Motion control method for unmanned transfer trolley for livestock and poultry died of illness and transfer trolley
By combining binocular cameras and the A-Star algorithm, the problems of positioning accuracy and computing resource requirements of unmanned transport vehicles were solved, and efficient and safe path planning was achieved.
Patent Information
- Application Number
- CN202411093180.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2024-11-26
AI Technical Summary
In existing technologies, the positioning accuracy of unmanned transport vehicles is affected by external signal interference and requires high computing resources, making it difficult to achieve accurate positioning and efficient path planning.
A stereo camera is used to acquire image information of the same environment from different directions. A 3D model is constructed by feature point matching and depth information calculation. Combined with target detection and tracking algorithms, the A* algorithm is used to generate the optimal path, reducing the computational resource requirements.
It achieves high-precision positioning and path planning for unmanned transport vehicles, avoids external signal interference, and improves the safety and efficiency of path planning.
Smart Images

Figure CN119024836A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of positioning and navigation, and in particular to a motion control method for an unmanned transport vehicle for dead livestock and poultry and a transport vehicle. Background Art
[0002] With the rapid development of modern agriculture and animal husbandry, the scale of livestock and poultry farming has continued to expand, and the problem of handling dead livestock and poultry has become increasingly prominent. Traditionally, the handling of dead livestock and poultry usually requires manual intervention, which not only increases the cost of handling, but also during the manual handling of dead livestock and poultry, operators may be exposed to pathogens, increasing the risk of infection. In addition, manual handling is inefficient and cannot meet the needs of large-scale farms.
[0003] In recent years, with the continuous progress in the fields of autonomous driving technology, sensor technology and artificial intelligence, the technical level of unmanned vehicles has gradually matured, providing technical support for the development of unmanned transport vehicles for dead livestock and poultry. The emergence of unmanned transport vehicles can greatly reduce the risk of manual processing and improve safety and processing efficiency. Unmanned transport vehicles use sensors to perceive the environment and their own status, and then achieve autonomous movement towards the target in an environment with obstacles. Positioning is to determine the position of the vehicle relative to the global coordinates in the working environment and its own posture, which is the basic link of the navigation of the loading vehicle.
[0004] At present, there are many technologies used for car navigation and positioning, mainly the following two. UWB (Ultra-Wideband) space beacon positioning is a technology that uses ultra-wideband signals for positioning. By arranging multiple UWB beacons in space, the distance between the beacon and the receiver and the signal propagation time difference are used to accurately locate the receiver. However, UWB signals are easily affected by multipath effects during propagation, that is, the signal reaches the receiver through multiple paths, resulting in signal attenuation, delay expansion and other problems, reducing positioning accuracy. At the same time, UWB signals may also be interfered with by other wireless devices, such as Wi-Fi, Bluetooth, etc., affecting the stability and accuracy of the positioning system.
[0005] Another commonly used navigation and positioning technology is SLAM (Simultaneous Localization and Mapping). SLAM is a technology that achieves autonomous positioning and map construction in an unknown environment. By fusing sensor information such as cameras, lidar, etc., the SLAM system can estimate the position and posture of mobile devices such as robots, vehicles or drones in real time, and simultaneously build a map of the surrounding environment. In SLAM, it usually includes two parts: the front end is responsible for real-time estimation of position and construction of local maps, and the back end is responsible for optimizing the entire trajectory and map. SLAM technology has the advantages of no need to pre-arrange beacons and strong adaptability, and is widely used in unmanned driving, robot navigation and other fields. However, the SLAM algorithm requires a lot of computing resources and time, has high hardware performance requirements and high computational complexity, and is prone to system delays.
[0006] Therefore, how to avoid interference from external signals on the unmanned transport vehicle while reducing the unmanned transport vehicle's requirements for computing resources and achieving precise positioning of the unmanned transport vehicle has become a technical problem that needs to be solved urgently. Summary of the invention
[0007] The main purpose of the present invention is to provide a motion control method and a transport vehicle for an unmanned transport vehicle for dead livestock and poultry, aiming to avoid interference of external signals on the unmanned transport vehicle while reducing the requirements of the unmanned transport vehicle for computing resources and achieving precise positioning of the unmanned transport vehicle.
