A mowing robot navigation method and system

Abstract

The application discloses a cutting robot navigation method and system, relates to the cutting robot navigation field, and is used for improving the navigation precision of the cutting robot and comprises the following steps: acquiring environment information of different types in front; the environment information of different types comprises image data and radar point cloud data; determining target crop path information independently recognized according to respective environment information; respectively evaluating the reliability scores of the target crop path information independently recognized according to respective environment information; comparing the spatial geometric consistency of the target crop path information independently recognized according to respective environment information and respective reliability scores, and quantifying the overall credibility of current environment information; when the overall credibility is greater than or equal to a credibility threshold, navigation is performed according to the environment information of different types; and when the overall credibility is less than the credibility threshold, navigation is performed according to internal dead reckoning.

Description

Technical Field

[0001] This invention relates to the field of navigation for green-cutting robots, and more particularly to a navigation method and system for green-cutting robots. Background Technology

[0002] In scenarios where harvesting robots autonomously navigate farmland, their navigation systems typically combine satellite positioning with local environmental perception modules. However, some field surfaces often exhibit subtle undulations that are difficult to detect with the naked eye, and the degree of soil compaction varies in different areas. When the harvesting robot moves through such fields, the contact between its tracks or wheels and the ground changes. For example, when traversing a relatively loose, moist patch of soil, the wheels may experience slight slippage; while when encountering a locally compacted, harder area, the robot's body may experience slight pitch or roll angle changes, leading to a momentary deviation between the robot's actual ground speed and the system commands. Simultaneously, these changes in the robot's attitude cause the height and tilt angle of its onboard local perception sensors (such as stereo vision cameras and LiDAR) relative to the ground to become less constant, introducing potential error accumulation for subsequent precise navigation.

[0003] Furthermore, in dense areas, the leaves of adjacent plants may obscure each other, making it difficult to identify individual plant features and hindering the accurate extraction of clear "channel" boundaries during visual processing. In sparse areas, the gaps between plants are too large, even resulting in broken rows, blurring the concept of "rows" and potentially causing the system to misjudge or lose its path. Similarly, the echo signal from the lidar is also affected. In dense areas, the echo point cloud may be too dense to distinguish individual plants, while in sparse areas, there may be too few echo points to construct a complete row structure. This unevenness in crop growth introduces inherent uncertainty into the local perception data, further exacerbating the difficulty of fine-tuning the navigation system.

[0004] Furthermore, the shape, color, and height of non-target plants such as weeds in the field may resemble the target crop (silage corn). The cutting robot may misidentify these weeds or free-growing seedlings as part of the target crop, or treat them as interference. This could cause the robot to make incorrect lateral adjustments, deviating from the actual corn planting rows. The lidar may also receive echoes from these weeds, confusing them with corn plants, further interfering with the judgment of the actual crop boundaries.

[0005] Finally, during prolonged operation, the inherent characteristics of the harvesting robot and external environmental factors cumulatively impact its sensor performance. For example, a large amount of particulate matter gradually adheres to and accumulates on the protective cover of the robot's stereo vision camera lens and the transmission and reception windows of the LiDAR, causing a continuous decline in the quality of data acquired by the local perception system. This means that even if the robot is in a relatively ideal crop-growing area, its navigation system may fail to acquire sufficiently reliable local environmental information due to the reduced "vision" of its sensors. This affects its ability to determine the row center and make fine adjustments, potentially causing the robot to deviate from its preset path or even damage the crops. Summary of the Invention

[0006] This invention provides a navigation method for a green-cutting robot, which improves the navigation accuracy of the green-cutting robot.

[0007] Firstly, to address the aforementioned technical problems, this invention provides a navigation method for a crop-cutting robot, comprising: acquiring different types of environmental information ahead; the different types of environmental information include image data and radar point cloud data; determining target crop path information independently identified based on their respective environmental information, and evaluating the reliability scores of the target crop path information independently identified based on their respective environmental information; comparing the spatial geometric consistency of the target crop path information independently identified based on their respective environmental information with their respective reliability scores to quantify the overall credibility of the current environmental information; when the overall credibility is greater than or equal to a credibility threshold, navigation is performed based on the different types of environmental information; when the overall credibility is less than the credibility threshold, navigation is performed based on internal dead reckoning.

[0008] Optionally, the reliability scores of target crop path information independently identified based on their respective environmental information are evaluated separately, including: when the environmental information is image data, performing binarization processing on the image data; extracting edges from the image data based on the binarized image data; identifying and fitting multiple line segments from the edges; using the line segments to indicate the possible boundaries of the crop; and calculating the image reliability score based on the number of line segments, parallelism, length, and edge sharpness of the multiple line segments.

[0009] Optionally, an image reliability score is calculated based on the number of line segments, parallelism, length, and edge sharpness of the image. This includes calculating the image reliability score using the image reliability score calculation formula: F1 = w1 * (S / S) MAX F1 = (w1 + w2 * L + w3 * Q) / (w1 + w2 * Q); where F1 represents the image reliability score, w1, w2, and w3 represent weighting coefficients, and S represents the number of valid line segments. MAXL represents the expected maximum number of line segments; Q represents the average line segment parallelism; and L represents the average edge sharpness. The average line segment parallelism can be determined by calculating the standard deviation of the angles between all fitted line segments. The average edge sharpness can be determined by calculating the average gradient value of the crop row edge region.

[0010] Optionally, the reliability scores of target crop path information independently identified based on their respective environmental information are evaluated separately, including: when the environmental information is radar point cloud data, preprocessing the radar point cloud data; the preprocessing includes: noise point filtering, and removing points below a specific height threshold based on the installation height of the lidar and ground slope information to exclude ground points; identifying crop point clusters in the radar point cloud data; performing straight line fitting on the center points or boundary points of these point clusters to identify the spatial boundary lines formed by crop stems; and calculating the radar reliability score based on the completeness of the clustering results, the linearity of the fitted straight line, and the uniformity of the crop row spacing.

