Agricultural robot visual navigation method and system based on vehicle speed adaptive gimbal view angle
By dynamically adjusting the combination of camera perspective and depth information, the accuracy and robustness issues of existing agricultural robot visual navigation technology under different operating speeds and complex scenarios are solved, achieving high-precision navigation control.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGSU UNIV
- Filing Date
- 2026-05-11
- Publication Date
- 2026-07-10
AI Technical Summary
Existing visual navigation technologies for agricultural robots suffer from limitations in navigation accuracy, which is affected by the camera's installation height and pitch angle. They cannot adapt to navigation requirements at different operating speeds and lack robustness in complex scenarios.
By acquiring the agricultural robot's speed in real time, the horizontal pitch angle of the gimbal is dynamically adjusted to change the camera's vertical view. Combining depth information and adaptive color feature enhancement methods, a binary image of the crop rows is generated. The crop row skeleton line is generated using a regional centroid slicing connection method. Finally, the heading angle and lateral deviation are calculated in the world coordinate system to generate reliable navigation commands.
It achieves high-precision navigation under different operating speeds and complex scenarios, improves the absolute accuracy and environmental adaptability of the navigation system, and ensures that navigation data can be directly used for the control of unmanned agricultural machinery.
Smart Images
Figure CN122363233A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of autonomous driving technology for agricultural robots, specifically relating to a three-dimensional precise positioning and navigation method and system that integrates real-time driving speed, dynamic adjustment of gimbal view, and depth visual information. Background Technology
[0002] Unmanned agricultural vehicles (UAGVs) are key equipment for realizing smart agriculture, and high-precision, robust machine vision navigation technology is the foundation for their autonomous execution of precise row-to-row operations in the field (such as tillage, plant protection, and harvesting). Existing machine vision navigation technologies mostly rely on fixed-mounted RGB cameras, extracting the centerline of crop rows through image morphological processing to calculate the heading angle and lateral deviation of the agricultural robot. However, this approach has significant limitations: First, the lateral deviations output in previous studies are mostly pixel coordinate differences or model inference values, which cannot be directly converted into displacements in the real physical world. Furthermore, navigation accuracy is easily affected by the camera's mounting height and pitch angle, making it difficult to meet the requirements for precise row-to-row autonomous operations. Second, traditional algorithms typically use fixed-view camera systems, which cannot adapt to the differentiated navigation information needs of agricultural robots at different operating speeds. For example, during low-speed, fine-grained operations, attention needs to be paid to the details of nearby crop rows to achieve precise row-to-row navigation; while during high-speed operations or relocation, the overall trend of distant crop rows needs to be considered to ensure stability under high-speed conditions. Therefore, visual navigation systems using fixed-view conditions struggle to maintain optimal performance across the entire speed range. Furthermore, existing methods still lack robustness in complex scenarios such as dense, interrupted, and sparse crop rows. Therefore, researching a visual navigation method that can dynamically optimize the perspective of the crop row detection camera based on the real-time vehicle speed is of great significance for improving the machine vision navigation performance of agricultural robots in various operational scenarios. Summary of the Invention
[0003] To address the shortcomings of existing technologies, this invention provides a visual navigation method and system for agricultural robots that dynamically adjusts the observation angle of the camera gimbal based on vehicle speed adaptation.
[0004] The present invention achieves the above-mentioned technical objectives through the following technical means.
[0005] A visual navigation method for agricultural robots based on vehicle speed adaptive gimbal view:
[0006] Real-time acquisition of the agricultural robot's speed V;
[0007] Based on the travel speed V, the gimbal controller controls the horizontal pitch angle of the gimbal, thereby dynamically adjusting the vertical viewing angle of the camera.
[0008] Under the camera's field of view after gimbal adjustment, color and depth images of the field scene are acquired, and the color images are processed using an adaptive color feature enhancement and segmentation method to obtain crop row binarized images.
[0009] Based on the perspective of the operation scene from far to near, the crop row skeleton line is generated in the binarized image by using the regional centroid slicing and connection method.
[0010] The maximum outer envelope method is used to extract the overall contour of all crop row skeleton lines, and the contour of the target crop row within the operation range of the agricultural robot is selected. Then, the least squares method is used to fit the global navigation line of the crop row, and the pixel coordinates of the endpoints of the global navigation line are transformed to the world coordinate system by combining depth information.