[0008] In order to achieve the above object, the present invention proposes a motion control method for an unmanned transport vehicle for dead livestock and poultry, comprising the following steps: Obtain image information from two different directions of the same environment, compare the image information from two different directions of the same environment, match the corresponding feature points in the two image information, and obtain depth information of the corresponding feature points in the same environment; Build a three-dimensional model of the current environment based on the depth information of each object; Use target detection algorithm and target tracking algorithm to identify each object in the current environment and obtain the position parameters and motion state parameters of each object; According to the three-dimensional model, the position parameters of each object and its own motion state, the optimal motion path is generated by the A-star algorithm.
[0009] In one embodiment of the present application, comparing image information in different directions of the same environment, matching feature points therein, and obtaining depth information of each object in the same environment includes the following steps: acquiring intrinsic parameters of two image acquisition devices; acquiring external parameters of two image acquisition devices; The feature points in the two images are corrected and aligned using the intrinsic and extrinsic parameters of the two image acquisition devices, and the corresponding feature points are converted into coordinate parameters in a three-dimensional space using a stereo matching algorithm; According to the parallax and the distance between the two image acquisition devices, the depth information of the object is calculated using the triangulation principle.
[0010] In one embodiment of the present application, the method further includes: correcting and aligning feature points in two images using intrinsic parameters and extrinsic parameters of two image acquisition devices; It also includes converting the two image information into grayscale or color space.
[0011] In one embodiment of the present application, after the two image information are converted into grayscale images or color spaces, denoising processing is performed on the image information acquired by the two image acquisition devices respectively.
[0012] In one embodiment of the present application, the intrinsic parameters include: At least one of the focal length, optical center position, and distortion coefficient of the image acquisition device.
[0013] In one embodiment of the present application, the extrinsic parameters include a translation vector and a rotation matrix.
[0014] In one embodiment of the present application, the three-dimensional model of the current environment is constructed using at least one of a parallax algorithm and a triangulation algorithm.
[0015] The present application also discloses an unmanned transport vehicle for dead livestock and poultry, comprising: The vehicle body is provided with a binocular camera, a wireless communication unit, a single board computer, and a control unit, wherein: Binocular camera, used to obtain image information from two different directions of the same environment; A wireless communication unit, communicatively connected to the binocular camera, and used for uploading the image information collected by the binocular camera to a server to implement analysis of the image information; A single-board computer, wherein the single-board computer is connected to a server via a wireless communication unit, and is used to receive a target and its own coordinate vectors sent by the server, and to generate a control instruction according to the target and its own coordinate vectors; The control unit is communicatively connected to the single board computer, receives the control instruction, and controls the motor and steering gear of the vehicle body to complete the angle and direction control of the vehicle body.
[0016] By adopting the above technical solution, through the heuristic search of the A-star algorithm, the optimal path from the starting point to the end point can be efficiently found to ensure the shortest path and the lowest cost. Combined with the target detection and tracking algorithm, it can update the object information in the environment in real time, dynamically adjust the path, and adapt to environmental changes. Through image information acquisition and feature point matching, high-precision point cloud data can be generated to build a detailed three-dimensional environment model. Through the target tracking algorithm, the motion state of the object can be tracked in real time, the dynamic information of the object can be obtained, and reliable data support can be provided for path planning. Combined with the motion state parameters of the object, the path can be adjusted dynamically to avoid collisions with moving objects and improve the safety of path planning. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The present invention is described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 It is a schematic structural diagram of the first embodiment of the present invention. DETAILED DESCRIPTION
[0018] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the following specific embodiments are only used to explain the present invention and do not constitute a limitation to the present invention.
[0019] like Figure 1 As shown, in order to achieve the above-mentioned purpose, the present invention proposes a motion control method for an unmanned transport vehicle for dead livestock and poultry, comprising the following steps: Obtain image information from two different directions of the same environment, compare the image information from two different directions of the same environment, match the corresponding feature points in the two image information, and obtain depth information of the corresponding feature points in the same environment; Build a three-dimensional model of the current environment based on the depth information of each object; Use target detection algorithm and target tracking algorithm to identify each object in the current environment and obtain the position parameters and motion state parameters of each object; According to the three-dimensional model, the position parameters of each object and its own motion state, the optimal motion path is generated by the A-star algorithm.