[0011] Optionally, based on the completeness of the clustering results, the linearity of the fitted line, and the uniformity of the crop row spacing, a radar reliability score is calculated, including: calculating the radar reliability score according to the radar reliability score calculation formula; the radar reliability score calculation formula includes: F2=w4*(J / J MAX F2 + w5*N + w6*H; where F2 represents the radar reliability score, w4, w5, and w6 represent weighting coefficients; J represents the number of effective clusters; J MAX The expected maximum number of clusters is represented by N; the mean linearity of the fit is represented by H; the mean linearity of the fit is represented by the reciprocal of the standard deviation of the crop row spacing; the mean linearity of the fit is determined by the root mean square error of the fit residuals.

[0012] Optionally, the spatial geometric consistency of target crop path information independently identified based on their respective environmental information is compared, including: sampling within the target crop path information independently identified based on their respective environmental information to obtain a series of corresponding point pairs; calculating the average Euclidean distance deviation between the point pairs; the average Euclidean distance deviation is expressed as:

[0013]

[0014] Among them, O L This represents the average Euclidean distance deviation, where N is the number of sampling points, and X... i Here, x is the x-coordinate of the sampling point, sqrt represents the square root function, and fv(X) i fl(X) represents the function representation of target crop path information independently identified from image data; i ) represents a function that independently identifies the target crop path information based on radar point cloud data.

[0015] Optionally, the overall credibility of the current environmental information can be quantified, including: determining the overall credibility based on the average Euclidean distance deviation; the overall credibility satisfies the following relationship:

[0016]

[0017] F Z Indicates overall credibility; O Lmax This indicates the maximum permissible deviation.

[0018] Optionally, navigation is performed based on internal dead reckoning, including: acquiring a virtual trajectory; the virtual trajectory is a smooth extension of the position and attitude of the cutting robot when the overall confidence level is less than the confidence level threshold; following the virtual trajectory based on steering commands to navigate based on internal dead reckoning; the steering commands are determined based on the deviation between the dead reckoning results and the virtual trajectory.

[0019] Optionally, the method further includes: identifying crop pixel regions in image data; identifying crop point clusters in radar point cloud data; projecting the crop point clusters onto the image plane to obtain projection points; for each projection point, determining whether it falls within the central confidence region of the corresponding crop pixel region; calculating the proportion of projection points falling outside the central confidence region to the total number of projection points to obtain the projection deviation ratio; when the projection deviation ratio exceeds a preset threshold, determining that there is a consistency deviation phenomenon, and correcting the overall confidence level based on the consistency deviation phenomenon.

[0020] Secondly, the present invention provides a navigation system for a green-cutting robot, the system comprising:

[0021] The environmental information acquisition module is used to acquire different types of environmental information ahead; these different types of environmental information include image data and radar point cloud data.

[0022] The path identification and evaluation module is used to determine the target crop path information independently identified based on their respective environmental information; and to evaluate the reliability score of the target crop path information independently identified based on their respective environmental information.

[0023] The perception information mutual verification module is used to compare the spatial geometric consistency of target crop path information independently identified based on their respective environmental information and their respective reliability scores to quantify the overall credibility of the current environmental information.

[0024] The navigation strategy adjustment module is used to navigate based on different types of environmental information when the overall credibility is greater than or equal to the credibility threshold.

[0025] When the overall credibility is less than the credibility threshold, navigation is performed based on internal dead reckoning.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] The navigation method for a crop-cutting robot disclosed in this application acquires different types of environmental information ahead (including image data and radar point cloud data), independently identifies the target crop path information, and evaluates the reliability scores of each. Based on this, the spatial geometric consistency of the paths identified by different types of environmental information and their respective reliability scores are compared to quantify the overall credibility of the current environmental information. When the overall credibility reaches a preset threshold, the system navigates based on multi-source environmental information; when the overall credibility is below the threshold, it switches to internal dead reckoning for navigation.

[0028] This method effectively addresses the problem in existing technologies where the reliability of local perception data is insufficient due to the complex and variable farmland environment (such as undulating ground, uneven crop growth, interference from non-target plants, and degraded sensor performance), thus affecting navigation accuracy and robustness. Specifically, by introducing a fusion and mutual verification mechanism of multi-source environmental information (images and radar point clouds), this application overcomes the limitations of a single sensor in harsh environments (such as changes in lighting, dust, and crop occlusion), improving the comprehensiveness and accuracy of environmental perception. By independently assessing the reliability of each sensor's data and quantifying it in conjunction with spatial geometric consistency, the system can determine the quality of the current perception information in real time, avoiding navigation errors caused by low-quality data. Furthermore, when external perception information is unreliable, it smoothly switches to internal dead reckoning navigation, ensuring the continuity and safety of robot operations and preventing operation interruptions or crop damage due to perception failure. In summary, the method of this application significantly improves the navigation robustness, reliability, and adaptability of the harvesting robot in complex farmland environments, ensuring efficient and precise agricultural operations. Attached Figure Description

[0029] Figure 1 This is a schematic flowchart of a green-cutting robot navigation method provided in an embodiment of the present invention;

[0030] Figure 2 This is a schematic diagram of another navigation method for a green-cutting robot provided in an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of a navigation system for a greening robot provided in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0033] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0034] The following specific embodiments will provide a detailed description and explanation of the navigation method for a greening robot provided in this application.

[0035] Reference Figure 1 This invention provides a navigation method for a green-cutting robot, comprising the following steps:

[0036] S1, obtain different types of environmental information ahead.

