A visual detection unmanned aerial vehicle path tracking guidance design method

Abstract

The application discloses a kind of unmanned aerial vehicle path tracking guidance design methods of visual detection, comprising the following steps: obtaining airborne camera image, visual detection result, flight state information and control constraint information and preprocessing;Topological coding is carried out to visual detection result;Error space reachability is constructed according to flight state information and control constraint information;Double-domain consistency constraint intersection is executed, and double-domain consistent error cone is generated;In the cone stability and topological switching hysteresis mechanism constraint executes cone-in optimization, generates guidance error quantity, executes cone extension rollback when cone degradation marker triggers;Guidance error quantity is generated based on path tracking guidance instruction and is issued to execute, and execution feedback is collected to return to flow update.The application adopts double-domain consistent error cone guidance method, realizes unmanned aerial vehicle path tracking driven by visual detection, and has the advantages of strong anti-interference, error stability and smooth control.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) guidance technology, and more particularly to a visual detection-based UAV path tracking guidance design method. Background Technology

[0002] In existing UAV path tracking and guidance technologies, common solutions typically extract target positions, path lines, or regional features from images acquired by airborne cameras, then combine this with flight status information to generate control errors and output guidance commands. Some solutions introduce filtering, prediction, or thresholding to mitigate the impact of visual noise, short-term occlusion, and detection jitter on control stability; others focus on path planning or trajectory tracking control based on flight status and control constraints to improve executability and flight safety. While these technologies have been able to accomplish basic vision-guided flight tasks in engineering applications, their processing chains primarily rely on direct mapping from detection results to control quantities, failing to adequately utilize the connectivity, breaks, and directional continuity within visual evidence.

[0003] However, existing technologies still have significant shortcomings in complex scenarios: when path detection results are affected by occlusion, obstacle interference, or semantic region missegmentation, visual detection results are prone to jumps, breaks, or topology misjudgments, leading to increased fluctuations in guidance error. Simultaneously, there is often a lack of unified error space constraint expression between visual detection results and the UAV's current flight status and control constraints, resulting in insufficient matching of the generated error in terms of power reachability and control continuity. Furthermore, existing solutions are generally inadequate in their jitter suppression during topology switching, backoff handling under degraded conditions, and execution feedback-driven rule and threshold update mechanisms, making it difficult to balance path tracking accuracy, guidance stability, and anti-interference capability within a continuous guidance cycle.

[0004] Therefore, how to provide a visual detection-based UAV path tracking guidance design method is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a vision-based path tracking guidance design method for unmanned aerial vehicles (UAVs). This invention employs a dual-domain consistent error cone guidance method to achieve vision-driven UAV path tracking, which has the advantages of strong anti-interference, stable error, and smooth control.

[0006] A visual detection-based UAV path tracking guidance design method according to an embodiment of the present invention includes the following steps: Acquire airborne camera images, visual inspection results, flight status information, and control constraint information, and perform preprocessing to form a standardized visual status input set; Based on the standardized visual state input set, topological encoding is performed on the visual detection results to generate a topological representation of path evidence; Based on the flight state information and control constraint information in the standardized visual state input set, error space reachability constraints are constructed, and dynamic reachability constraint representations are generated; The bi-domain consistency constraint is applied to the topological representation of path evidence and the dynamic reachability constraint representation to generate a bi-domain consistency error cone, and the cone stability and cone degradation flag are output. Based on the dual-domain consistent error cone, intra-cone optimization is performed under the constraints of cone stability and topology switching hysteresis mechanism to generate guidance error quantity, and cone extension backtracking is performed when the cone degradation mark is triggered; Based on the guidance error, path tracking guidance commands are generated and issued for execution, and execution feedback is collected for feedback updates.

[0007] Optionally, the generation of the normalized visual state input set specifically includes: Collect airborne camera images, visual inspection results, flight status information and control constraint information within the same guidance cycle, establish timestamps, source identifiers and cycle identifiers, and form a raw visual status data set; The original visual state data set is subjected to unified time base processing and mapped to a unified guidance timing sequence to generate a unified timing data set. Image preprocessing is performed on airborne camera images in a unified time-series dataset, including distortion correction, brightness normalization, and invalid edge cropping, to generate a preprocessed image set; The visual detection results in the unified temporal data set are processed to unify the format, converting different detection output formats into a unified detection description structure. The detection result positions are remapped based on the preprocessed image set to generate a unified detection result set. Perform state preprocessing on the flight state information in the unified time series data set, including abnormal state removal, missing state completion, and state sampling reorganization, to generate a preprocessed flight state set; Perform constraint preprocessing on the control constraint information in the unified time series data set, including unifying constraint types, validating constraint boundaries, and unifying constraint units, to generate a preprocessed control constraint set; Consistency screening and correlation encapsulation are performed based on the preprocessed image set, unified detection result set, preprocessed flight state set, and preprocessed control constraint set to form a standardized visual state input set.

[0008] Optionally, the generation of the path evidence topology representation specifically includes: Read the unified detection result set from the standardized visual state input set, extract the corresponding data of path line detection results, obstacle detection results and semantic region detection results under the same unified guidance time sequence, and form a topologically encoded input set; Based on the topologically encoded input set, the path detection results are processed into segments and endpoints are merged to obtain a set of candidate path segments. Adjacency records are established for each candidate path segment to form a basic set of path connection. Based on the path connection base set, connectivity determination and branch separation processing are performed on adjacent path candidate segments to generate a branch candidate set, and the initial results of branch connectivity are generated according to the connection relationship of each path candidate segment. The obstacle detection results are mapped to the spatial range of the branch candidate set. The obstacle detection results that overlap, obstruct, or block the path line candidate segment are associated and labeled to generate the obstacle-associated branch set. The semantic region detection results are mapped to the spatial range of the branch candidate set. Regional consistency screening is performed on the passable, impassable and interference regions corresponding to the branch candidate set to generate a semantically constrained branch set. Based on the obstacle-associated branch set and the semantically constrained branch set, breakage identification is performed on the branch candidate set to generate breakage information, including breakage location, breakage length and breakage type. The initial results of branch connectivity are then corrected to obtain the branch connectivity. Based on the extension direction changes of candidate path segments in each branch candidate set and the directional continuity of adjacent candidate path segments, directional consistency calculation and screening are performed to generate directional consistency. Based on the connection relationship of branches, the information of breaks, and the consistency of directions, the path evidence topology is encapsulated to form a topological representation.

