] The remote monitoring terminal is communicatively connected to the processing unit, and the remote monitoring terminal is used to receive and display the spatial distribution index of the daily feed after the pushing robot pushes the feed.

[0080] The remote monitoring terminal is communicatively connected to the processing unit, and the remote monitoring terminal is used to receive and display the spatial distribution index of the daily feed after the pusher robot pushes.

[0081] The online detection system for the spatial distribution index of feed after a feeding robot, provided in this application, firstly involves fixing an image acquisition unit (such as a waterproof and dustproof camera) to the rear of the feeding robot. After the robot completes its feeding action, it immediately captures left and right views of the feed area in real time, forming an image of the actual feed distribution. The acquired image of the actual feed distribution can be transmitted in real time to the processing unit via wired or wireless communication (such as CAN bus, industrial Ethernet, 5G / Wi-Fi). The processing unit (such as an industrial computer or a high-performance embedded controller) executes the following core tasks according to a predetermined program:

[0082] Step a: After receiving the image, perform preprocessing such as denoising and correction to obtain a preprocessed image, and further convert it into point cloud data describing the three-dimensional morphology of the feed surface.

[0083] Step b: Process the generated 3D point cloud (e.g., use the CSF algorithm to separate the ground, use RANSAC to fit the plane and project), and calculate the convex hull area of ​​the feed point cloud projection contour on the cross section perpendicular to the direction of travel. This is the actual area of ​​the daily feed distribution region.

[0084] Step c: Retrieve the area of ​​the ideal ration distribution region (e.g., a rectangular cross-sectional area, such as 0.4 square meters) that has been pre-set and stored from local storage or the server.

[0085] Step d: Compare the actual area with the ideal area, and calculate the final spatial distribution index of the feed pushed by the pushing robot (a value between 0 and 1) based on their overlap.

[0086] The calculated spatial distribution index of the feed after the pusher robot can be fed back to the main control system of the pusher robot in real time, serving as the basis for adjusting the roller parameters (downward pressure, speed) of the pusher robot in the next work cycle. The index can also be sent to a remote monitoring terminal (such as PC, APP or cloud) for visualization (such as numbers, curves, color blocks) so that managers can grasp the overall work quality in real time.

[0087] The system provided in this application integrates image acquisition, data processing, index calculation, result feedback, and monitoring, enabling quality assessment without manual intervention. The image acquisition unit is responsible for front-end perception and specializes in acquiring high-definition images. The processing unit, as the computing core, centrally processes complex algorithms, ensuring computational efficiency and stability. The system is not only used for monitoring but also optimizes the feeding quality of the feeding robot through its processing results (the spatial distribution index of feed after the feeding robot pushes), ensuring that feed is pushed in place, thereby reducing feed loss caused by feed being pushed out of the feeding area and ensuring the feeding quality of dairy cows.

[0088] In a preferred embodiment, the image acquisition unit is a binocular camera, for example, an image acquisition using a ZED2i binocular camera. The advantage of a binocular camera is that it can simultaneously acquire left and right views of the food area, thus improving acquisition efficiency.

[0089] In a preferred embodiment, the processing unit is an industrial computer, such as the FEVM-FN60 mini industrial computer.

[0090] Furthermore, it also includes an inertial measurement unit and a motion correction module. The inertial measurement unit is mounted on the pushing robot and is used to collect the angular velocity and acceleration data of the pushing robot in real time. The motion correction module is connected to the image acquisition unit, the inertial measurement unit, and the processing unit, respectively, and is configured as follows:

[0091] Receive the actual daily food distribution image and the corresponding acquisition timestamp acquired by the image acquisition unit;

[0092] Receive attitude data synchronized with the timestamp collected by the inertial measurement unit;

[0093] The robot's motion trajectory during image exposure is calculated based on the posture data, and the actual food distribution image is subjected to inverse motion compensation to obtain a clear image after deblurring correction.

[0094] The processing unit is further configured to receive and perform parallax calculation and subsequent processing based on the clear image.

[0095] In this embodiment, the inertial measurement unit can be an IMU sensor. The IMU sensor collects the motion data of the pushing robot in real time, and combined with the motion correction module, performs inverse motion compensation on the actual food distribution image. This effectively eliminates image blurring and geometric deformation caused by vibrations, turning, or road bumps of the pushing robot, thereby obtaining a clearer and more stable image of the actual food distribution. This provides higher-quality data input for subsequent parallax calculation, 3D point cloud reconstruction, and area calculation, significantly improving the measurement accuracy and reliability of the entire detection system.