[0037] The different types of environmental information include image data and radar point cloud data.

[0038] As one possible implementation, the cutting robot could be equipped with a vision camera and a LiDAR. The system could acquire image data of the front using the robot's vision camera and radar point cloud data of the front using the LiDAR.

[0039] The vision camera and lidar can be installed at the front of the cutting robot and precisely calibrated to ensure that their data can be accurately converted into the cutting robot's own coordinate system.

[0040] S2. Determine the target crop path information independently identified based on their respective environmental information; and evaluate the reliability score of the target crop path information independently identified based on their respective environmental information.

[0041] In this context, target crop path information refers to the center line or boundary line of the crop row that is identified based on this environmental information and instructs the robot to travel along it. For example, in a cornfield, target crop path information could be a geometric representation of the channels between rows formed by corn plants.

[0042] The reliability score is a quantitative assessment of the quality of target crop path information independently identified for each type of environmental information. The higher the score, the more reliable and accurate the path information is.

[0043] I. The process of independently identifying target crop path information based on image data may include:

[0044] Image preprocessing: First, the original image is color space converted (e.g., from RGB to HSV), and filtering operations such as Gaussian blur are applied to reduce image noise.

[0045] Crop region extraction: By utilizing a specific range of the green channel in the HSV color space (e.g., H value between 30-90, S value between 40-255, and V value between 30-255), the image is binarized to initially separate potential crop regions.

[0046] Edge and line detection: The Canny edge detection algorithm is applied to the binarized image to extract all salient edges. Subsequently, the Hough Transform or Random Sample Consensus (RANSAC) algorithm is used to identify and fit multiple straight or curve segments from these edges. These line segments represent the possible boundaries of crop rows, thus obtaining target crop path information independently identified based on image data.

[0047] II. The process of independently identifying target crop path information based on radar point cloud data may include:

[0048] Point cloud filtering and ground removal: First, noise points are filtered out from the raw point cloud data (e.g., through statistical filtering or radius filtering). Since the crop-cutting robot primarily focuses on crop rows, the system removes points below a specific height threshold (e.g., 0.1 meters) based on the LiDAR installation height and ground slope information to exclude ground points.

[0049] Crop point clustering and row identification: Density-based clustering algorithms (e.g., DBSCAN) are applied to the remaining point cloud data to cluster points belonging to the same plant or clump of crops. Subsequently, straight line fitting (e.g., least squares or RANSAC) is performed on the center points or boundary points of these point clusters to identify the spatial boundary lines formed by crop stems, thus obtaining target crop path information independently identified based on radar point cloud data.

[0050] Furthermore, the system calculates an image reliability score based on the number of identified line segments, their parallelism, length, and continuity in the image. The radar reliability score is determined based on the completeness of the clustering results, the linearity of the fitted lines, and the uniformity of the crop row spacing.

[0051] S3. Compare the spatial geometric consistency of the target crop path information independently identified based on their respective environmental information and their respective reliability scores to quantify the overall credibility of the current environmental information.

[0052] Spatial geometric consistency refers to the degree of matching in spatial location and morphology between target crop path information independently identified by different types of sensors (such as image and radar). Higher consistency indicates that the perception results of different sensors on the same scene are more consistent.

[0053] Among them, overall credibility is a quantitative indicator of the overall credibility of the current navigation data after comprehensively considering the reliability scores of different environmental information and their spatial geometric consistency.

[0054] As one possible implementation, the system can sample from the target crop path information independently identified based on their respective environmental information to obtain a series of corresponding point pairs; calculate the average Euclidean distance deviation between the point pairs; and determine the overall reliability based on the average Euclidean distance deviation.

[0055] The overall credibility satisfies the following relationship:

[0056]

[0057] F Z Indicates overall credibility; O Lmax This indicates the maximum permissible deviation. F1 represents the image reliability score. F2 represents the radar reliability score. O L This represents the average Euclidean distance deviation.

[0058] S4. When the overall credibility is greater than or equal to the credibility threshold, navigation is performed based on different types of environmental information.

[0059] The credibility threshold can be set as needed and is not restricted.

[0060] As one possible implementation, the system can employ a high-gain proportional-integral-derivative (PID) controller or a model predictive control (MPC) algorithm, taking the lateral and heading deviations between the robot's current position and the target path as inputs to calculate precise steering and speed commands for robot navigation.

[0061] For example, a steering instruction can be expressed as:

[0062] Steering command = High lateral deviation K_p + High heading deviation K_d

[0063] Among them, K_p and K_d are the higher control gains.

[0064] S5. When the overall credibility is less than the credibility threshold, navigation is performed based on internal dead reckoning.

[0065] Internal dead reckoning is a navigation method based on the robot's own kinematic model and data from internal sensors (such as encoders and inertial measurement units, IMUs). It serves as a backup or supplementary navigation method when external sensing data is unreliable.

[0066] As one possible implementation, the system can acquire a virtual trajectory and follow it based on steering commands to navigate using internal dead reckoning.

[0067] The turning instruction is determined based on the deviation between the dead reckoning result and the virtual trajectory.

[0068] Among them, the virtual trajectory is a smooth extension based on the position and posture of the cutting robot when the overall credibility is less than the credibility threshold.

[0069] As another possible implementation, the system can acquire a virtual trajectory when the overall credibility is less than the credibility threshold for a period of time that is less than the duration threshold, and follow the virtual trajectory based on the turning command to perform navigation based on internal dead reckoning.

[0070] The duration threshold can be set as needed. For example, it can be 3 seconds.