[0009] Optionally, the generation of the dynamic reachability constraint representation specifically includes: Read the preprocessed flight state set and preprocessed control constraint set from the normalized visual state input set to form the dynamic constraint construction input set; Based on dynamic constraints, an input set is constructed to perform guidance cycle state pairing and constraint pairing, establish the correspondence between each flight state information and each control constraint information under the same unified guidance timing, and generate a state constraint pairing set. Based on the state constraint pairing set, motion consistency screening is performed on position state information, velocity state information, attitude state information, angular velocity state information, and altitude state information. State pairing items that are inconsistent with the current guidance cycle are eliminated, and a valid state constraint pairing set is generated. Based on the lateral maneuver constraint information, heading change constraint information, and turning capability constraint information in the effective state constraint pairing set, construct the lateral maneuver reachable boundary and the heading adjustment reachable boundary, and generate the maneuver reachable boundary set; Based on the velocity change constraint information and guidance command change rate constraint information in the effective state constraint pairing set, construct the velocity adjustment reachability boundary and the command change reachability boundary, and generate the adjustment reachability boundary set; Based on the set of maneuverable reachable boundaries and the set of adjustable reachable boundaries, and combined with the effective state constraint pairing set, error space mapping is performed to generate a set of candidate constraints in the error space; Boundary closure processing, conflict constraint resolution, and temporal continuity screening are performed on the candidate constraint set in the error space to form reachable constraints in the error space; The constraint fields are encapsulated based on the error space reachability constraint to obtain the dynamic reachability constraint representation.

[0010] Optionally, the generation of the dual-domain consistent error cone, cone stability, and cone degradation marker specifically includes: Read the path evidence topology representation and dynamic reachability constraint representation, correspond to the current guidance cycle according to the unified guidance timing, extract the branch connectivity relationship, breakage information, directional consistency, error space reachability constraint and corresponding relationship, and form a dual-domain consistency constraint intersection input set; Based on the dual-domain consistency constraint intersection input set, the branch connectivity, breakage information and directional consistency are mapped to the error space composed of lateral deviation and heading deviation, generating a visual candidate constraint set; Based on the intersection input set of dual-domain consistency constraints, the visual candidate constraint set is paired with the error space reachability constraints and their corresponding relationships to generate a dual-domain paired constraint set. Based on the set of dual-domain pairing constraints, the intersection of dual-domain consistency constraints is performed to filter out visual candidate constraints that conflict with the reachability constraints in the error space, and generate a set of dual-domain consistency candidate cones. Based on the dual-domain consistent candidate cone set, cone parameters are extracted and uniformly encapsulated for each candidate result to generate a dual-domain consistent error cone. Based on the matching results of each candidate result in the dual-domain consistent candidate cone set with the branch connectivity, breakage information, directional consistency and error space reachability constraints, stability screening is performed to generate cone stability and write it into the dual-domain consistent error cone. Degradation determination is performed based on the intersection results of the two-domain consistent candidate cone sets, a cone degradation marker is generated, and written into the two-domain consistent error cone.

[0011] Optionally, the dual-domain consistent error cone includes a cone center, cone axis direction, cone opening, topological connectivity mode label, cone stability, and cone degeneration marker. The center value of each candidate position is determined as the cone center, the cone axis direction is determined based on the composite direction of the directional constraint direction information corresponding to each candidate result in the error space, and the cone opening is determined based on the maximum angle between each candidate position and the cone center direction and the cone axis direction.

[0012] Optionally, the degradation determination includes empty intersections, excessively wide intersections, and cone stability below the stability threshold.

[0013] Optionally, the generation of the guidance error specifically includes: Read the dual-domain consistency error cone of the current guidance cycle, extract the cone center, cone axis direction, cone opening, topological connectivity mode label, cone stability, and cone degradation marker to form the cone-in-cone optimal input set; A topology switching hysteresis mechanism is established based on the in-cone optimal input set to determine the input and generate the topology switching determination result. Based on the topology switching determination result, execute the topology switching hysteresis mechanism constraint and generate the topology connectivity mode label after hysteresis constraint; Based on the topological connectivity mode label, cone center, cone axis direction, cone opening and cone stability after hysteresis constraint, intra-cone selection is performed to form a set of intra-cone candidate guidance results. The results are then filtered according to the continuity constraint and guidance command change rate constraint information of the current guidance cycle to generate the guidance error amount. When the cone degradation flag is triggered, a backtracking trigger determination is performed. The dual-domain consistency error cone retained in the previous guidance cycle and the dynamic reachability constraint representation of the current guidance cycle are read. Based on the cone center, cone axis direction and cone opening retained in the previous guidance cycle, cone extension backtracking is performed to generate a backtracked dual-domain consistency error cone. Based on the consistent error cone of the backtracking dual domain, the backtracking cone is selected in conservative mode. The backtracking guidance error is generated by the guidance command change rate constraint information and the velocity change constraint information together, and replaces the guidance error of the current guidance cycle.