[0020] Specifically, to obtain image information, select a high-resolution camera or a stereo camera to ensure that clear image information can be captured. Install the camera on a mobile device (such as a robot or an unmanned vehicle) to ensure that images can be obtained from different directions. In the same environment, image information is collected from two different directions. It is manifested as a binocular camera that collects image information in different directions in the same environment from two different angles to ensure that the viewing angles in the two directions are somewhat different.
[0021] Perform denoising on the collected images to remove noise and improve image quality. Convert color images to grayscale images to simplify the subsequent feature extraction process.
[0022] Feature point extraction uses the ORB (rapid robust feature) algorithm to extract key feature points from the two images. The extracted feature points are described to generate feature vectors. Feature point matching uses the KNN (K nearest neighbor) algorithm to match the feature points in the two images. The RANSAC (random sampling consensus) algorithm is used to remove incorrectly matched feature points and retain correctly matched feature point pairs. Depth information calculation Based on the matched feature point pairs, the disparity value (i.e., the position difference between the two feature points in the two images) is calculated. The depth information of each feature point is calculated using the disparity value and the baseline distance of the camera.
[0023] 3D environment modeling generates point cloud data of the environment based on the depth information of feature points, and each point contains 3D coordinates (X, Y, Z). The point cloud data is filtered and denoised to improve its quality.
[0024] The 3D model construction uses algorithms such as Delaunay triangulation or Poisson surface reconstruction to convert point cloud data into a 3D mesh model. The texture information of the original image is mapped onto the 3D mesh model to generate a 3D environment model with texture.
[0025] Object detection and tracking uses the YOLO (You Only Look Once) object detection algorithm to identify objects in the environment. Generate bounding boxes for detected objects and mark the locations of the objects. Use the KCF (Kernelized Correlation Filters) algorithm to track detected objects. Obtain the location parameters (such as coordinates, size) and motion state parameters (such as speed, direction) of each object.
[0026] Path planning takes the 3D environment model, the position parameters and motion state parameters of the object as input data, and uses the A-star algorithm for path planning.
[0027] Define the nodes in the search space, each node contains information such as position and cost.
[0028] Use the evaluation function F(n), where g(n) is the actual cost from the starting point to the current node, and h(n) is the estimated cost from the current node to the end point.
[0029] During the search process, the node with the smallest F value is preferentially selected for expansion until the optimal path from the starting point to the end point is found. The Bezier curve method is added to smooth the generated path, reducing sharp turns in the path to make the path more natural and more suitable for actual movement. After completing the global planning, the DWA algorithm is added as a local planning. By evaluating the position distance, route results, and the relationship to obstacles during the walking process, obstacle avoidance is achieved when encountering obstacles. By adopting the above technical solution, through the heuristic search of the A-star algorithm, the optimal path from the starting point to the end point can be efficiently found to ensure the shortest path and the lowest cost. Combined with the target detection and tracking algorithm, it can update the object information in the environment in real time, dynamically adjust the path, and adapt to environmental changes. Through image information acquisition and feature point matching, high-precision point cloud data can be generated to build a detailed three-dimensional environment model. Through the target tracking algorithm, the motion state of the object can be tracked in real time, the dynamic information of the object can be obtained, and reliable data support can be provided for path planning. Combined with the motion state parameters of the object, the path can be adjusted dynamically to avoid collisions with moving objects and improve the safety of path planning.
[0030] In one embodiment of the present application, comparing image information in different directions of the same environment, matching feature points therein, and obtaining depth information of each object in the same environment includes the following steps: acquiring intrinsic parameters of two image acquisition devices; acquiring external parameters of two image acquisition devices; The feature points in the two images are corrected and aligned using the intrinsic and extrinsic parameters of the two image acquisition devices, and the corresponding feature points are converted into coordinate parameters in a three-dimensional space using a stereo matching algorithm; According to the parallax and the distance between the two image acquisition devices, the depth information of the object is calculated using the triangulation principle.