[0071] In some embodiments, if the duration of an overall confidence level less than the confidence level threshold exceeds the duration threshold, it indicates that the robot is unable to obtain any reliable local environmental information. The system will then trigger an emergency shutdown procedure or send an alarm to a remote operator via the wireless communication module, requesting manual intervention to prevent further path deviation or potential crop damage.

[0072] The navigation method for the harvesting robot in this application significantly improves the robot's navigation robustness and accuracy in complex farmland environments by introducing reliability assessment and spatial geometric consistency comparison of multi-source environmental information, and dynamically adjusting the navigation strategy based on the quantified overall credibility. Traditional navigation methods often rely excessively on single sensors or simple fusion, which can easily lead to navigation failure when sensor data is interfered with or degraded in quality. For example, in scenarios with sparse crops, weed interference, or contaminated sensors, a single vision or lidar system may not provide accurate path information. This application, by independently identifying and reliably assessing image data and radar point cloud data, and further verifying them through spatial geometric consistency, can more comprehensively and accurately determine the overall credibility of the current environmental information. When the credibility is high, it fully utilizes multi-source sensing data for precise navigation; when the credibility is low, it promptly switches to a more stable internal dead reckoning mode to avoid navigation deviations caused by erroneous sensing data. This adaptive navigation strategy enables the harvesting robot to maintain stable and reliable operation capabilities under various complex and changing farmland conditions, effectively solving the problem of insufficient navigation robustness in existing technologies and improving operational efficiency and safety.

[0073] In one possible design, such as Figure 2 As shown, in order to evaluate the reliability score of target crop path information independently identified based on their respective environmental information, this application may further include the following steps:

[0074] S101. When the environmental information is image data, perform binarization processing on the image data.

[0075] The purpose of binarization is to convert pixels in an image into only two possible values ​​(usually black and white) based on their grayscale or color values, thereby simplifying image information and highlighting the contrast between crops and the background.

[0076] As one possible implementation, the system can set a global or local threshold to set pixels in the image above the threshold to white (representing crops or foreground) and pixels below the threshold to black (representing background).

[0077] S102. Based on the image data processed by binarization, extract the edges in the image data.

[0078] Edge extraction aims to identify regions in an image where there are significant changes in brightness or color; these regions typically correspond to the boundaries of objects.

[0079] For example, classic edge detection algorithms such as the Canny operator, Sobel operator, or Prewitt operator can be used to identify edge information in an image. These algorithms determine the edge location by calculating the gradient of image pixels, thus obtaining a clear crop boundary outline.

[0080] S103. Identify and fit multiple line segments from the edges.

[0081] The line segments are used to indicate the possible boundaries of the crop.

[0082] As one possible implementation, the system can utilize line detection algorithms such as the Hough Transform to identify and fit continuous line segments from discrete edge pixels. These fitted line segments are considered as potential boundary representations of crop rows or individual crop stalks.

[0083] In this way, complex crop image information can be abstracted into a series of geometric line segments, which facilitates subsequent analysis and processing.

[0084] S104. Calculate the image reliability score based on the number of line segments, parallelism, length, and edge sharpness of the image.

[0085] The number of line segments reflects the richness of the identified crop boundaries; parallelism measures the alignment between these line segments, typically crop rows are approximately parallel; length indicates the integrity of the identified crop boundaries; and image edge sharpness reflects image quality and the clarity of the crop boundaries. These parameters, taken together, quantify the reliability of image data in indicating crop paths.

[0086] As one possible implementation, the system can calculate the image reliability score based on the image reliability score calculation formula;

[0087] The formula for calculating the image reliability score is:

[0088] F1 = w1 * (S / S) MAX )+w2*L+w3*Q

[0089] Where F1 represents the image reliability score, w1, w2, and w3 represent weighting coefficients; S represents the number of valid line segments; S MAX L represents the expected maximum number of line segments; Q represents the average line segment parallelism; and L represents the average edge sharpness. The average line segment parallelism can be determined by calculating the standard deviation of the angles between all fitted line segments. The average edge sharpness can be determined by calculating the average gradient value of the crop row edge region.

[0090] The above technical solution provides an effective method for harvesting robots to evaluate the reliability of target crop path information based on image data. This method, through image preprocessing, feature extraction, and geometric fitting, can accurately identify crop boundaries in complex farmland environments and transform them into quantifiable line segment features. By comprehensively considering the number, parallelism, length, and edge sharpness of line segments, an image reliability score can be objectively and accurately calculated, providing a reliable basis for subsequent navigation decisions. This helps improve the navigation accuracy and robustness of harvesting robots under different lighting conditions, crop growth statuses, and environmental complexity, avoiding navigation deviations or erroneous decisions caused by poor image information quality.

[0091] In one possible design, in order to separately evaluate the reliability scores of target crop path information independently identified based on their respective environmental information, this application further includes the following steps:

[0092] S201. When the environmental information is radar point cloud data, preprocess the radar point cloud data.

[0093] The preprocessing includes: filtering out noise points and removing points below a specific height threshold based on the installation height of the lidar and the ground slope information to exclude ground points.

[0094] Specifically, when the system senses an external environmental fluctuation event, such as a sudden drop in temperature or a sudden change in solar radiation intensity, it will immediately extract the power data sequence associated with the period before and after the occurrence of the fluctuation event from the historical operating data of the intelligent environmental control equipment.

[0095] S202. Identify crop clusters in radar point cloud data.

[0096] Crop point clusters refer to dense collections of points formed in a point cloud by one or more crop stems and leaves.

[0097] As one possible approach, crop point cluster identification can be achieved through various clustering algorithms, such as DBSCAN (density-based spatial clustering with noise), K-means, or connectivity-based region growing algorithms. These algorithms can group points belonging to the same crop together to form independent crop point clusters based on their spatial density and distance relationships.

