Coverage path planning method based on region shrinkage

Abstract

The invention relates to the technical field of unmanned aerial vehicle path planning, in particular to a coverage path planning method based on region contraction, which comprises an image segmentation module, a boundary feature extraction module, a contraction ring generation module, a path optimization module and a path selection module. According to the method, semantic segmentation and boundary feature extraction are performed on a high-resolution image, step-by-step contraction rings and key nodes are generated, a path is optimized through path smoothing and redundant node elimination, and finally, an optimal coverage path is dynamically selected according to task requirements. The method does not need to pre-decompose a non-convex or mixed region, remarkably improves the efficiency, supports multi-target optimization path selection by combining with a dynamic weight distribution and real-time adjustment mechanism, reduces redundant flight and coverage blind areas, improves the flight stability by adopting a cubic spline interpolation algorithm, is suitable for agricultural spraying, topographic mapping and other scenes, and has a wide application prospect. And the method has relatively high practicability and popularization value.

Description

Technical Field

[0001] The present invention belongs to the technical field of path planning, and particularly relates to a coverage path planning method based on region contraction. Background Art

[0002] The coverage path planning technology is widely used in fields such as UAV inspection, agricultural plant protection, and environmental monitoring. Its core goal is to achieve efficient full coverage of the target area by optimizing the path design. However, traditional coverage path planning algorithms have significant deficiencies in dealing with irregular regions (such as non-convex regions or multi-connected regions). These methods usually rely on manual annotation to identify the region boundary, and fail to make full use of the multi-scale feature extraction technology of high-resolution images, resulting in low boundary recognition efficiency and difficulty in dynamically updating the coverage parameters.

[0003] Chinese Patent Publication No. CN116797665A discloses a UAV adaptive coverage path planning method for aircraft surface inspection, which uses a UAV equipped with a camera to perform visual inspection of aircraft surface damage; the viewpoint generation module generates a pose sample set and a corresponding viewpoint sample set by performing surface adaptive discretization sampling on the existing aircraft point cloud model, and screens the samples according to distance, collision, and coverage area; the search space construction module connects the screened pose samples to construct the UAV search space; the path search module searches for a path in the search space according to the constructed weighted heuristic reward function; the coverage evaluation module calculates the coverage rate using the voxel volume method during path search, and finally generates a UAV damage inspection path that meets the target coverage rate.

[0004] In addition, most of the existing technologies do not combine real-time image data (such as terrain features, obstacle distribution, etc.) collected by the UAV camera to dynamically adjust the path parameters, which easily leads to problems such as redundant flight or coverage blind spots, seriously affecting the coverage efficiency and operation quality.

[0005] On the other hand, for the coverage task of non-convex regions or convex and non-convex mixed regions, traditional methods usually require complex regional pre-decomposition operations, dividing the target region into multiple regular sub-regions and then planning paths separately. This pre-decomposition process not only increases the computational complexity but also may affect the coverage effect due to inaccurate decomposition. At the same time, the existing methods lack the utilization of the characteristic of gradual region contraction when generating the coverage path, and it is difficult to form an efficient coverage path by connecting the key nodes on each level of the contraction ring. Therefore, developing a coverage path planning method that does not require regional pre-decomposition and can adaptively process convex and non-convex regions has become an urgent technical problem to be solved.

[0006] The present invention aims to solve the above problems. By integrating image segmentation technology with path planning algorithms, it dynamically generates regional boundary features using the image data collected in real time by drones, and optimizes the calculation of virtual offset vertices in combination with the generation method of gradually shrinking rings, thereby improving the efficiency and accuracy of coverage path planning. This method not only avoids the need for pre-decomposition of irregular regions, but also can dynamically adjust path parameters according to real-time image information, effectively reducing redundant flights and coverage blind spots, and providing a more efficient and reliable solution for coverage tasks in complex scenarios. Summary of the Invention

[0007] The purpose of the present invention is to provide a coverage path planning method based on regional contraction to solve the problems mentioned in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solutions: A coverage path planning method based on regional contraction, characterized in that it includes an image segmentation module, a boundary feature extraction module, a contraction ring generation module, a path optimization module, and a path selection module; the image segmentation module is used for semantic segmentation of the high-resolution images collected by drones; the boundary feature extraction module is used to extract regional boundary features from the segmented images; the contraction ring generation module is used to generate gradually shrinking rings and calculate the positions of key nodes; the path optimization module is used to smooth the generated path and remove redundant nodes; the path selection module is used to select the optimal coverage path according to task requirements.