[0031] Specifically, the intrinsic parameters and extrinsic parameters of the image acquisition device are acquired.
[0032] The intrinsic parameters include the focal length, principal point coordinates (optical center), distortion coefficient, etc. of the camera. Use a checkerboard or other calibration plate to calibrate the camera. By taking multiple images of the calibration plate and calibrating using tools such as OpenCV, the intrinsic parameter matrix and distortion coefficient of the camera are calculated. Prepare a calibration plate (such as a checkerboard) and take multiple images of the calibration plate from different angles. Use calibration software (such as OpenCV) to calibrate and obtain the intrinsic parameter matrix and distortion coefficient.
[0033] The external parameters include the relative position and attitude between the cameras (rotation matrix and translation vector). Through stereo calibration, the relative position and attitude between the two cameras are obtained.
[0034] Take multiple images of the calibration plate in the overlapping area of the two cameras' field of view. Use a stereo calibration tool (such as the stereoCalibrate function in OpenCV) to calculate the extrinsic parameters (rotation matrix and translation vector) of the two cameras.
[0035] Use intrinsic and extrinsic parameters to correct and align the images so that the two images are on the same plane. Use the intrinsic parameters and distortion coefficients of the camera to correct the image distortion. Use extrinsic parameters to align the images of the two cameras to the same plane. Use the ORB algorithm to extract feature points from the corrected image. Extract feature points from the two images separately. Generate descriptors of the feature points for subsequent matching.
[0036] The KNN algorithm is used to match the feature points in the two images. The KNN algorithm is used to find the nearest neighbor feature point of each feature point in the other image. The RANSAC algorithm is used to remove incorrect matches and retain the correctly matched feature point pairs.
[0037] The three-dimensional coordinate conversion uses a stereo matching algorithm to convert the matched feature point pairs into coordinate parameters in three-dimensional space. The disparity value is calculated based on the matched feature point pairs. The three-dimensional coordinates of each feature point are calculated using the disparity value and the baseline distance of the camera.
[0038] Depth information calculation, parallax refers to the position difference of the same object in two images. For each pair of matching feature points, the position difference between the two images is calculated, that is, the disparity value. Depth calculation uses the principle of triangulation to calculate the depth information of the object based on the parallax and the baseline distance between the two cameras.
[0039] The calculation formula is as follows:
[0040] in, is the focal length of the camera, is the baseline distance between the two cameras, is the parallax value.
[0041] Get the focal length of the camera Distance from baseline .
[0042] According to the disparity value, the depth information of each feature point is calculated using the above formula.
[0043] The above technical solution can accurately calculate the depth information of objects through stereo vision and triangulation principle with high accuracy and small error. Using high-resolution camera and advanced feature point extraction algorithm, more detailed information can be captured and high-precision depth map can be generated. Using efficient feature point extraction and matching algorithm, image data can be processed quickly and real-time depth information calculation can be realized.
[0044] In one embodiment of the present application, the method further includes: correcting and aligning feature points in two images using intrinsic parameters and extrinsic parameters of two image acquisition devices; It also includes converting the two image information into grayscale or color space.
[0045] With the above technical solution, the grayscale image contains only one channel (grayscale value), which greatly reduces the amount of data and the computational complexity compared to the three channels (RGB) of the color image. Due to the reduction in data volume, the speed of processing steps such as feature extraction, matching, and depth calculation is significantly improved. By converting to other color spaces (such as HSV, Lab), color information can be separated from brightness information, making color features more prominent. On specific channels of color spaces (such as HSV), feature point detection may be more accurate.
[0046] In one embodiment of the present application, after the two image information are converted into grayscale images or color spaces, denoising processing is performed on the image information acquired by the two image acquisition devices respectively.
[0047] Specifically, the denoising processing method includes at least one of Gaussian filtering, median filtering, bilateral filtering, and non-local mean filtering.
[0048] Gaussian filtering is a linear smoothing filter that reduces high-frequency noise in an image through convolution operations. The weight of the filter is determined by a Gaussian function, with the largest weight center and decreasing with increasing distance.
[0049] When using Gaussian filtering, select an appropriate Gaussian kernel size and standard deviation. Perform a convolution operation on the Gaussian kernel and the image to generate a smoothed image.