[0098] S203. Perform straight line fitting on the center points or boundary points of these point clusters to identify the spatial boundary lines formed by crop stems.

[0099] These spatial boundary lines typically represent the direction of crop rows.

[0100] As one possible approach, line fitting can employ least squares, the RANSAC algorithm, or other robust fitting methods to extract a representative line model from a discrete cluster of points. The choice of center or boundary points of the cluster for fitting depends on the specific crop morphology and fitting strategy, with the aim of accurately representing the spatial location and orientation of crop rows.

[0101] S204. Calculate the radar reliability score based on the completeness of the clustering results, the linearity of the fitted line, and the uniformity of the crop row spacing.

[0102] The completeness of clustering results measures the degree of match between the number of identified crop point clusters and the expected number of crop rows, for example, by calculating the ratio of the number of effective clusters to the expected maximum number of clusters. The linearity of the fitted line reflects the consistency between the fitted line and the actual crop rows, for example, by calculating the root mean square error of the fitting residuals. The uniformity of crop row spacing is determined by analyzing the distance distribution between adjacent fitted lines, for example, by calculating the reciprocal of the standard deviation of the row spacing; the smaller the standard deviation, the better the uniformity. These indicators collectively reflect the clarity and reliability of crop row features in radar point cloud data, thus quantifying the radar reliability score.

[0103] As one possible implementation, the system can calculate the radar reliability score based on the radar reliability score calculation formula;

[0104] The formula for calculating radar reliability score includes:

[0105] F2=w4*(J / J MAX )+w5*N+w6*H;

[0106] Where F2 represents the radar reliability score, w4, w5, and w6 represent weighting coefficients; J represents the number of effective clusters; J MAX The expected maximum number of clusters is represented by N; the mean linearity of the fit is represented by H; the mean linearity of the fit is represented by the reciprocal of the standard deviation of the crop row spacing; the mean linearity of the fit is determined by the root mean square error of the fit residuals.

[0107] This application's solution, tailored to the characteristics of radar point cloud data, employs a specialized preprocessing, crop cluster identification, spatial boundary line fitting, and multi-dimensional reliability assessment process. This effectively addresses the accuracy limitations of traditional general assessment methods when processing radar data. Specifically, noise point filtering and ground point removal ensure data purity, leading to more accurate subsequent crop identification. Crop cluster identification and spatial boundary line fitting efficiently extract key geometric information of crop rows from radar data. Furthermore, by comprehensively considering the completeness of clustering results, the linearity of the fitted lines, and the uniformity of crop row spacing, the quality and reliability of the crop path information identified by radar data can be comprehensively and objectively evaluated, thus providing a more reliable basis for subsequent navigation decisions.

[0108] The aforementioned technical solution significantly improves the accuracy and robustness of crop path information reliability assessment in radar point cloud data for harvesting robots. This targeted assessment method enables the robot to more accurately determine the availability of current radar sensing data, avoiding navigation deviations caused by poor radar data quality. Especially in environments where image data may be limited due to insufficient lighting or dust, accurate radar reliability score assessment ensures the robot can still navigate based on reliable sensing information, thereby improving the adaptability and operational efficiency of the harvesting robot in complex farmland environments and reducing the risk of misoperation.

[0109] In some preferred embodiments, a specific example is given below. Assume a crop-harvesting robot is operating in a farmland, its onboard LiDAR continuously acquiring point cloud data. When the robot enters an area where crops are sparsely grown or where weeds are present, traditional general reliability assessment methods may not accurately determine the reliability of the radar data. However, using the solution of this application, the radar point cloud data is first preprocessed. For example, sporadic noise points are removed through statistical filtering, and ground points below 0.1 meters are removed based on the LiDAR's installation height to ensure that only crop and obstacle points are retained. Subsequently, the DBSCAN clustering algorithm is used to identify crop clusters in the point cloud. For example, if three rows of crops are expected, the system will attempt to identify three main clusters. Next, a straight line is fitted to the center point of each identified crop cluster to obtain three spatial boundary lines representing the direction of the crop rows. Finally, the system calculates a radar reliability score. For example, if only two complete crop clusters are identified (low clustering completeness), the fitted straight line has significant curvature (poor linearity), or the spacing between adjacent crop rows fluctuates greatly (poor uniformity), the calculated radar reliability score will be low. This low radar reliability score will be comprehensively evaluated along with the reliability scores of other environmental information and may cause the robot to switch to internal dead reckoning navigation mode to avoid navigation errors caused by unreliable radar data, thereby ensuring the robot's safe and efficient operation in complex environments.

[0110] In one possible design, to compare the spatial geometric consistency of target crop path information independently identified based on their respective environmental information, this application further includes the following steps:

[0111] S301. Sampling is performed within the target crop path information independently identified based on their respective environmental information to obtain a series of corresponding point pairs.

[0112] After acquiring the target crop path information independently identified from image data and radar point cloud data, it is necessary to spatially compare these two path information sets. For this purpose, sampling can be performed along these path information sets. Sampling can employ a uniform sampling method, selecting a series of sampling points at fixed intervals along the length of the path; or an adaptive sampling method, increasing the sampling density in areas with significant changes in path curvature to more accurately capture the geometric features of the path. Through sampling, a series of corresponding point pairs can be obtained, where each point pair contains one point from the image path information and a corresponding point from the radar path information.

[0113] S302. Calculate the average Euclidean distance deviation between point pairs.

[0114] The average Euclidean distance deviation is expressed as:

[0115]

[0116] Among them, O L This represents the average Euclidean distance deviation, where N is the number of sampling points, and X... i Here, x is the x-coordinate of the sampling point, sqrt represents the square root function, and fv(X) i fl(X) represents the function representation of target crop path information independently identified from image data; i ) represents a function that independently identifies the target crop path information based on radar point cloud data.