[0014] Optionally, the generation of the reflow update specifically includes: The guidance error value of the current guidance cycle is read, and combined with the preprocessed flight state set and the preprocessed control constraint set, a guidance command generation input set is formed. Based on the input set of guidance command generation, path tracking guidance command generation is performed. Guidance error is processed by determining guidance direction, limiting guidance amplitude and guidance change rate, and then path tracking guidance command is generated. The path tracking guidance command is issued and executed, and the execution feedback set corresponding to the current guidance cycle is collected. Based on the execution feedback set and the topological connectivity mode label after the current guidance cycle's dual-domain consistent error cone, cone stability, cone degradation flag, and hysteresis constraint, a backflow association is performed to generate a backflow update input set. Based on the reflux update input set, the cone stability evaluation rule and topology switching hysteresis mechanism parameters are updated to generate the updated cone stability evaluation rule and updated topology switching hysteresis mechanism parameters; The degradation decision parameters are updated based on the reflux update input set, generating updated degradation decision parameters, including the values ​​of the opening threshold and stability threshold used in subsequent guidance cycles; The updated cone stability assessment rules, updated topology switching hysteresis mechanism parameters, and updated degradation judgment parameters are written into the feedback update results of subsequent guidance cycles to complete the feedback update after the path tracking guidance command is issued and executed.

[0015] The beneficial effects of this invention are: This invention proposes a visual detection-based UAV path tracking guidance design method. By uniformly preprocessing and associating visual detection results, flight state information, and control constraint information, a standardized visual state input set is constructed. This is further used to form a path evidence topology representation and a dynamic reachability constraint representation. Finally, a dual-domain consistent error cone is generated through the intersection of dual-domain consistency constraints, thus achieving collaborative constraints between visual evidence and dynamic executability conditions. Compared to existing technologies that directly generate guidance error quantities from visual detection results, this invention can structurally constrain the generation process of guidance error quantities in complex scenarios such as path line breaks, obstacle interference, and semantic region perturbations. This utilizes branch connectivity, breakage information, and directional consistency to provide better continuity, stability, and executability of the generated guidance error quantities.

[0016] Meanwhile, this invention introduces cone stability, cone degradation markers, topology switching hysteresis mechanism constraints, and cone extension backoff processing on the basis of the dual-domain consistent error cone. Combined with execution feedback to form a backflow update, it updates the cone stability evaluation rules, topology switching hysteresis mechanism parameters, and degradation judgment parameters in subsequent guidance cycles. This ensures stable output during the guidance process even under conditions of visual noise, short-term loss, frequent topology switching, and intersection degradation. Therefore, this invention effectively reduces guidance error jitter and control fluctuations caused by unreasonable switching, improves the smoothness and robustness of path tracking guidance commands, and enhances the UAV's continuous path tracking capability and anti-interference capability in complex visual environments. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a visual detection-based UAV path tracking guidance design method proposed in this invention; Figure 2 This is a flowchart illustrating the generation of a dual-domain consistent error cone through the intersection of dual-domain consistency constraints in a visual detection-based UAV path tracking guidance design method proposed in this invention. Figure 3 This is a flowchart of the in-cone selection and cone extension back-off of a visual detection-based UAV path tracking guidance design method proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-3 A visual detection-based UAV path tracking guidance design method includes the following steps: Acquire airborne camera images, visual inspection results, flight status information, and control constraint information, and perform preprocessing to form a standardized visual status input set; Based on the standardized visual state input set, topological encoding is performed on the visual detection results to generate a topological representation of path evidence; Based on the flight state information and control constraint information in the standardized visual state input set, error space reachability constraints are constructed, and dynamic reachability constraint representations are generated; The bi-domain consistency constraint is applied to the topological representation of path evidence and the dynamic reachability constraint representation to generate a bi-domain consistency error cone, and the cone stability and cone degradation flag are output. Based on the dual-domain consistent error cone, intra-cone optimization is performed under the constraints of cone stability and topology switching hysteresis mechanism to generate guidance error quantity, and cone extension backtracking is performed when the cone degradation mark is triggered; Based on the guidance error, path tracking guidance commands are generated and issued for execution, and execution feedback is collected for feedback updates.

[0020] In this embodiment, the generation of the standardized visual state input set specifically includes: Collect airborne camera images, visual inspection results, flight status information and control constraint information within the same guidance cycle, establish timestamps, source identifiers and cycle identifiers, and form a raw visual status data set; The visual detection results include path detection results, obstacle detection results, and semantic region detection results. The flight status information includes position status information, velocity status information, attitude status information, angular velocity status information, and altitude status information. The control constraint information includes lateral maneuver constraint information, heading change constraint information, turning capability constraint information, velocity change constraint information, and guidance command change rate constraint information. The original visual state data set is subjected to unified time base processing and mapped to a unified guidance timing sequence to generate a unified timing data set. Image preprocessing is performed on airborne camera images in a unified time-series dataset, including distortion correction, brightness normalization, and invalid edge cropping, to generate a preprocessed image set; The visual detection results in the unified temporal data set are processed to unify the format, converting different detection output formats into a unified detection description structure. The detection result positions are remapped based on the preprocessed image set to generate a unified detection result set. Perform state preprocessing on the flight state information in the unified time series data set, including abnormal state removal, missing state completion, and state sampling reorganization, to generate a preprocessed flight state set; Perform constraint preprocessing on the control constraint information in the unified time series data set, including unifying constraint types, validating constraint boundaries, and unifying constraint units, to generate a preprocessed control constraint set; Based on the preprocessed image set, unified detection result set, preprocessed flight state set, and preprocessed control constraint set, consistency screening and correlation encapsulation are performed to form a standardized visual state input set; Consistency screening includes removing data with temporal mismatches, spatial mismatches, and threshold out-of-bounds errors.