[0096] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A method for online detection of the spatial distribution index of food rations after a feeding robot pushes food, characterized in that, Includes the following steps: S10. Real-time acquisition of actual feed distribution images after the feeding robot pushes the feed; S20. The actual food distribution image is preprocessed to obtain a preprocessed image; S30. Generate three-dimensional point cloud data from the preprocessed image; S40. Calculate the actual area of ​​the daily food distribution region based on the three-dimensional point cloud data; S50. Based on the preset ideal diet distribution area and the degree of overlap between the ideal diet distribution area and the actual diet distribution area, calculate the diet spatial distribution index after the pusher robot pushes the food, and judge the pushing quality of the pusher robot based on the difference between the preset ideal diet spatial distribution index and the diet spatial distribution index after the pusher robot pushes the food.

2. The online detection method for the spatial distribution index of the feed after the pusher robot pushes the feed according to claim 1, characterized in that, The actual food distribution image is a depth image. Step S20 includes: converting the actual food distribution image into a grayscale image, performing Gaussian filtering, denoising, image correction, and disparity calculation on the grayscale image to obtain the preprocessed image. The Gaussian filtering process uses the horizontal Sobel operator to calculate the horizontal gradient features of the image.

3. The online detection method for the spatial distribution index of the feed after the pusher robot pushes the feed, as described in claim 2, is characterized in that, The preprocessed image is obtained by calculating disparity using the SGBM local matching algorithm.

4. The online detection method for the spatial distribution index of the feed after the pusher robot pushes the feed according to claim 1, characterized in that, In step S40, the step of "calculating the actual daily food distribution area based on the three-dimensional point cloud data" includes: processing the three-dimensional point cloud data, separating the daily food point cloud, and calculating the projected area of ​​the daily food point cloud on a cross-section perpendicular to the direction of travel during the unit travel time of the pushing robot, which is taken as the actual daily food distribution area.

5. The online detection method for the spatial distribution index of the feed after the pusher robot pushes the feed, as described in claim 4, is characterized in that, The ground point cloud and the food point cloud are separated using the Cloth Simulation Filter (CSF) algorithm, and the point cloud plane is fitted using the Random Sample Consensus (RANSAC) algorithm. The segmented food point cloud is then projected onto this plane to calculate the projected area.

6. The online detection method for the spatial distribution index of the feed after the pusher robot pushes the feed according to claim 4, characterized in that, The calculation of the projected area is specifically as follows: the boundary is extracted from the projected two-dimensional point set using the convex hull algorithm, and the area enclosed by the convex hull is calculated.

7. The online detection method for the spatial distribution index of the feed after the pusher robot pushes the feed according to claim 1, characterized in that, In step S50, the spatial distribution index of the feed after the pusher robot pushes the feed is calculated using the following formula: σ= ; In the formula, σ is the spatial distribution index of the feed after the pusher robot pushes the feed, S1 is the area of ​​the ideal feed distribution area, and S2 is the area of ​​the current actual feed distribution area.

8. An online detection system for the spatial distribution index of feed after a feeding robot pushes food, characterized in that, The method for online detection of the spatial distribution index of feed after a feeding robot pushes feed, as described in any one of claims 1 to 7, includes an image acquisition unit, a processing unit, and a remote monitoring terminal. The image acquisition unit is installed behind the feeding robot and is used to capture real-time images of the feed pile after the robot pushes feed, obtaining an image of the actual feed distribution. The processing unit is communicatively connected to the image acquisition unit and is configured to: a. Receive and preprocess the actual food distribution image to obtain a preprocessed image, and generate three-dimensional point cloud data from the preprocessed image; b. Calculate the actual area of ​​the daily food distribution region based on the aforementioned three-dimensional point cloud data; c. Retrieve the area of ​​the ideal daily food distribution region that has been pre-stored; d. Calculate the spatial distribution index of the feed pushed by the feeding robot based on the preset ideal feed distribution area and the degree of overlap between the ideal feed distribution area and the actual feed distribution area; The remote monitoring terminal is communicatively connected to the processing unit, and the remote monitoring terminal is used to receive and display the spatial distribution index of the daily feed after the pusher robot pushes.

9. The online detection system for the spatial distribution index of feed after the pusher robot as described in claim 8, characterized in that, The image acquisition unit is a binocular camera.

10. The online detection system for the spatial distribution index of feed after the pusher robot is pushed according to claim 8, characterized in that, The processing unit is an industrial control computer.

Images