[0117] This application's solution samples target crop path information independently identified from image data and radar point cloud data, and calculates their average Euclidean distance deviation, thereby achieving a quantitative assessment of the spatial geometric consistency of path information from two different sources. When the crop paths independently identified by the two sensor data are highly consistent in space, the Euclidean distance deviation between their corresponding point pairs will be small, resulting in a smaller average Euclidean distance deviation O_L. Conversely, if there are significant spatial differences between the two paths, the average Euclidean distance deviation O_L will be larger. This quantitative method provides an objective indicator for judging whether the paths identified by different sensor data corroborate each other, thus providing a key input for subsequent overall reliability quantification.

[0118] The above technical solution provides a precise and quantifiable method to assess the spatial geometric consistency between target crop paths identified by different types of environmental information (such as image data and radar point cloud data). This quantitative assessment avoids subjective judgment, enabling the harvesting robot to make a more accurate assessment of the reliability of current environmental information based on objective data. Therefore, when subsequently quantifying the overall reliability of current environmental information, the advantages of different sensor data can be more effectively integrated, improving the robustness and accuracy of navigation decisions, especially in complex farmland environments where single sensor data may have limitations or uncertainties.

[0119] In one possible design, in order to perform navigation based on internal dead reckoning, this application also includes:

[0120] S401, Obtain the virtual trajectory.

[0121] The virtual trajectory is a smooth extension based on the position and posture of the green-cutting robot when the overall credibility is less than the credibility threshold.

[0122] Specifically, a virtual trajectory can be understood as a predicted or smoothed path calculated by the robot system based on historical motion data from its internal sensors (such as odometry and inertial measurement units, IMUs) and its current state when external environmental perception information is unreliable. This virtual trajectory aims to provide a temporary, relatively stable navigation benchmark to address situations where external perception data is insufficient or inaccurate. The smooth extension of the virtual trajectory means that when external perception data is unreliable, the robot does not completely stop or move randomly, but rather predicts and generates a smooth, continuous path based on its most recent valid motion trends and attitude information to maintain navigation consistency.

[0123] S402: Follow the virtual trajectory based on the turning command to navigate according to internal dead reckoning.

[0124] The steering command is determined based on the deviation between the dead reckoning result and the virtual trajectory. This means the system continuously monitors the difference between the robot's current position and attitude, calculated internally through dead reckoning, and the preset virtual trajectory. Once a deviation is detected, a corresponding steering command is generated to guide the robot to adjust its direction and attitude, making it conform to the virtual trajectory as closely as possible. The dead reckoning result is the robot's current position and attitude calculated in real time based on its own kinematic model and internal sensor data (such as wheel speed encoders, gyroscopes, etc.).

[0125] This application's solution introduces a virtual trajectory as a navigation reference when external environmental information is unreliable. It then generates turning commands by combining the deviation between the internal dead reckoning results and this virtual trajectory. This allows the robot to maintain basic path tracking capabilities even in the absence of highly reliable external perception data. This mechanism ensures that the robot maintains a certain degree of autonomous navigation capability even under complex or harsh environmental conditions, avoiding complete stagnation or deviation from the predetermined path. Through the smooth extension of the virtual trajectory, the robot can achieve a smooth transition from multi-sensor fusion navigation to internal dead reckoning navigation, ensuring the continuity and stability of navigation.

[0126] Through the aforementioned technical solution, when the overall reliability of external environmental information is low, the logging robot can smoothly transition to a navigation mode based on internal dead reckoning. This method avoids navigation interruptions or path loss due to unreliable external perception data, significantly improving the robot's robustness and operational continuity in complex environments. By smoothly extending the virtual trajectory and using deviation-based steering commands, the robot achieves more stable and predictable autonomous navigation, effectively reducing operational risks and ensuring the smooth progress of logging operations.

[0127] However, in practical applications, even with high macroscopic spatial geometric consistency, subtle calibration errors between sensors, dynamic pose drift, or the complexity of the local environment (e.g., crop shading, changes in illumination) can still lead to inconsistencies in image data and radar point cloud data at the level of detail. Such inconsistencies, if not fully identified and quantified, can result in inaccurate assessments of the overall reliability of current environmental information, thereby affecting the robustness and accuracy of navigation decisions.

[0128] In response, this application further proposes a navigation method for a green-cutting robot, the method including:

[0129] S501, Identify crop pixel regions in image data.

[0130] Among them, identifying crop pixel regions in image data can be achieved using image segmentation techniques, such as semantic segmentation models based on deep learning. These models are trained to accurately distinguish between crop and non-crop regions in an image, thereby identifying crop pixel regions in the image data.

[0131] S502, Identify crop clusters in radar point cloud data.

[0132] Among them, identifying crop clusters in radar point cloud data can be achieved by using clustering algorithms, such as DBSCAN or K-means, to process the radar point cloud data and identify clusters representing single or multiple crop stems.

[0133] S503. Project the crop point clusters onto the image plane to obtain the projection points.

[0134] As one possible implementation, the system can transform the point coordinates in the 3D radar point cloud to the 2D image coordinate system, so as to project the crop point clusters onto the image plane and obtain the projection points.

[0135] S504. For each projection point, determine whether it falls within the central confidence region of the corresponding crop pixel area.

[0136] The central confidence region can be understood as a core region within the crop pixel area. Its size and shape can be adjusted according to actual application requirements. For example, it can be defined as a region extending outward from the geometric center of the crop pixel area within a certain pixel range, or it can be obtained from the original crop pixel area through morphological operations (such as erosion). This aims to eliminate areas with blurred or uncertain crop edges, focusing on the consistency of the main crop part.