[0009] As a further solution of the present invention, the contraction ring generation module includes a virtual offset vertex calculation unit; the virtual offset vertex calculation unit is used to dynamically adjust the offset distance according to the boundary features and generate the key node coordinates of the gradually shrinking rings, and the formula is as follows: ; Wherein, represents the coordinate of the th node on the th contraction ring (the previous contraction ring node), is the offset step, represents the coordinate of the th node on the (k - 1)th contraction ring (the previous contraction ring node), represents the gradient direction of the boundary feature function at the node

[0010] As a further solution of the present invention, the contraction ring generation module further includes a boundary constraint correction unit; the boundary constraint correction unit is used to perform boundary integrity verification on the generated contraction rings and correct the positions of the nodes that exceed the boundary, and the formula is as follows: ; Among them, represents the distance from the node to the boundary , is the maximum allowable deviation value, is the correction coefficient, is the corrected node coordinate.

[0011] As a further solution of the present invention, the path optimization module includes a redundant node elimination unit; the redundant node elimination unit is used to eliminate redundant nodes according to the Euclidean distance and included angle relationship between adjacent nodes, and the formula is as follows: ; When and , eliminate the node , where is the included angle between the nodes , is the reference benchmark node (the starting point for comparison), is the node to be evaluated (the redundant point that may be eliminated), is the subsequent adjacent node ( the next point of ), this formula calculates the turning angle formed by three consecutive path nodes, is the minimum included angle threshold, , is the Euclidean distance between the nodes , is the maximum distance threshold.

[0012] As a further solution of the present invention, the path optimization module further includes a path smoothing unit; the path smoothing unit is used to smooth the optimized path and generate a smooth path using the cubic spline interpolation algorithm, and the formula is as follows: ; Among them, , , , is the interpolation coefficient, which is determined by the node coordinates and the path continuity conditions.

[0013] As a further solution of the present invention, the path selection module includes a dynamic weight allocation unit; the dynamic weight allocation unit is used ; Among them, represents the path length cost, represents the time cost, represents the energy consumption cost, , , , are the weight coefficients.

[0014] As a further solution of the present invention, the path selection module further includes a real-time adjustment unit; the real-time adjustment unit is used to update the path weight in real time according to the UAV sensor data.

[0015] As a further solution of the present invention, the path smoothing unit further adopts an anti-interference cubic spline interpolation algorithm, and the specific calculation formula is: ; Wherein, is the anti-interference coefficient, is the noise suppression term, is the noise point weight, dynamically estimated by the Kalman filter, is the highest order of the main path polynomial, determining the complexity of the basic spline, is the number of detected noise points, is the spline interpolation coefficient, is the path parameterization variable, is the parameter position of the noise point.

[0016] As a further solution of the present invention, the dynamic weight allocation unit further introduces an energy recovery cost term , and the specific calculation formula is: ; Wherein, is calculated through the UAV dynamics model, specifically the kinetic energy recovery amount during descent or turning, satisfying , is the energy conversion efficiency, is the UAV mass, is the instantaneous speed.

[0017] Compared with the prior art, the beneficial effects of the present invention are: The present invention does not need to pre-decompose non-convex or mixed type regions, and directly generates a coverage path through step-by-step shrinking rings, significantly improving the algorithm efficiency. Secondly, by combining image segmentation technology to dynamically extract boundary features, it avoids the dependence on manual annotation and improves the automation degree of path planning. Thirdly, through the dynamic weight allocation and real-time adjustment mechanism, it supports multi-objective optimized path selection in complex environments, reducing redundant flights and coverage blind spots. In addition, the cubic spline interpolation algorithm is used to generate a smooth path, improving the flight stability of the UAV and reducing mechanical wear. This method is applicable to various scenarios, such as agricultural spraying, terrain mapping, etc., and has strong practicality and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 : Schematic flow diagram of the method of the present invention; Figure 2: Schematic diagram of the shrinkage ring generation module of the present invention; Figure 3 : Processing flowchart of the path optimization module of the present invention. Detailed implementation manner