[0011] Based on the global navigation line of the crop row and the center line of the vehicle body in the world coordinate system, the heading angle and lateral deviation of the agricultural robot are calculated.
[0012] The autonomous vehicle coordinate system is transformed based on the center position of the front axle of the autonomous vehicle in order to correct the heading angle and physical lateral deviation required for actual control.
[0013] Based on the validity of the navigation results, control commands are generated and sent to the steering controller.
[0014] Furthermore, based on the travel speed V, the gimbal controller controls the horizontal pitch angle of the gimbal, thereby dynamically adjusting the camera's vertical viewing angle, specifically as follows:
[0015] If V < 0.5 m / s, adjust the camera's vertical angle to 60 degrees and focus on the nearby crop rows within 2-5 meters in front of the vehicle;
[0016] If 0.5m / s≤V<1.5m / s, adjust the camera's vertical angle to 45 degrees and focus on the nearby crop rows within a range of 5-10 meters in front of the vehicle;
[0017] If V≥1.5m / s, adjust the camera's vertical viewing angle to 30 degrees and observe the distant trend within a range of 10-20 meters in front of the vehicle.
[0018] Furthermore, the adaptive color feature enhancement and segmentation method specifically involves: calculating the TGBR super green index of the color image and using the Otsu method for adaptive threshold segmentation to obtain a crop row binarized image.
[0019] Furthermore, a regional centroid slicing connection method is used to generate crop row skeleton lines. Specifically, the observation area is divided into three sub-regions in the vertical direction: upper, middle, and lower, with a 10% overlap between adjacent sub-regions. The connected domains within each sub-region are calculated, and the centroid coordinates of the connected domains are obtained. Within each sub-region, it is determined whether the longitudinal and lateral offsets between the centroid coordinates are subject to constraints. Centroids that meet the constraints are connected with line segments to form crop row skeleton lines.
[0020] Furthermore, the process of filtering out the target crop row outlines within the operating range of the agricultural robot involves: calculating the geometric features of the skeleton line outlines of each crop row; if the crop row outline height is less than 0.7 times the image height or the outline width is greater than 0.3 times the image width, then the crop row does not meet the navigation scenario and is removed.
[0021] Furthermore, the pixel coordinates of the endpoints of the global navigation line are transformed to the world coordinate system. Specifically, the endpoints of the global navigation line of the crop row are selected, the depth value corresponding to the point in the aligned depth image is queried, the camera calibration parameters are obtained using the Zhang Zhengyou calibration method, and the coordinates of the endpoints of the global navigation line of the crop row in the world coordinate system are calculated by inverse transformation of the perspective projection model.
[0022] Furthermore, the heading angle that the agricultural robot needs to adjust °, where P_w1.x represents the x-coordinate of the bottom endpoint of the crop row navigation line in the world coordinate system. This represents the ordinate of the bottom endpoint of the crop row navigation line in the world coordinate system. This represents the x-coordinate of the top endpoint of the crop row navigation line in the world coordinate system. The vertical coordinate of the top endpoint of the crop row navigation line in the world coordinate system is represented; the lateral deviation ΔX required to be adjusted by the agricultural robot is ΔX = P_w1.x - P_c1.x, where P_c1.x is the horizontal coordinate of the bottom endpoint of the vehicle centerline in the world coordinate system.
[0023] Furthermore, the vehicle coordinate system transformation is achieved by converting the lateral deviation ΔX and heading angle θ into the actual control deviations ΔX_control and θ_control in the vehicle coordinate system with the front axle center as the origin through spatial translation.
[0024] Furthermore, the control command format is "#VIS, Lateral Deviation, Heading Angle, Status Identifier\r\n", where the status identifier "YY" indicates valid data and "XX" indicates invalid data. The actual control deviation is verified: if |θ_control|>20° or crop row navigation line fitting fails, the status identifier is "XX", and the "GNSS+machine vision" combined navigation safety strategy is enabled; otherwise, the status identifier is "YY", and ΔX_control and θ_control are sent to the steering controller of the agricultural robot.
[0025] A visual navigation system for agricultural robots based on vehicle speed adaptive gimbal perspective includes a depth camera, a Doppler velocimeter, a navigation industrial control computer, and a steering controller. The depth camera is mounted on a two-axis gimbal, the Doppler velocimeter is mounted above the left front wheel of the agricultural robot, and the navigation industrial control computer and steering controller are integrated and mounted at the center of the agricultural robot. The controller of the two-axis gimbal, the depth camera, the Doppler velocimeter, and the steering controller are all connected to the navigation industrial control computer through a serial communication module.