[0137] S505. Calculate the proportion of the number of projection points that fall outside the central confidence region to the total number of projection points, and obtain the projection deviation ratio.

[0138] S506. When the projection deviation ratio exceeds the preset threshold, it is determined that there is a consistency deviation phenomenon, and the overall credibility is corrected according to the consistency deviation phenomenon.

[0139] The preset threshold can be determined based on experience or through experiments, and is used to define the acceptable range of consistency between sensors.

[0140] Once a consistency deviation is detected, the previously calculated overall reliability can be corrected accordingly. Correction methods may include, but are not limited to, reducing the calculated overall reliability score, or increasing the weight of dead reckoning navigation in subsequent navigation decisions to reflect the current decline in the reliability of multi-sensor fusion information.

[0141] Through the above technical solution, this application can effectively improve the robustness and reliability of the greening robot navigation system in complex environments. Specifically, by introducing cross-modal consistency verification of image and radar data, the system can more finely evaluate the quality of multi-sensor fusion information, making up for the shortcomings of relying solely on spatial geometric consistency and independent reliability scores for evaluation. When there are slight deviations or local inconsistencies in sensor data, this solution can promptly detect and quantify such deviations, thereby correcting the overall reliability and avoiding navigation risks caused by inaccurate perception information. Thus, even under conditions of limited sensor performance or complex and variable environments, the system can make more prudent and accurate navigation strategy adjustments, ensuring that the greening robot can safely and efficiently operate along the target crop path, significantly improving navigation accuracy and safety.

[0142] In some preferred embodiments, a specific example is given below. Suppose a harvesting robot is navigating in a field. After long-term operation, its image sensor and lidar sensor may experience slight relative pose drift due to vibration or temperature changes, resulting in a slight discrepancy in their spatial perception of the same crop row.

[0143] When the robot acquires information about the environment ahead:

[0144] First, the image data is processed to identify crop pixel regions, for example, by accurately delineating the boundaries of crop rows using a semantic segmentation model.

[0145] Meanwhile, radar point cloud data is processed to identify crop point clusters, for example, by using clustering algorithms to identify point cloud sets representing crop stems.

[0146] Next, these crop point clusters are projected onto the image plane, generating a series of projection points.

[0147] The system checks whether each projection point falls within the center confidence region of the corresponding crop pixel region in the image. For example, if a crop row is identified in the image, its center confidence region is defined as the 50% width range of the center of the crop row pixel region.

[0148] If a significant portion of the projection points (e.g., exceeding a preset threshold of 10%) are found to fall outside the central confidence region, this indicates an inconsistency between the image sensor and the radar sensor in their perception of the crop location, i.e., a consistency bias exists.

[0149] Based on this consistency deviation phenomenon, the system will correct the previously calculated overall credibility score. For example, even if the original overall credibility score is high, it will be reduced by a preset penalty value, or directly multiplied by a correction coefficient less than 1.

[0150] Therefore, the corrected overall reliability will more accurately reflect the true reliability of the current multi-sensor fusion information. If the corrected overall reliability is lower than the reliability threshold, the system will promptly switch to internal dead reckoning for navigation, thereby avoiding navigation errors caused by relying on inaccurate fusion information and ensuring that the robot can continue to operate safely and stably until sensor consistency is restored or environmental conditions improve.

[0151] like Figure 3 As shown, this embodiment of the invention also provides a navigation system for a green-cutting robot. The system includes:

[0152] The environmental information acquisition module is used to acquire different types of environmental information ahead; these different types of environmental information include image data and radar point cloud data.

[0153] The path identification and evaluation module is used to determine the target crop path information independently identified based on their respective environmental information; and to evaluate the reliability score of the target crop path information independently identified based on their respective environmental information.

[0154] The perception information mutual verification module is used to compare the spatial geometric consistency of target crop path information independently identified based on their respective environmental information and their respective reliability scores to quantify the overall credibility of the current environmental information.

[0155] The navigation strategy adjustment module is used to navigate based on different types of environmental information when the overall credibility is greater than or equal to the credibility threshold.

[0156] When the overall credibility is less than the credibility threshold, navigation is performed based on internal dead reckoning.

[0157] This application also provides a computer-readable storage medium. All or part of the processes in the above method embodiments can be executed by a computer program instructing related hardware. This program can be stored in the computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. The computer-readable storage medium can be an internal storage unit of the task execution device (including a data sending end and / or a data receiving end) of any of the foregoing embodiments, such as the hard disk or memory of the task execution device. The computer-readable storage medium can also be an external storage device of the terminal device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the terminal device. Further, the computer-readable storage medium can include both the internal storage unit of the task execution device and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the task execution device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0158] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0159] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, essentially, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0160] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A navigation method for a green-cutting robot, characterized in that, include: Acquire different types of environmental information ahead; these different types of environmental information include image data and radar point cloud data. Determine the target crop path information that is independently identified based on its respective environmental information, and evaluate the reliability score of the target crop path information that is independently identified based on its respective environmental information. By comparing the spatial geometric consistency and reliability scores of target crop path information independently identified based on their respective environmental information, the overall credibility of the current environmental information is quantified. When the overall credibility is greater than or equal to the credibility threshold, navigation is performed based on the different types of environmental information; When the overall confidence level is less than the confidence threshold, navigation is performed based on internal dead reckoning. The evaluation of the reliability scores of target crop path information independently identified based on their respective environmental information includes: When the environmental information is image data, the image data is binarized. Based on the binarized image data, edges are extracted from the image data. Multiple line segments are identified and fitted from the edges; these line segments are used to indicate possible boundaries of the crop. The image reliability score is calculated based on the number of line segments, parallelism, length, and edge sharpness of the image. The step of calculating the image reliability score based on the number of line segments, parallelism, length, and edge sharpness of the image includes: Calculate the image reliability score according to the image reliability score calculation formula; The formula for calculating the image reliability score is: ； Where F1 represents the image reliability score, w1, w2, and w3 represent weighting coefficients; S represents the number of valid line segments; SMAX represents the expected maximum number of line segments; L represents the average line segment parallelism; Q represents the average edge sharpness; the average line segment parallelism is determined by calculating the standard deviation of the angle between all fitted line segments; the average edge sharpness is determined by calculating the average gradient value of the crop row edge region.