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Refer to Figures 1-3 As shown, the present invention provides a coverage path planning method based on regional shrinkage, which is applicable to path planning tasks of unmanned aerial vehicles in complex scenarios, such as agricultural spraying, topographic surveying and mapping and other fields. First, starting from the overall process, the specific implementation methods of the image segmentation module, boundary feature extraction module, shrinkage ring generation module, path optimization module and path selection module are described, and in-depth analysis is carried out in combination with actual application scenarios.

[0021] In the specific implementation process, the operation process of the whole method is as Figure 1 As shown, it includes five main functional modules: an image segmentation module, a boundary feature extraction module, a shrinkage ring generation module, a path optimization module and a path selection module. First, the unmanned aerial vehicle collects image data of the target area through a mounted high-resolution camera. These image data are transmitted to the image segmentation module, which processes the image using semantic segmentation technology and divides the target area into different semantic categories, such as farmland areas, obstacle areas, etc. In order to improve the segmentation accuracy, the deep learning model U-Net is used as the core algorithm in this embodiment, which has strong feature extraction ability and pixel-level segmentation effect. The segmented image provides basic data support for subsequent boundary feature extraction.

[0022] The boundary feature extraction module is responsible for extracting regional boundary features from the segmented image. The extraction of boundary features is completed by calculating the image gradient and combining edge detection algorithms. Specifically, in this embodiment, the Sobel operator is used to calculate the gradient of the segmented image to obtain the gradient direction and gradient amplitude of each pixel point. Subsequently, through non-maximum suppression and double-threshold detection methods, significant boundary points are further screened out. These boundary points constitute the contour information of the target area and lay the foundation for the subsequent generation of shrinkage rings step by step. The output result of the boundary feature extraction module is a set containing the coordinates of the boundary points, and these coordinates will be passed as input to the shrinkage ring generation module.

[0023] The core task of the shrinkage ring generation module is to generate shrinkage rings at different levels based on boundary features and calculate the positions of key nodes on each level of shrinkage rings. This module includes a virtual offset vertex calculation unit and a boundary constraint correction unit. The function of the virtual offset vertex calculation unit is to dynamically adjust the offset distance to ensure that the shrinkage ring can adapt to the complex boundaries of non-convex regions. The specific calculation formula is as follows: ; Where represents the coordinates of the -th node on the -th level of shrinkage ring, is the offset step size,, represents the coordinates of the i-th node on the (k-1)-th level of shrinkage ring (the node of the previous level of shrinkage ring), represents the gradient direction of the boundary feature function at the node . This formula dynamically adjusts the offset step size through the gradient direction to ensure that the shrinkage ring can closely fit the boundary of the target area. In practical applications, the value of the offset step size needs to be adjusted according to the size and complexity of the target area, usually set between 0.5 and 2 meters. In addition, the boundary constraint correction unit is used to perform boundary integrity verification on the generated shrinkage ring and correct the positions of nodes that exceed the boundary. The correction formula is as follows: ; Where represents the distance from the node to the boundary , is the maximum allowable deviation value, is the correction coefficient, is the corrected node coordinates. Through the above formula, the situation that the shrinkage ring nodes deviate from the actual boundary can be effectively avoided, so as to ensure that the generated shrinkage ring is consistent with the boundary of the target area.

[0024] The main task of the path optimization module is to smooth the generated path and remove redundant nodes to improve the feasibility and efficiency of the path. This module includes a redundant node removal unit and a path smoothing unit. The redundant node removal unit calculates the Euclidean distance and angular relationship between adjacent nodes and removes unnecessary redundant nodes. The specific calculation formula is as follows: ; When and , remove the node , where is the included angle between the nodes , is the reference benchmark node (the starting point for comparison), is the node to be evaluated (a redundant node that may be removed), is the subsequent adjacent node ( the next point), and the formula calculates the turning angle formed by three consecutive path nodes, is the minimum angle threshold, , is the node 's Euclidean distance, is the maximum distance threshold. This formula effectively reduces path redundancy through dual constraints of angle and distance. In practical applications, the angle threshold and the distance threshold need to be adjusted according to the flight performance of the UAV and the mission requirements. For example, in the agricultural spraying scenario, can be set to 30 degrees, can be set to 5 meters. The path smoothing unit then uses the cubic spline interpolation algorithm to smooth the optimized path and generate a smooth path. The interpolation formula is as follows: ; where, , , , are interpolation coefficients, determined by the node coordinates and path continuity conditions. This algorithm ensures that the path has second-order continuity between nodes and improves the flight stability of the UAV.