[0026] The depth camera captures color and depth images of the field scene;
[0027] The Doppler velocimeter acquires the current speed of the agricultural robot in real time;
[0028] The navigation industrial control computer is used to acquire binarized images of crop rows, generate crop row skeleton lines, fit global navigation lines of crop rows, and calculate the heading angle and lateral deviation that the agricultural robot needs to adjust.
[0029] The steering controller is used to receive and parse machine vision navigation deviation data, and then send it to the steering actuator.
[0030] The advantages of this invention compared to the prior art are:
[0031] (1) Compared with traditional visual navigation methods that only output pixel deviation, this invention integrates camera depth information Zc and directly outputs the lateral deviation in the physical world through three-dimensional coordinate calculation, which fundamentally improves the absolute accuracy of visual navigation.
[0032] (2) Compared with the traditional fixed-view camera system, the present invention innovatively introduces real-time vehicle speed as a control variable and dynamically adjusts the camera pitch observation angle through a two-axis gimbal, so that the system can achieve fine close-range line control at low speed and ensure overall stable tracking of line trend at high speed, thus realizing adaptive optimal visual navigation angle control under different operating scenarios.
[0033] (3) Compared with traditional crop row extraction methods, the regional centroid slicing connection method proposed in this invention can more effectively generate robust guide lines in complex scenarios where crop rows are dense, sparse, broken, or have interference, thereby improving the system's environmental adaptability.
[0034] (4) Compared with existing systems, this invention can ensure that the output data of the navigation system can be directly used for the control of unmanned agricultural machinery by designing a self-vehicle coordinate transformation, anomaly judgment and protocol-based communication mechanism, thereby improving the practical performance of the visual navigation system. Attached Figure Description
[0035] Figure 1 This is a flowchart of the agricultural robot visual navigation method based on vehicle speed adaptive gimbal perspective described in this invention.
[0036] Figure 2 This is a schematic diagram illustrating the dynamic adjustment of the gimbal's viewing angle at different vehicle speeds as described in this invention (comparison of low speed, medium speed, and high speed).
[0037] Figure 3 This refers to the crop row binarized image obtained based on TGBR features and adaptive threshold segmentation as described in this invention;
[0038] Figure 4 This is a schematic diagram illustrating the process of generating local skeleton lines using the regional centroid slicing connection method described in this invention.
[0039] Figure 5(a) is a schematic diagram of the overall outline of the extraction of all crop row skeleton lines based on the maximum outer envelope of the present invention;
[0040] Figure 5(b) is a schematic diagram of the global navigation line of crop rows fitted by the least squares method according to the present invention;
[0041] Figure 6 This is a schematic diagram illustrating the calculation of heading angle and lateral deviation based on the navigation line (blue) and the vehicle centerline (green) as described in this invention.
[0042] Figure 7 This is a schematic diagram illustrating the principle of the vehicle coordinate system correction described in this invention. Detailed Implementation
[0043] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited thereto.
[0044] The purpose of this invention is to solve the problem that existing visual navigation methods have fixed observation angles and cannot adapt to the requirements of multi-scenario operations. It proposes a visual navigation method and system for agricultural robots based on vehicle speed adaptive gimbal perspective, which makes it more in line with the current multi-scenario operation needs of agricultural robots.
[0045] like Figure 1As shown, this invention proposes a visual navigation method for agricultural robots based on vehicle speed adaptive gimbal perspective. This method dynamically adjusts the gimbal's observation perspective through real-time vehicle speed perception and combines this with acquired depth information to achieve precise navigation line positioning and vehicle coordinate transformation, ultimately generating reliable visual navigation commands. Specifically, it includes the following steps:
[0046] Step 1: System Initialization and Speed Sensing. Start the Intel RealSense D435i depth camera mounted on the two-axis gimbal and initialize the gimbal controller. Obtain the agricultural robot's real-time speed V (m / s) using a Doppler velocimeter, with a sampling frequency of at least 10Hz.