[0050] Median filtering is a nonlinear filtering method that can effectively remove salt and pepper noise by taking the median of the pixel values in a local window to replace the central pixel value.
[0051] When using median filtering, select an appropriate window size (such as 3x3 or 5x5). For each pixel, take the median value of all pixel values in its neighborhood window and replace the pixel value.
[0052] Bilateral filtering combines the spatial domain smoothing of Gaussian filtering and the similarity of pixel value domain, which can remove noise while retaining edge details.
[0053] When using bilateral filtering, select appropriate spatial domain Gaussian kernel and pixel value domain Gaussian kernel. Calculate the weighted average of each pixel, and the weight is determined by the spatial distance and the pixel value difference.
[0054] Non-local mean filtering removes noise by calculating the weighted average of similar blocks in the image, which can retain more image details.
[0055] When using non-local mean filtering, select appropriate search window and similarity window sizes. For each pixel, search for similar blocks in its neighborhood, calculate the weighted average, and replace the pixel value.
[0056] The advantages of denoising by using the above technical solution are that the noise in the image can be effectively removed through denoising, the signal-to-noise ratio of the image can be improved, and the image can be made clearer. By using a suitable denoising algorithm (such as bilateral filtering, non-local mean filtering), the details and edge information in the image can be retained while denoising. After denoising, the noise in the image is reduced, and the feature point extraction algorithm can detect the real feature points more accurately, reducing the probability of mis-extraction and mis-matching.
[0057] In one embodiment of the present application, the intrinsic parameters include: At least one of the focal length, optical center position, and distortion coefficient of the image acquisition device.
[0058] In one embodiment of the present application, the extrinsic parameters include a translation vector and a rotation matrix.
[0059] In one embodiment of the present application, the three-dimensional model of the current environment is constructed using at least one of a parallax algorithm and a triangulation algorithm.
[0060] The present application also discloses an unmanned transport vehicle for dead livestock and poultry, comprising: The vehicle body is provided with a binocular camera, a wireless communication unit, a single board computer, and a control unit, wherein: Binocular camera, used to obtain image information from two different directions of the same environment; A wireless communication unit, communicatively connected to the binocular camera, and used for uploading the image information collected by the binocular camera to a server to implement analysis of the image information; A single-board computer, wherein the single-board computer is connected to a server via a wireless communication unit, and is used to receive a target and its own coordinate vectors sent by the server, and to generate a control instruction according to the target and its own coordinate vectors; The control unit is communicatively connected to the single board computer, receives the control instruction, and controls the motor and steering gear of the vehicle body to complete the angle and direction control of the vehicle body.
[0061] Specifically, the vehicle body is the basic structure of the entire system and carries all other components. It includes basic mechanical structures such as chassis, wheels, motors and steering gears. The binocular camera is installed on the vehicle body and is responsible for obtaining image information in two different directions in the same environment. Through the binocular perspective of the camera, the depth information of the environment can be collected. The wireless communication unit is connected to the binocular camera and is used to upload the image information collected by the camera to the server. It is also responsible for communicating with the server and receiving data sent by the server.
[0062] The single board computer is connected to the server through a wireless communication unit. It receives the target and its own coordinate vectors sent by the server and generates control instructions based on this information. The single board computer is the "brain" of the entire system, responsible for calculations and decisions.
[0063] The control unit is connected to the single-board computer for communication and receives control instructions generated by the single-board computer. It controls the motor and steering gear of the vehicle body, thereby adjusting the angle and direction of the vehicle body.
[0064] The binocular camera provides stereoscopic vision and can obtain depth information of the environment. Through the binocular camera, the system can perceive the surrounding environment more accurately and plan the path. The wireless communication unit ensures real-time data transmission between the unmanned vehicle and the server. Through wireless communication, the vehicle system can upload real-time image data to the server and receive instructions from the server to form a closed-loop control.
[0065] As the core computing unit of the system, the single-board computer is responsible for processing the coordinate vectors received from the server and generating corresponding control instructions. Its computing power directly affects the response speed and control accuracy of the system. The control unit receives instructions from the single-board computer, controls the motor and steering gear of the vehicle body, and realizes precise control of the vehicle body. The performance of the motor and steering gear directly determines the motion performance and stability of the unmanned vehicle.