[0021] In this embodiment, the generation of the path evidence topology representation specifically includes: Read the unified detection result set from the standardized visual state input set, extract the corresponding data of path line detection results, obstacle detection results and semantic region detection results under the same unified guidance time sequence, and form a topologically encoded input set; Based on the topologically encoded input set, the path detection results are processed into segments and endpoints are merged to obtain a set of candidate path segments. Adjacency records are established for each candidate path segment to form a basic set of path connection. Based on the path connection base set, connectivity determination and branch separation processing are performed on adjacent path candidate segments to generate a branch candidate set, and the initial results of branch connectivity are generated according to the connection relationship of each path candidate segment. The obstacle detection results are mapped to the spatial range of the branch candidate set. The obstacle detection results that overlap, obstruct, or block the path line candidate segment are associated and labeled to generate the obstacle-associated branch set. The semantic region detection results are mapped to the spatial range of the branch candidate set. Regional consistency screening is performed on the passable, impassable and interference regions corresponding to the branch candidate set to generate a semantically constrained branch set. Based on the obstacle-associated branch set and the semantically constrained branch set, breakage identification is performed on the branch candidate set to generate breakage information, including breakage location, breakage length and breakage type. The initial results of branch connectivity are then corrected to obtain the branch connectivity. Based on the extension direction changes of candidate path segments in each branch candidate set and the directional continuity of adjacent candidate path segments, directional consistency calculation and screening are performed to generate directional consistency. Based on the connection relationship of branches, the information of breaks, and the consistency of directions, the path evidence topology is encapsulated to form a topological representation.

[0022] In this embodiment, the generation of the dynamic reachability constraint representation specifically includes: Read the preprocessed flight state set and preprocessed control constraint set from the normalized visual state input set to form the dynamic constraint construction input set; The input set for constructing dynamic constraints includes position state information, velocity state information, attitude state information, angular velocity state information, altitude state information, lateral maneuvering constraint information, heading change constraint information, turning capability constraint information, velocity change constraint information, and guidance command change rate constraint information. Based on dynamic constraints, an input set is constructed to perform guidance cycle state pairing and constraint pairing, establish the correspondence between each flight state information and each control constraint information under the same unified guidance timing, and generate a state constraint pairing set. Based on the state constraint pairing set, motion consistency screening is performed on position state information, velocity state information, attitude state information, angular velocity state information, and altitude state information. State pairing items that are inconsistent with the current guidance cycle are eliminated, and a valid state constraint pairing set is generated. Based on the lateral maneuver constraint information, heading change constraint information, and turning capability constraint information in the effective state constraint pairing set, construct the lateral maneuver reachable boundary and the heading adjustment reachable boundary, and generate the maneuver reachable boundary set; Based on the velocity change constraint information and guidance command change rate constraint information in the effective state constraint pairing set, construct the velocity adjustment reachability boundary and the command change reachability boundary, and generate the adjustment reachability boundary set; Based on the set of maneuverable reachable boundaries and the set of adjustable reachable boundaries, and combined with the effective state constraint pairing set, error space mapping is performed to generate a set of candidate constraints in the error space; Among them, the error space mapping is performed by mapping the set of maneuverable reachable boundaries and the set of adjustable reachable boundaries to the error space composed of lateral deviation and heading deviation. The working condition is corrected by combining the speed state information, attitude state information, angular velocity state information and altitude state information in the effective state constraint pairing set, generating error direction constraints, error magnitude constraints and error change constraints, and associating and encapsulating them into a set of candidate constraints for the error space. Boundary closure processing, conflict constraint resolution, and temporal continuity screening are performed on the candidate constraint set in the error space to form reachable constraints in the error space; The constraint fields are encapsulated based on the error space reachability constraint to obtain the dynamic reachability constraint representation; The constraint field encapsulation maps the error space reachability constraints to the current guidance cycle according to a unified guidance timing sequence. It also associates and organizes the set of maneuverable reachable boundaries, the set of adjustable reachable boundaries, and the error space mapping results, maintaining the correspondence with the effective state constraint pairing set. These are written into the same encapsulation structure in a unified order to form a dynamic reachability constraint representation that includes the error space reachability constraints and their corresponding relationships.

[0023] In this embodiment, the generation of the dual-domain consistent error cone, cone stability, and cone degeneration marker specifically includes: Read the path evidence topology representation and dynamic reachability constraint representation, correspond to the current guidance cycle according to the unified guidance timing, extract the branch connectivity relationship, breakage information, directional consistency, error space reachability constraint and corresponding relationship, and form a dual-domain consistency constraint intersection input set; Based on the dual-domain consistency constraint intersection input set, the branch connectivity, breakage information and directional consistency are mapped to the error space composed of lateral deviation and heading deviation, generating a visual candidate constraint set; The visual candidate constraint set includes connectivity constraints corresponding to branch connectivity, fracture constraints corresponding to fracture information, and directional constraints corresponding to directional consistency. Based on the intersection input set of dual-domain consistency constraints, the visual candidate constraint set is paired with the error space reachability constraints and their corresponding relationships to generate a dual-domain paired constraint set. Based on the set of dual-domain pairing constraints, the intersection of dual-domain consistency constraints is performed to filter out visual candidate constraints that conflict with the reachability constraints in the error space, and generate a set of dual-domain consistency candidate cones. Based on the dual-domain consistent candidate cone set, cone parameters are extracted and uniformly encapsulated for each candidate result to generate a dual-domain consistent error cone. Based on the matching results of each candidate result in the dual-domain consistent candidate cone set with the branch connectivity, breakage information, directional consistency and error space reachability constraints, stability screening is performed to generate cone stability and write it into the dual-domain consistent error cone. The stability screening takes each candidate result in the dual-domain consistent candidate cone set as the object, performs branch connectivity consistency screening, retains candidate results that match the branch connectivity of the current guidance cycle, performs fracture information consistency screening, filters out candidate results corresponding to fracture location, fracture length and fracture type, performs direction consistency screening, retains candidate results that match the direction consistency, and performs matching screening with error space reachability constraints. Based on the matching completeness and matching continuity of each candidate result in branch connectivity, fracture information, direction consistency and error space reachability constraints, the cone stability is generated. Degradation determination is performed based on the intersection results of the two-domain consistent candidate cone sets, a cone degradation marker is generated, and written into the two-domain consistent error cone.