[0047] Step Two: Vehicle Speed Adaptive Gimbal View Adjustment. Based on the real-time driving speed V, the gimbal controller controls the gimbal's observation angle (i.e., horizontal pitch angle), thereby dynamically adjusting the camera's vertical viewing angle. For example... Figure 2 As shown:
[0048] If V < 0.5 m / s (low-speed fine operation), the gimbal controller controls the horizontal pitch angle of the gimbal to adjust the vertical viewing angle of the camera to 60 degrees, focusing on the nearby crop rows within 2-5 meters in front of the vehicle;
[0049] If 0.5m / s≤V<1.5m / s (medium-speed operation), the gimbal controller controls the horizontal pitch angle of the gimbal to adjust the vertical viewing angle of the camera to 45 degrees, focusing on the nearby crop rows within 5-10 meters in front of the vehicle;
[0050] If V≥1.5m / s (high-speed driving), the gimbal controller controls the horizontal pitch angle of the gimbal, adjusting the vertical viewing angle of the camera to 30 degrees, mainly observing the distant trend within a range of 10-20 meters in front of the vehicle.
[0051] Step 3: Image Acquisition and Crop Row Feature Extraction. Under the camera's adjusted pan-tilt-zoom (PTZ) viewpoint, the camera simultaneously acquires RGB and depth image streams of the field scene. From the acquired images, the RGB images are converted to TGBR (Transient Transformer Baseline) supergreen index (TGBR = 2G-RB) maps to enhance the distinguishability of crop rows from other background features. Otsu's method is used to adaptively threshold the TGBR supergreen index map to obtain a preliminary binarized image of the crop rows. Figure 3 The morphological filtering is performed using the closing operation to remove interline speckle noise.
[0052] Step 4: Based on the binarized image of crop rows, generate skeleton lines by connecting centroid slices from different regions: Divide the observation area vertically into three sub-regions: upper, middle, and lower. Each sub-region should have 10% overlap to avoid interruptions in the crop rows. Then, use connected component analysis to calculate the centroid coordinates of the connected components within each sub-region. Specifically, within each sub-region, determine whether the longitudinal and lateral offsets between centroids meet the conditions in Table 1. Connect the centroids that meet the conditions with line segments to ultimately form the crop row skeleton lines (e.g., ...). Figure 4 (As shown).
[0053] Table 1. Centroid Connectivity Relationships
[0054] Intercentrifugal connectivity Horizontal offset absolute value Vertical offset absolute value upper part <20 pixels <40 pixels Central <30 pixels <80 pixels lower part <50 pixels <120 pixels
[0055] Step 5: Global navigation line fitting. The overall contour of all crop row skeleton lines is extracted using the maximum outer envelope method, as shown in Figure 5(a). Then, the geometric features (width and height, in pixels) of each crop row skeleton line contour are calculated. If the contour height of a crop row is less than 0.7 times the image height, or the contour width is greater than 0.3 times the image width, then the crop row does not meet the navigation scenario and is removed. Finally, the contours of the target crop rows within the working range of the agricultural robot are selected, and then the global navigation line of the crop row is fitted using the least squares method, as shown in Figure 5(b).
[0056] Step Six: Calculation of 3D coordinates for the global navigation line of the crop row and transformation of the vehicle's coordinate system. Combined with... Figure 6 , Figure 7 :
[0057] Select the endpoint of the global navigation line of the crop row, query the depth value corresponding to the point in the aligned depth image, that is, obtain the depth value Zc of the bottom endpoint, use Zhang Zhengyou calibration method to obtain camera calibration parameters (including intrinsic parameter matrix I, rotation matrix R, translation vector T), and calculate the world coordinates of the endpoint of the global navigation line of the crop row in the world coordinate system through inverse transformation of perspective projection model.
[0058] Using the same method, calculate the world coordinates of the endpoints of the vehicle's centerline.
[0059] Calculate the lateral deviation ΔX = P_w1.x - P_c1.x, and the heading angle. °, where P_c1.x is the x-coordinate of the bottom endpoint of the vehicle centerline in the world coordinate system, and P_w1.x represents the x-coordinate of the bottom endpoint of the crop navigation line in the world coordinate system. This represents the ordinate of the bottom endpoint of the crop row navigation line in the world coordinate system. This represents the x-coordinate of the top endpoint of the crop row navigation line in the world coordinate system. This indicates the ordinate of the top endpoint of the crop row navigation line in the world coordinate system.
[0060] Self-vehicle coordinate transformation: Based on the geometric relationship between the camera installation position (X_c, Y_c, Z_c) and the front axle center position (X_v, Y_v, Z_v) of the agricultural robot, the lateral deviation ΔX and heading angle θ are converted into actual control deviations ΔX_control and θ_control in the self-vehicle coordinate system with the front axle center as the origin through spatial translation.