[0066] By adopting the above technical solutions, through automated control and real-time data transmission, the unmanned transport vehicle can efficiently complete the transport of dead livestock and poultry, reducing labor costs and operational risks. The stereoscopic vision provided by the binocular camera and the powerful computing power of the single-board computer enable the system to perform accurate path planning and obstacle avoidance, ensuring the safety and reliability of the transport process. The precise control of the motor and servo by the control unit enables the unmanned vehicle to flexibly respond to various complex environments and has strong adaptability.
[0067] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.
Claims
1. A motion control method for an unmanned transport vehicle for dead livestock and poultry, characterized in that: The following steps are involved: Obtain image information from two different directions of the same environment, compare the image information from two different directions of the same environment, match the corresponding feature points in the two image information, and obtain depth information of the corresponding feature points in the same environment; Build a three-dimensional model of the current environment based on the depth information of each object; Use target detection algorithm and target tracking algorithm to identify each object in the current environment and obtain the position parameters and motion state parameters of each object; According to the three-dimensional model, the position parameters of each object and its own motion state, the optimal motion path is generated by the A-star algorithm.
2. The motion control method for an unmanned transport vehicle for dead livestock and poultry according to claim 1, characterized in that: Compare image information in different directions of the same environment, match the feature points therein, and obtain depth information of each object in the same environment, including the following steps: acquiring intrinsic parameters of two image acquisition devices; acquiring external parameters of two image acquisition devices; The feature points in the two images are corrected and aligned using the intrinsic and extrinsic parameters of the two image acquisition devices, and the corresponding feature points are converted into coordinate parameters in a three-dimensional space using a stereo matching algorithm; According to the parallax and the distance between the two image acquisition devices, the depth information of the object is calculated using the triangulation principle.
3. The motion control method for an unmanned transport vehicle for dead livestock and poultry according to claim 2, characterized in that: The method further includes correcting and aligning the feature points in the two images using the intrinsic parameters and extrinsic parameters of the two image acquisition devices: It also includes converting the two image information into grayscale or color space.
4. The motion control method for an unmanned transport vehicle for dead livestock and poultry according to claim 3, characterized in that: After the two image information are converted into grayscale images or color spaces, the image information acquired by the two image acquisition devices are subjected to denoising processing respectively.
5. The motion control method for an unmanned transport vehicle for dead livestock and poultry according to claim 2, characterized in that: The internal parameters include: At least one of the focal length, optical center position, and distortion coefficient of the image acquisition device.
6. The motion control method for an unmanned transport vehicle for dead livestock and poultry according to claim 2, characterized in that: The external parameters include a translation vector and a rotation matrix.
7. The motion control method for an unmanned transport vehicle for dead livestock and poultry according to claim 1, characterized in that: The three-dimensional model of the current environment is constructed using at least one of a parallax algorithm and a triangulation algorithm.
8. An unmanned transport vehicle for dead livestock and poultry, characterized in that: include: The vehicle body is provided with a binocular camera, a wireless communication unit, a single board computer, and a control unit, wherein: Binocular camera, used to obtain image information from two different directions of the same environment; A wireless communication unit, communicatively connected to the binocular camera, and used for uploading the image information collected by the binocular camera to a server to implement analysis of the image information; A single-board computer, wherein the single-board computer is connected to a server via a wireless communication unit, and is used to receive a target and its own coordinate vectors sent by the server, and to generate a control instruction according to the target and its own coordinate vectors; The control unit is communicatively connected to the single board computer, receives the control instruction, and controls the motor and steering gear of the vehicle body to complete the angle and direction control of the vehicle body.
Citation Information
Patent Citations
Binocular vision based reconstruction method
CN107945220A
Robot gear chamfering machining method based on parallel binocular stereoscopic vision
CN113902810A
Path planning method and device for mine unmanned vehicle and storage medium
CN115509223A
Robot path planning method and system in radioactive environment
CN116989793A
Four-rotor unmanned aerial vehicle visual target recognition and positioning method based on binocular camera
CN118135526A