[0024] In this embodiment, the dual-domain consistent error cone includes a cone center, a cone axis direction, a cone opening, a topological connectivity mode label, a cone stability, and a cone degeneration marker. The center value of each candidate position is determined as the cone center. The cone axis direction is determined based on the composite direction of the directional constraint direction information corresponding to each candidate result in the error space. The cone opening is determined based on the maximum angle between each candidate position and the cone center direction and the cone axis direction.

[0025] In this embodiment, the degradation determination includes empty intersection, excessively wide intersection, and cone stability below the stability threshold. Where intersection is empty, it means that the set of candidate cones for bi-domain consistency is empty after the bi-domain consistency constraint intersection is empty; intersection is too wide, it means that the cone opening of the bi-domain consistency error cone is greater than the opening threshold; cone stability is lower than the stability threshold, it means that the cone stability is less than the stability threshold under the current guidance cycle.

[0026] In this embodiment, the generation of guidance error specifically includes: Read the dual-domain consistency error cone of the current guidance cycle, extract the cone center, cone axis direction, cone opening, topological connectivity mode label, cone stability, and cone degradation marker to form the cone-in-cone optimal input set; A topology switching hysteresis mechanism is established based on the in-cone optimal input set to determine the input and generate the topology switching determination result. The inputs for the topology switching hysteresis mechanism include the topology connectivity mode label and cone stability of the current guidance cycle, as well as the topology connectivity mode label and cone stability retained from the previous guidance cycle. The topology switching determination result is generated based on the difference between the topology connectivity mode label of the current guidance cycle and the topology connectivity mode label retained in the previous guidance cycle, the comparison between the cone stability and the stability threshold of the current guidance cycle, and the determination of the number of retention cycles of the topology connectivity mode label of the current guidance cycle. Based on the topology switching determination result, execute the topology switching hysteresis mechanism constraint and generate the topology connectivity mode label after hysteresis constraint; The topology switching hysteresis mechanism constraint performs a hold process on the topology connectivity mode label of the current guidance cycle that does not meet the topology switching determination, and performs a switching process on the topology connectivity mode label of the current guidance cycle that meets the topology switching determination. Based on the topological connectivity mode label, cone center, cone axis direction, cone opening and cone stability after hysteresis constraint, intra-cone selection is performed to form a set of intra-cone candidate guidance results. The results are then filtered according to the continuity constraint and guidance command change rate constraint information of the current guidance cycle to generate the guidance error amount. When the cone degradation flag is triggered, a backtracking trigger determination is performed. The dual-domain consistency error cone retained in the previous guidance cycle and the dynamic reachability constraint representation of the current guidance cycle are read. Based on the cone center, cone axis direction and cone opening retained in the previous guidance cycle, cone extension backtracking is performed to generate a backtracked dual-domain consistency error cone. Based on the consistent error cone of the backtracking dual domain, the backtracking cone is selected in conservative mode. The backtracking guidance error is generated by the guidance command change rate constraint information and the velocity change constraint information together, and replaces the guidance error of the current guidance cycle.