2. The navigation method for a green-cutting robot according to claim 1, characterized in that, The reliability scores of target crop pathway information independently identified based on their respective environmental information were evaluated separately, including: When the environmental information is radar point cloud data, the radar point cloud data is preprocessed; the preprocessing includes: noise point filtering, and removing points below a specific height threshold based on the installation height of the lidar and the ground slope information to exclude ground points; Identify crop clusters in the radar point cloud data; By fitting straight lines to the center or boundary points of these point clusters, the spatial boundary lines formed by crop stems can be identified. The radar reliability score is calculated based on the completeness of the clustering results, the linearity of the fitted line, and the uniformity of the crop row spacing.

3. The navigation method for a green-cutting robot according to claim 2, characterized in that, The radar reliability score is calculated based on the completeness of the clustering results, the linearity of the fitted line, and the uniformity of the crop row spacing, including: Calculate the radar reliability score according to the radar reliability score calculation formula; The formula for calculating the radar reliability score includes: ； Where F2 represents the radar reliability score, w4, w5, and w6 represent weighting coefficients; J represents the number of effective clusters; JMAX represents the expected maximum number of clusters; N represents the average linearity of the fit; H represents the reciprocal of the standard deviation of the crop row spacing; the average linearity of the fit is determined by the root mean square error of the fitting residuals.

4. The navigation method for a green-cutting robot according to claim 1, characterized in that, The comparison of the spatial geometric consistency of target crop path information independently identified based on their respective environmental information includes: Sampling is performed within the target crop path information independently identified based on their respective environmental information to obtain a series of corresponding point pairs; Calculate the average Euclidean distance deviation between the point pairs; The average Euclidean distance deviation is expressed as: ； Where OL represents the average Euclidean distance deviation, N is the number of sampling points, Xi is the abscissa of the sampling points, sqrt represents the square root function, fv(Xi) represents the function representation of the target crop path information independently identified based on image data, and fl(Xi) represents the function representation of the target crop path information independently identified based on radar point cloud data.

5. The navigation method for a green-cutting robot according to claim 4, characterized in that, The overall credibility of the quantified current environmental information includes: The overall confidence level is determined based on the average Euclidean distance deviation. The overall credibility satisfies the following relationship: ； FZ represents the overall confidence level; OLmax represents the maximum permissible deviation.

6. The navigation method for a green-cutting robot according to claim 1, characterized in that, The navigation based on internal dead reckoning includes: Obtain a virtual trajectory; the virtual trajectory is a smooth extension based on the position and posture of the cutting robot when the overall confidence level is less than the confidence level threshold; The virtual trajectory is followed based on the turning command to navigate according to internal dead reckoning; the turning command is determined based on the deviation between the dead reckoning result and the virtual trajectory.

7. The navigation method for a green-cutting robot according to claim 1, characterized in that, The method further includes: Identify crop pixel regions in the image data; Identify crop clusters in the radar point cloud data; The crop point clusters are projected onto the image plane to obtain the projection points; For each projection point, determine whether it falls within the central confidence region of the corresponding crop pixel area; The projection deviation ratio is obtained by calculating the proportion of projection points that fall outside the central confidence region to the total number of projection points. When the projection deviation ratio exceeds a preset threshold, it is determined that there is a consistency deviation phenomenon, and the overall credibility is corrected based on the consistency deviation phenomenon.

8. A navigation system for a green-cutting robot, characterized in that, The system includes: An environmental information acquisition module is used to acquire different types of environmental information ahead; the different types of environmental information include image data and radar point cloud data. The path identification and evaluation module is used to determine the target crop path information independently identified based on their respective environmental information; and to evaluate the reliability score of the target crop path information independently identified based on their respective environmental information. The perception information mutual verification module is used to compare the spatial geometric consistency of target crop path information independently identified based on their respective environmental information and their respective reliability scores to quantify the overall credibility of the current environmental information. The navigation strategy adjustment module is used to perform navigation based on the different types of environmental information when the overall credibility is greater than or equal to the credibility threshold. When the overall confidence level is less than the confidence threshold, navigation is performed based on internal dead reckoning. The evaluation of the reliability scores of target crop path information independently identified based on their respective environmental information includes: When the environmental information is image data, the image data is binarized. Based on the binarized image data, edges are extracted from the image data. Multiple line segments are identified and fitted from the edges; these line segments are used to indicate possible boundaries of the crop. The image reliability score is calculated based on the number of line segments, parallelism, length, and edge sharpness of the image. The step of calculating the image reliability score based on the number of line segments, parallelism, length, and edge sharpness of the image includes: Calculate the image reliability score according to the image reliability score calculation formula; The formula for calculating the image reliability score is: ； Where F1 represents the image reliability score, w1, w2, and w3 represent weighting coefficients; S represents the number of valid line segments; SMAX represents the expected maximum number of line segments; L represents the average line segment parallelism; Q represents the average edge sharpness; the average line segment parallelism is determined by calculating the standard deviation of the angle between all fitted line segments; the average edge sharpness is determined by calculating the average gradient value of the crop row edge region.

Images