[0061] Step 7: Exception Handling and Command Transmission. Data is encapsulated in the format "#VIS, Lateral Deviation, Heading Angle, Status Identifier\r\n", where "YY" indicates valid data and "XX" indicates invalid data. First, the calculation results are verified: if |θ_control|>20° or the crop row navigation line fitting fails (the program has no crop row global navigation line data input, so fitting is considered failed), the status identifier is "XX", and the "GNSS + Machine Vision" combined navigation safety strategy (existing technology) is enabled. Otherwise, the status identifier is "YY", and closed-loop control is performed using machine vision navigation deviation data (including ΔX_control and θ_control). This data is sent via serial port to the agricultural robot's steering controller, which parses the status identifier and sends it to the steering actuator for execution.
[0062] This invention discloses an agricultural robot visual navigation system based on a vehicle speed adaptive gimbal perspective. The system implements the machine vision navigation method using C++ and OpenCV, enabling real-time data interaction with the underlying vehicle motion controller to assist or independently complete autonomous driving navigation operations in the field. The system includes a two-axis gimbal, a depth camera, a Doppler velocimeter, a navigation industrial control computer, a serial communication module, and a steering controller. The depth camera is mounted on the two-axis gimbal, the Doppler velocimeter is mounted above the left front wheel of the agricultural robot, and the navigation industrial control computer, serial communication module, and steering controller are integrated and mounted at the center of the agricultural robot. The controller of the two-axis gimbal, the depth camera, the Doppler velocimeter, and the steering controller are connected to the navigation industrial control computer via the serial communication module.
[0063] Depth cameras capture color and depth images of field scenes;
[0064] The Doppler velocimeter obtains the current speed of the agricultural robot in real time;
[0065] The navigation industrial control computer is used to acquire binarized images of crop rows, generate crop row skeleton lines, fit global navigation lines for crop rows, and calculate the heading angle and lateral deviation that the agricultural robot needs to adjust.
[0066] The steering controller receives and parses machine vision navigation deviation data, and then sends it to the steering actuator.
[0067] Ultimately, this invention significantly improves the visual navigation performance of agricultural robots in multiple operational scenarios by implementing adaptive speed control from the gimbal's perspective, connecting partitioned center-of-gravity slices, and achieving precise 3D positioning and autonomous vehicle coordinate transformation. This provides an effective technical solution for the precise and autonomous row-by-row operation of unmanned agricultural machinery in the field.
[0068] The embodiments described above are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the protection scope of the present invention.
Claims
1. A visual navigation method for agricultural robots based on vehicle speed adaptive gimbal viewpoint, characterized in that: Real-time acquisition of the agricultural robot's speed V; Based on the travel speed V, the gimbal controller controls the horizontal pitch angle of the gimbal, thereby dynamically adjusting the vertical viewing angle of the camera. Under the camera's field of view after gimbal adjustment, color and depth images of the field scene are acquired, and the color images are processed using an adaptive color feature enhancement and segmentation method to obtain crop row binarized images. Based on the perspective of the operation scene from far to near, the crop row skeleton line is generated in the binarized image by using the regional centroid slicing and connection method. The maximum outer envelope method is used to extract the overall contour of all crop row skeleton lines, and the contour of the target crop row within the operation range of the agricultural robot is selected. Then, the least squares method is used to fit the global navigation line of the crop row, and the pixel coordinates of the endpoints of the global navigation line are transformed to the world coordinate system by combining depth information. Based on the global navigation line of the crop row and the center line of the vehicle body in the world coordinate system, the heading angle and lateral deviation of the agricultural robot are calculated. The autonomous vehicle coordinate system is transformed based on the center position of the front axle of the autonomous vehicle in order to correct the heading angle and physical lateral deviation required for actual control. Based on the validity of the navigation results, control commands are generated and sent to the steering controller.
2. The agricultural robot visual navigation method according to claim 1, characterized in that, Based on the travel speed V, the gimbal controller controls the horizontal pitch angle of the gimbal, thereby dynamically adjusting the vertical viewing angle of the camera, specifically: If V < 0.5 m / s, adjust the camera's vertical angle to 60 degrees and focus on the nearby crop rows within 2-5 meters in front of the vehicle; If 0.5m / s≤V<1.5m / s, adjust the camera's vertical angle to 45 degrees and focus on the nearby crop rows within a range of 5-10 meters in front of the vehicle; If V≥1.5m / s, adjust the camera's vertical viewing angle to 30 degrees and observe the distant trend within a range of 10-20 meters in front of the vehicle.