[0027] In this embodiment, the generation of the reflow update specifically includes: The guidance error value of the current guidance cycle is read, and combined with the preprocessed flight state set and the preprocessed control constraint set, a guidance command generation input set is formed. The input set for generating guidance commands includes guidance error, position status information, velocity status information, attitude status information, angular velocity status information, altitude status information, velocity change constraint information, and guidance command change rate constraint information. Based on the input set of guidance command generation, path tracking guidance command generation is performed. Guidance error is processed by determining guidance direction, limiting guidance amplitude and guidance change rate, and then path tracking guidance command is generated. The guidance direction determination and guidance amplitude limit are jointly constrained by the guidance error and the preprocessing control constraint set, and the guidance change rate limit is constrained by the guidance command change rate constraint information. The path tracking guidance command is issued and executed, and the execution feedback set corresponding to the current guidance cycle is collected. The execution feedback set includes the execution results of path tracking guidance commands, flight status change results, and path tracking results corresponding to guidance error amounts; Based on the execution feedback set and the topological connectivity mode label after the current guidance cycle's dual-domain consistent error cone, cone stability, cone degradation flag, and hysteresis constraint, a backflow association is performed to generate a backflow update input set. Based on the reflux update input set, the cone stability evaluation rule and topology switching hysteresis mechanism parameters are updated to generate the updated cone stability evaluation rule and updated topology switching hysteresis mechanism parameters; The cone stability evaluation rule update is based on the consistency between the path tracking results in the execution feedback set and the guidance error amount in the current guidance cycle, the cone degradation marker triggering status, and the cone stability execution in the current guidance cycle. When the consistency meets the preset consistency conditions and the cone degradation marker is not triggered, the maintenance process is executed; otherwise, the screening requirements for matching integrity and matching continuity in cone stability generation are improved. The topology switching hysteresis mechanism parameters include a stability threshold and the number of hold periods used to generate the topology switching determination result. The topology switching hysteresis mechanism parameters are updated based on the hold or switch processing corresponding to the topology switching determination result, the path tracking results in the execution feedback set, and the cone degradation flag triggering status. When the consistency of the path tracking result decreases or the cone degradation flag is triggered after the switch processing, the stability threshold and the number of hold periods are updated upwards. When the hold processing is continuous and the consistency of the path tracking result remains stable, the stability threshold or the number of hold periods is updated downwards. The degradation decision parameters are updated based on the reflux update input set, generating updated degradation decision parameters, including the values ​​of the opening threshold and stability threshold used in subsequent guidance cycles; The opening threshold update is performed based on the triggering of the excessively wide intersection in the cone degradation marker, the cone opening of the two-domain consistency error cone, and the path tracking results in the execution feedback set. When the number of excessively wide intersection triggers increases and the consistency of the path tracking results does not decrease, the opening threshold is updated upward. When excessively wide intersection is not triggered, the consistency of the path tracking results decreases, and the cone opening is close to the opening threshold, the opening threshold is updated downward. The stability threshold is updated in subsequent guidance cycles based on the triggering of cone stability falling below the stability threshold in the cone degradation marker, the cone stability in the current guidance cycle, and the path tracking results in the execution feedback set. When the number of triggering events where the cone stability falls below the stability threshold increases and the path tracking results corresponding to the backtracking guidance error remain stable, the stability threshold is downgraded in subsequent guidance cycles. When the cone stability falls below the stability threshold but is not triggered, the consistency of the path tracking results decreases, and the cone stability is close to the stability threshold, the stability threshold is upgraded in subsequent guidance cycles. The updated cone stability assessment rules, updated topology switching hysteresis mechanism parameters, and updated degradation judgment parameters are written into the feedback update results of subsequent guidance cycles to complete the feedback update after the path tracking guidance command is issued and executed.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to a low-altitude patrol flight path tracking scenario in a riverside industrial park. This scenario includes road edges, green belt boundaries, and temporary fencing. During flight, shadow occlusion, glare interference, localized line breaks, and semantic region confusion are common issues. Common practices often directly translate short-term visual jumps into guidance errors, leading to command jitter, course corrections, and unstable path alignment. The test was conducted at a closed flight site in East China, covering morning and evening hours over several consecutive weekdays. Images from the airborne camera, visual detection results, flight status information, control constraint information, path tracking guidance commands, and execution feedback logs were collected under clear weather, sidelight, and low-light conditions for before-and-after comparative analysis.

[0029] In application, the acquired images and state data are first subjected to unified temporal processing and preprocessing to form a standardized visual state input set. Then, the path line detection results, obstacle detection results, and semantic region detection results are converted into a path evidence topology representation, enabling connectivity, breakage information, and directional consistency to participate in subsequent guidance. Subsequently, a dynamic reachability constraint representation is generated by combining flight state information and control constraint information, and then intersected with the path evidence topology representation through a two-domain consistency constraint to obtain a two-domain consistency error cone. Cone stability and cone degradation markers are output simultaneously. During flight, when the visual evidence is continuous and stable, the guidance error obtained by the optimal selection within the cone remains smooth. When occlusion, breakage, or intersection degradation occurs, the topology switching hysteresis mechanism suppresses frequent switching, cone extension backtracking maintains continuous guidance, and feedback reflow updates the cone stability evaluation rules, topology switching hysteresis mechanism parameters, and degradation judgment parameters. Comparison of flight logs, image logs, and guidance logs shows that, under the same flight path and time period, the present invention reduces the fluctuation trend of guidance error, decreases the number of sudden changes in guidance commands, improves the continuity of path tracking, and makes the recovery process after short-term visual loss more stable. It can effectively solve the problems of unstable error conversion and susceptibility to noise in complex visual environments.

[0030] Table 1. Comparison of Path Tracking Guidance Performance in Complex Visual Interference Scenarios

[0031] As shown in Table 1, the improvement of the method in complex visual interference scenarios does not rely on a single indicator, but rather on the simultaneous improvement of path tracking accuracy, control smoothness, topological stability, and the ability to withstand degradation conditions. The average lateral deviation of path tracking decreased from 0.84m to 0.51m, and the standard deviation of lateral deviation decreased from 0.47m to 0.26m, indicating that not only did the average error decrease, but the error fluctuation also converged significantly. This is directly related to the intersection of path evidence topological representation, dynamic reachability constraint representation, and dual-domain consistency constraint in the claims, because the visual detection results are no longer directly mapped to the guidance error, but are first screened by path connectivity, breakage, directional consistency, and dynamic reachability, reducing the deviation amplification caused by local false detections and short-term breakages.

[0032] Regarding control smoothness, the average peak change rate of heading correction commands decreased from 21.3 deg / s² to 13.8 deg / s², the number of guidance command abrupt changes decreased from 17 to 8, and the number of topology mis-switches decreased from 11 to 4. This change is consistent with the topology switching hysteresis mechanism constraint and cone-based optimization logic in the claims. Conventional methods tend to frequently switch path determinations between adjacent guidance cycles when path detection results are affected by occlusion, reflection, or semantic region confounding, leading to back-and-forth corrections of control commands. This invention introduces topology connectivity mode labels, cone stability, and information retained from the previous guidance cycle, giving the switching and holding processes clear constraints, thereby significantly suppressing the propagation of topology jitter to the control layer.

[0033] In terms of anti-interference and continuity, the recovery time after short-term visual loss was shortened from 2.9s to 1.6s, the proportion of sustainable tracking time under degraded conditions increased from 71.4% to 86.7%, and the route completion rate increased from 92.8% to 97.1%. These results reflect that the cone extension backoff triggered by the cone degradation marker did indeed play a role. When degradation occurs at the intersection of dual-domain consistency constraints, this invention does not directly output the unstable error amount. Instead, it utilizes the cone center, cone axis direction, and cone opening retained from the previous guidance cycle, combined with the current dynamic reachability constraint, to perform in-cone selection during the backoff. This maintains continuous and executable guidance error output during periods when visual evidence is temporarily unreliable, avoiding mission interruption or significant yaw.