[0025] Moreover, during the flight of the UAV, path smoothing is vulnerable to sensor noise or environmental interference (such as electromagnetic interference, misdetection of obstacles), resulting in path jitter. The anti-interference cubic spline interpolation algorithm is used to suppress noise and improve path robustness. The path smoothing unit further adopts the anti-interference cubic spline interpolation algorithm, and the specific calculation formula is: ; where, is the anti-interference coefficient, with a value range of 0.1 - 0.5 (determined by experiments, taking a larger value in a high-noise environment), is the highest order of the main path polynomial, which determines the complexity of the basic spline, is the number of detected noise points, is the spline interpolation coefficient, is the path parameterization variable, is the parameter position of the noise point, is the noise suppression term as a Dirac function, indicating suppression is applied at the noise point location, is the noise point weight, dynamically estimated by the Kalman filter, with an initial value set to 1 and adaptively adjusted according to the noise intensity. The Kalman filter dynamic estimation is based on the position deviation of the input noise point (Difference between current and historical coordinates), through the state equation: ; where is the Kalman gain filter gain, determined by the noise covariance matrix, is the observation matrix, thus outputting the updated weight , is the noise point position deviation. If < 0.2, it is determined as an invalid noise point and not suppressed. The interpolated path is sent to the UAV flight control system to correct the flight trajectory in real time. For example, in the farmland spraying scenario, the path jitter of the UAV caused by wind interference is reduced by 60% through this algorithm. Through noise detection, anti-interference interpolation, and dynamic weight estimation of Kalman filtering, the path jitter is significantly reduced, suitable for high-interference environments.

[0026] The task of the path selection module is to select the optimal coverage path according to the task requirements and adjust the path weight in real time to cope with the changes in the dynamic environment. This module includes a dynamic weight allocation unit and a real-time adjustment unit. The dynamic weight allocation unit calculates the total path weight through the following formula: ; where represents the path length cost, represents the time cost, represents the energy consumption cost, , , , are weight coefficients. This formula realizes multi-objective optimized path selection through dynamic weight allocation. In practical applications, the values of the weight coefficients need to be adjusted according to the task priorities. For example, in the emergency rescue scenario, the weight of the time cost should be set to a higher value; while in the energy-constrained scenario, the weight of the energy consumption cost should be set to a higher value. The real-time adjustment unit then updates the path weight in real time according to the UAV sensor data to ensure the robustness of the path planning in the dynamic environment.

[0027] When the UAV is descending or turning, part of the kinetic energy can be recovered as electrical energy through motor back-dragging or potential energy conversion. The energy recovery cost item can be introduced to optimize the path selection to extend the endurance. The dynamic weight allocation unit further introduces the energy recovery cost item , and the specific calculation formula is: ; where Through the calculation of the UAV dynamics model, specifically the amount of kinetic energy recovered during descent or turning, it satisfies , is the energy conversion efficiency, is the mass of the UAV, is the instantaneous velocity (obtained in real-time through the IMU). Specifically, the weight coefficients are dynamically adjusted according to the mission mode. For example, in the emergency mode: = 0.7, = 0.2, = 0.1 (prioritizing time cost); in the energy-saving mode: = 0.6, = 0.3, = 0.1 (prioritizing energy consumption optimization); when the UAV is performing mountain mapping, select the path that includes the descent slope (recoverable energy), Calculate the total weight of the two candidate paths: Path A (steep slope descent): ; Path B (smooth flight): .