3. The agricultural robot visual navigation method according to claim 1, characterized in that, The adaptive color feature enhancement and segmentation method specifically involves: calculating the TGBR super green index of the color image and using the Otsu method for adaptive threshold segmentation to obtain a crop row binarized image.
4. The agricultural robot visual navigation method according to claim 1, characterized in that, The crop row skeleton line is generated by the regional centroid slicing connection method. Specifically, the observation area is divided into three sub-regions in the vertical direction: upper, middle and lower, with a 10% overlap between adjacent sub-regions; the connected domains in each sub-region are calculated, and the centroid coordinates of the connected domains are obtained; within each sub-region, it is determined whether the longitudinal and lateral offsets between the centroid coordinates are subject to constraints, and the centroids that meet the conditions are connected by line segments to form the crop row skeleton line.
5. The agricultural robot visual navigation method according to claim 1, characterized in that, The process of filtering out the target crop row outlines within the operating range of the agricultural robot involves: calculating the geometric features of the skeleton line outline of each crop row; if the crop row outline height is less than 0.7 times the image height or the outline width is greater than 0.3 times the image width, then the crop row does not meet the navigation scenario and is removed.
6. The agricultural robot visual navigation method according to claim 1, characterized in that, The pixel coordinates of the endpoints of the global navigation line are transformed to the world coordinate system. Specifically, the endpoints of the global navigation line of the crop row are selected, the depth value corresponding to the point in the aligned depth image is queried, the camera calibration parameters are obtained using the Zhang Zhengyou calibration method, and the coordinates of the endpoints of the global navigation line of the crop row in the world coordinate system are calculated by inverse transformation of the perspective projection model.
7. The agricultural robot visual navigation method according to claim 6, characterized in that, The heading angle that the agricultural robot needs to adjust °, where P_w1.x represents the x-coordinate of the bottom endpoint of the crop row navigation line in the world coordinate system. This represents the ordinate of the bottom endpoint of the crop row navigation line in the world coordinate system. This represents the x-coordinate of the top endpoint of the crop row navigation line in the world coordinate system. The vertical coordinate of the top endpoint of the crop row navigation line in the world coordinate system is represented; the lateral deviation ΔX required to be adjusted by the agricultural robot is ΔX = P_w1.x - P_c1.x, where P_c1.x is the horizontal coordinate of the bottom endpoint of the vehicle centerline in the world coordinate system.
8. The agricultural robot visual navigation method according to claim 7, characterized in that, The vehicle coordinate system transformation is achieved by converting the lateral deviation ΔX and heading angle θ into the actual control deviations ΔX_control and θ_control in the vehicle coordinate system with the front axle center as the origin through spatial translation.
9. The agricultural robot visual navigation method according to claim 8, characterized in that, The control command format is "#VIS, Lateral Deviation, Heading Angle, Status Identifier\r\n", where the status identifier "YY" indicates valid data and "XX" indicates invalid data. The actual control deviation is verified: if |θ_control|>20° or crop row navigation line fitting fails, the status identifier is "XX", and the "GNSS+machine vision" combined navigation safety strategy is enabled; otherwise, the status identifier is "YY", and ΔX_control and θ_control are sent to the steering controller of the agricultural robot.
10. A system for implementing the visual navigation method for agricultural robots according to any one of claims 1-9, characterized in that, The system includes a depth camera, a Doppler velocimeter, a navigation industrial control computer, and a steering controller. The depth camera is mounted on a two-axis gimbal, the Doppler velocimeter is mounted above the left front wheel of the agricultural robot, and the navigation industrial control computer and steering controller are integrated and mounted at the center of the agricultural robot. The controller of the two-axis gimbal, the depth camera, the Doppler velocimeter, and the steering controller are all connected to the navigation industrial control computer through a serial communication module. The depth camera captures color and depth images of the field scene; The Doppler velocimeter acquires the current speed of the agricultural robot in real time; The navigation industrial control computer is used to acquire binarized images of crop rows, generate crop row skeleton lines, fit global navigation lines of crop rows, and calculate the heading angle and lateral deviation that the agricultural robot needs to adjust. The steering controller is used to receive and parse machine vision navigation deviation data, and then send it to the steering actuator.