[0034] Furthermore, the number of parameter updates triggered by feedback in this invention is higher than that of conventional methods, and the dispersion of repeated flight results is lower. This indicates that the feedback update is not merely a formal record, but rather an effective correction of the cone stability assessment rules, topology switching hysteresis mechanism parameters, and degradation judgment parameters in subsequent guidance cycles. The performance improvement is due to the fact that the parameter update rules are not isolated tuning, but are jointly bound to the consistency of path tracking results, cone degradation marker triggering, and topology switching results. This allows the threshold and hold period number to converge and adjust according to changes in the scenario, thereby improving the consistency and robustness of multiple repeated flights.

[0035] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A visual detection-based UAV path tracking guidance design method, characterized in that, Includes the following steps: Acquire airborne camera images, visual inspection results, flight status information, and control constraint information, and perform preprocessing to form a standardized visual status input set; Based on the standardized visual state input set, topological encoding is performed on the visual detection results to generate a topological representation of path evidence; Based on the flight state information and control constraint information in the standardized visual state input set, error space reachability constraints are constructed, and dynamic reachability constraint representations are generated; The bi-domain consistency constraint is applied to the topological representation of path evidence and the dynamic reachability constraint representation to generate a bi-domain consistency error cone, and the cone stability and cone degradation flag are output. Based on the dual-domain consistent error cone, intra-cone optimization is performed under the constraints of cone stability and topology switching hysteresis mechanism to generate guidance error quantity, and cone extension backtracking is performed when the cone degradation mark is triggered; Based on the guidance error, path tracking guidance commands are generated and issued for execution, and execution feedback is collected for feedback updates.

2. The visual detection-based UAV path tracking guidance design method according to claim 1, characterized in that, The generation of the standardized visual state input set specifically includes: Collect airborne camera images, visual inspection results, flight status information and control constraint information within the same guidance cycle, establish timestamps, source identifiers and cycle identifiers, and form a raw visual status data set; The original visual state data set is subjected to unified time base processing and mapped to a unified guidance timing sequence to generate a unified timing data set. Image preprocessing is performed on airborne camera images in a unified time-series dataset, including distortion correction, brightness normalization, and invalid edge cropping, to generate a preprocessed image set; The visual detection results in the unified temporal data set are processed to unify the format, converting different detection output formats into a unified detection description structure. The detection result positions are remapped based on the preprocessed image set to generate a unified detection result set. Perform state preprocessing on the flight state information in the unified time series data set, including abnormal state removal, missing state completion, and state sampling reorganization, to generate a preprocessed flight state set; Perform constraint preprocessing on the control constraint information in the unified time series data set, including unifying constraint types, validating constraint boundaries, and unifying constraint units, to generate a preprocessed control constraint set; Consistency screening and correlation encapsulation are performed based on the preprocessed image set, unified detection result set, preprocessed flight state set, and preprocessed control constraint set to form a standardized visual state input set.

3. The visual detection-based UAV path tracking guidance design method according to claim 1, characterized in that, The generation of the path evidence topology representation specifically includes: Read the unified detection result set from the standardized visual state input set, extract the corresponding data of path line detection results, obstacle detection results and semantic region detection results under the same unified guidance time sequence, and form a topologically encoded input set; Based on the topologically encoded input set, the path detection results are processed into segments and endpoints are merged to obtain a set of candidate path segments. Adjacency records are established for each candidate path segment to form a basic set of path connection. Based on the path connection base set, connectivity determination and branch separation processing are performed on adjacent path candidate segments to generate a branch candidate set, and the initial results of branch connectivity are generated according to the connection relationship of each path candidate segment. The obstacle detection results are mapped to the spatial range of the branch candidate set. The obstacle detection results that overlap, obstruct, or block the path line candidate segment are associated and labeled to generate the obstacle-associated branch set. The semantic region detection results are mapped to the spatial range of the branch candidate set. Regional consistency screening is performed on the passable, impassable and interference regions corresponding to the branch candidate set to generate a semantically constrained branch set. Based on the obstacle-associated branch set and the semantically constrained branch set, breakage identification is performed on the branch candidate set to generate breakage information, including breakage location, breakage length and breakage type. The initial results of branch connectivity are then corrected to obtain the branch connectivity. Based on the extension direction changes of candidate path segments in each branch candidate set and the directional continuity of adjacent candidate path segments, directional consistency calculation and screening are performed to generate directional consistency. Based on the connection relationship of branches, the information of breaks, and the consistency of directions, the path evidence topology is encapsulated to form a topological representation.

4. The visual detection-based UAV path tracking guidance design method according to claim 1, characterized in that, The generation of the dynamic reachability constraint representation specifically includes: Read the preprocessed flight state set and preprocessed control constraint set from the normalized visual state input set to form the dynamic constraint construction input set; Based on dynamic constraints, an input set is constructed to perform guidance cycle state pairing and constraint pairing, establish the correspondence between each flight state information and each control constraint information under the same unified guidance timing, and generate a state constraint pairing set. Based on the state constraint pairing set, motion consistency screening is performed on position state information, velocity state information, attitude state information, angular velocity state information, and altitude state information. State pairing items that are inconsistent with the current guidance cycle are eliminated, and a valid state constraint pairing set is generated. Based on the lateral maneuver constraint information, heading change constraint information, and turning capability constraint information in the effective state constraint pairing set, construct the lateral maneuver reachable boundary and the heading adjustment reachable boundary, and generate the maneuver reachable boundary set; Based on the velocity change constraint information and guidance command change rate constraint information in the effective state constraint pairing set, construct the velocity adjustment reachability boundary and the command change reachability boundary, and generate the adjustment reachability boundary set; Based on the set of maneuverable reachable boundaries and the set of adjustable reachable boundaries, and combined with the effective state constraint pairing set, error space mapping is performed to generate a set of candidate constraints in the error space; Boundary closure processing, conflict constraint resolution, and temporal continuity screening are performed on the candidate constraint set in the error space to form reachable constraints in the error space; The constraint fields are encapsulated based on the error space reachability constraint to obtain the dynamic reachability constraint representation.