[0028] If = 0.6, then the weight of Path A is lower and it is preferentially selected. The actual recovered energy is monitored in real-time through the battery management system (BMS), compared with the model prediction value, and the μ value is dynamically corrected. Combining the energy recovery model with dynamic weight allocation, the path is optimized to extend the endurance, which is applicable to long-term inspection or energy-sensitive scenarios.

[0029] Through the collaborative work of the above five modules, the present invention realizes the complete process from image segmentation to optimal path selection. The specific implementation steps are as follows: First, the UAV collects high-resolution images of the target area and performs semantic segmentation through the image segmentation module; then, the boundary feature extraction module extracts the regional boundary features and transmits them to the shrinking ring generation module; next, the shrinking ring generation module generates gradually shrinking rings and calculates the positions of key nodes; subsequently, the path optimization module smooths the generated path and removes redundant nodes; finally, the path selection module selects the optimal coverage path according to the mission requirements and dynamically updates the path weight through the real-time adjustment unit. The entire process is as Figure 1 shown. The connection between each module is tight, ensuring the efficiency and accuracy of path planning.

[0030] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A coverage path planning method based on region contraction, characterized in that It includes an image segmentation module, a boundary feature extraction module, a shrinking ring generation module, a path optimization module, and a path selection module; the image segmentation module is used to perform semantic segmentation on the high-resolution images collected by the drone; the boundary feature extraction module is used to extract regional boundary features from the segmented images; the shrinking ring generation module is used to generate gradually shrinking rings and calculate the positions of key nodes; the path optimization module is used to smooth the generated path and remove redundant nodes; the path selection module is used to select the optimal coverage path according to the task requirements.

2. The coverage path planning method based on region contraction according to claim 1, wherein: The shrinking ring generation module includes a virtual offset vertex calculation unit; the virtual offset vertex calculation unit is used to dynamically adjust the offset distance according to the boundary features and generate the key node coordinates of the gradually shrinking rings. The formula is as follows: ； Among them, represents the coordinates of the th node on the th-level contraction ring, is the offset step size, represents the gradient direction of the boundary feature function at the node 3. The coverage path planning method based on region contraction according to claim 2, wherein: The shrinking ring generation module further includes a boundary constraint correction unit; the boundary constraint correction unit is used to perform boundary integrity verification on the generated shrinking rings and correct the node positions that exceed the boundary. The formula is as follows: ； Among them, represents the node to the boundary distance, is the maximum allowable deviation value, is the correction coefficient.

4. A coverage path planning method based on region contraction according to claim 1, characterized in that: The path optimization module includes a redundant node removal unit; the redundant node removal unit is used to remove redundant nodes according to the Euclidean distance and included angle relationship between adjacent nodes. The formula is as follows: ； When and remove the node .

5. A coverage path planning method based on region contraction according to claim 4, characterized in that: The path optimization module further includes a path smoothing unit; the path smoothing unit is used to smooth the optimized path and generate a smooth path using the cubic spline interpolation algorithm. The formula is as follows: Among them, , , , are interpolation coefficients, which are determined by node coordinates and path continuity conditions.

6. The coverage path planning method based on region contraction according to claim 1, wherein: The path selection module includes a dynamic weight allocation unit; the dynamic weight allocation unit uses ; Among them, represents the path length cost, represents the time cost, represents the energy consumption cost, , , , are weight coefficients.

7. A coverage path planning method based on regional contraction according to claim 6, characterized in that: The path selection module further includes a real-time adjustment unit; the real-time adjustment unit is used to update the path weights in real time according to the drone sensor data.

8. A coverage path planning method based on area contraction according to claim 5, characterized in that: The path smoothing unit further adopts an anti-interference cubic spline interpolation algorithm. The specific calculation formula is: ; Among them, is the anti-interference coefficient, is the noise suppression term, is the noise point weight, which is dynamically estimated by the Kalman filter.

9. The coverage path planning method based on area contraction according to claim 7, wherein: The dynamic weight allocation unit further introduces an energy recovery cost item , and the specific calculation formula is as follows: ; Among them, calculated by the UAV dynamics model, specifically the kinetic energy recovery amount during descent or turning, satisfies , is the energy conversion efficiency, is the UAV mass, is the instantaneous velocity.

Images