5. The visual detection-based UAV path tracking guidance design method according to claim 1, characterized in that, The generation of the dual-domain consistent error cone, cone stability, and cone degeneration marker specifically includes: Read the path evidence topology representation and dynamic reachability constraint representation, correspond to the current guidance cycle according to the unified guidance timing, extract the branch connectivity relationship, breakage information, directional consistency, error space reachability constraint and corresponding relationship, and form a dual-domain consistency constraint intersection input set; Based on the dual-domain consistency constraint intersection input set, the branch connectivity, breakage information and directional consistency are mapped to the error space composed of lateral deviation and heading deviation, generating a visual candidate constraint set; Based on the intersection input set of dual-domain consistency constraints, the visual candidate constraint set is paired with the error space reachability constraints and their corresponding relationships to generate a dual-domain paired constraint set. Based on the set of dual-domain pairing constraints, the intersection of dual-domain consistency constraints is performed to filter out visual candidate constraints that conflict with the reachability constraints in the error space, and generate a set of dual-domain consistency candidate cones. Based on the dual-domain consistent candidate cone set, cone parameters are extracted and uniformly encapsulated for each candidate result to generate a dual-domain consistent error cone. Based on the matching results of each candidate result in the dual-domain consistent candidate cone set with the branch connectivity, breakage information, directional consistency and error space reachability constraints, stability screening is performed to generate cone stability and write it into the dual-domain consistent error cone. Degradation determination is performed based on the intersection results of the two-domain consistent candidate cone sets, a cone degradation marker is generated, and written into the two-domain consistent error cone.

6. The visual detection-based UAV path tracking guidance design method according to claim 5, characterized in that, The dual-domain consistent error cone includes a cone center, cone axis direction, cone opening, topological connectivity mode label, cone stability, and cone degeneration marker. The center value of each candidate position is determined as the cone center. The cone axis direction is determined based on the synthesis direction of the directional constraint direction information corresponding to each candidate result in the error space. The cone opening is determined based on the maximum angle between each candidate position and the cone center direction and the cone axis direction.

7. The visual detection-based UAV path tracking guidance design method according to claim 1, characterized in that, The degradation criteria include empty intersections, excessively wide intersections, and cone stability below the stability threshold.

8. The visual detection-based UAV path tracking guidance design method according to claim 1, characterized in that, The generation of the guidance error specifically includes: Read the dual-domain consistency error cone of the current guidance cycle, extract the cone center, cone axis direction, cone opening, topological connectivity mode label, cone stability, and cone degradation marker to form the cone-in-cone optimal input set; A topology switching hysteresis mechanism is established based on the in-cone optimal input set to determine the input and generate the topology switching determination result. Based on the topology switching determination result, execute the topology switching hysteresis mechanism constraint and generate the topology connectivity mode label after hysteresis constraint; Based on the topological connectivity mode label, cone center, cone axis direction, cone opening and cone stability after hysteresis constraint, intra-cone selection is performed to form a set of intra-cone candidate guidance results. The results are then filtered according to the continuity constraint and guidance command change rate constraint information of the current guidance cycle to generate the guidance error amount. When the cone degradation flag is triggered, a backtracking trigger determination is performed. The dual-domain consistency error cone retained in the previous guidance cycle and the dynamic reachability constraint representation of the current guidance cycle are read. Based on the cone center, cone axis direction and cone opening retained in the previous guidance cycle, cone extension backtracking is performed to generate a backtracked dual-domain consistency error cone. Based on the consistent error cone of the backtracking dual domain, the backtracking cone is selected in conservative mode. The backtracking guidance error is generated by the guidance command change rate constraint information and the velocity change constraint information together, and replaces the guidance error of the current guidance cycle.

9. The visual detection-based UAV path tracking guidance design method according to claim 1, characterized in that, The generation of the reflow update specifically includes: The guidance error value of the current guidance cycle is read, and combined with the preprocessed flight state set and the preprocessed control constraint set, a guidance command generation input set is formed. Based on the input set of guidance command generation, path tracking guidance command generation is performed. Guidance error is processed by determining guidance direction, limiting guidance amplitude and guidance change rate, and then path tracking guidance command is generated. The path tracking guidance command is issued and executed, and the execution feedback set corresponding to the current guidance cycle is collected. Based on the execution feedback set and the topological connectivity mode label after the current guidance cycle's dual-domain consistent error cone, cone stability, cone degradation flag, and hysteresis constraint, a backflow association is performed to generate a backflow update input set. Based on the reflux update input set, the cone stability evaluation rule and topology switching hysteresis mechanism parameters are updated to generate the updated cone stability evaluation rule and updated topology switching hysteresis mechanism parameters; The degradation decision parameters are updated based on the reflux update input set, generating updated degradation decision parameters, including the values ​​of the opening threshold and stability threshold used in subsequent guidance cycles; The updated cone stability assessment rules, updated topology switching hysteresis mechanism parameters, and updated degradation judgment parameters are written into the feedback update results of subsequent guidance cycles to complete the feedback update after the path tracking guidance command is issued and executed.

Images