Anti-occlusion unmanned aerial vehicle target tracking method and system

Abstract

The application provides an anti-occlusion unmanned aerial vehicle target tracking method and system, relates to the technical field of unmanned aerial vehicle target tracking, and comprises the following steps: identifying the physical structure boundary of an industrial park based on a visible light image of the industrial park, constructing a topological addressing space according to the physical structure boundary of the industrial park, and generating an adjacency transition table; mapping the unmanned aerial vehicle hot spot coordinates in an infrared continuous image to the topological addressing space to generate a topological state sequence; determining a flight occlusion state according to the adjacency relationship of adjacent addresses in the topological state sequence, generating a frozen node set based on the end address of the topological state sequence in response to a hot spot loss event triggered by the flight occlusion state, and mapping the unmanned aerial vehicle hot spot coordinates to a reappearing area in response to a hot spot recovery event; checking the transition path of the frozen node set and the reappearing area in the adjacency transition table, and outputting a tracking recovery result. The application improves the anti-occlusion target tracking accuracy through topological path checking.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) target tracking technology, and in particular to an anti-obstruction UAV target tracking method and system. Background Technology

[0002] Existing UAV target tracking methods typically use camera equipment to capture continuous video images, extract the target's pixel coordinates through target detection algorithms, and combine kinematic models to predict the target's position in the next frame. Then, they use spatial distance calculations to perform data association to maintain the target's identity.

[0003] In the process of drone target monitoring in industrial parks, it is usually necessary to combine the visible light images reflecting the fixed structures of the park with the tracking of the hotspot positions formed by the drone in continuous infrared images. Facilities such as pipe corridors and cooling towers within industrial parks are not merely background elements; their structural boundaries define the visible area and passable relationships of the drone in a localized space. When a drone flies along the gaps in pipe corridors or the outer areas of towers, the positional changes of the infrared hotspots in the images need to correspond to these fixed structures to provide spatial basis for determining the continuation of the target after obstruction.

[0004] When tracking is based solely on infrared image coordinates, the tracking process typically relies on the continuous visibility of the hotspot in adjacent image frames. However, due to obstructions from pipe corridors or curved tower surfaces, the hotspot's continuity may be interrupted, causing a loss of direct image coordinate continuity between its pre-obstruction and post-obstruction positions. In this case, although the recovered hotspot reappears in the infrared image, there is a lack of connectivity criteria based on the park's physical structure between it and the pre-obstruction UAV location, making it difficult to match the tracking recovery results with the actual spatial transfer relationship. Summary of the Invention

[0005] This application provides a method and system for anti-occlusion UAV target tracking to address the problem of establishing spatial connectivity between the pre-occlusion location and the reappearance area after the hot spot is restored when the hot spot continuity is interrupted by a fixed structure in an industrial park, and to determine whether UAV target tracking can be restored based on this connectivity.

[0006] A first aspect of this application provides an anti-occlusion drone target tracking method, comprising: Based on visible light images of industrial parks, the physical structure boundaries of the parks are identified, a topological addressing space is constructed according to the physical structure boundaries of the parks, and an adjacency transfer table is generated. Map the coordinates of the UAV hotspots in the continuous infrared image to the topology addressing space to generate a topology state sequence; The flight obstruction state is determined based on the adjacency relationship of adjacent addresses in the topology state sequence. In response to the hot spot missing event triggered by the flight obstruction state, a set of frozen nodes is generated based on the end address of the topology state sequence. In response to the hot spot recovery event, the hot spot coordinates of the UAV are mapped to the reproduction area. Verify the transfer path between the frozen node set and the reproduced region in the adjacency transfer table, and output the tracking and recovery results.

[0007] Optionally, in one possible implementation of the first aspect, the visible light image recognition of the industrial park to identify the physical structure boundary of the park, constructing a topological addressing space based on the physical structure boundary of the park, and generating an adjacency transfer table includes: The outlines of parallel pipelines and hyperbolic tower walls in the visible light image are extracted as the physical structure boundaries of the park. Determine the physical gaps between adjacent parallel pipeline outlines, and define each physical gap as a pipe gallery gap region; Multiple equidistant offset lines are generated by extending outward along the hyperbolic tower wall outline, and the area between adjacent equidistant offset lines is defined as the outer annular area of ​​the tower. All the gap regions of the pipe gallery are merged with all the annular regions outside the tower to generate the topology addressing space; Extract the entrance arc segment of the outer annular region adjacent to the hyperbolic tower wall outline of the outlet endpoint of the pipe gallery gap area, and generate the adjacency transfer table based on the spatial connectivity between the outlet endpoint and the entrance arc segment.

[0008] Optionally, in one possible implementation of the first aspect, the step of extracting the outlet endpoint of the pipe gallery gap region and the entrance arc segment of the outer annular region adjacent to the hyperbolic tower wall profile, and generating the adjacency transfer table based on the spatial connectivity between the outlet endpoint and the entrance arc segment, includes: Extract the endpoint along the pipeline pointing to the annular region outside the tower as the outlet endpoint, and take the section of the annular region outside the tower adjacent to the hyperbolic tower wall outline that faces the pipe gallery gap region as the inlet arc segment; Generate corresponding addressing nodes for each of the pipe gallery gap regions and each of the tower outer annular regions; When there is a connecting segment between any two boundaries of the pipe gallery gap area that is not blocked by the outline of the parallel pipeline, an internal adjacent edge is established between the corresponding addressing nodes. In the normal arrangement sequence composed of the multiple equidistant offset lines, the internal adjacency edge is established between the addressing nodes corresponding to the outer annular region of the tower that are in adjacent positions; When there is a straight connecting segment without physical obstruction between the exit endpoint and the entrance arc segment, a cross-domain adjacency edge is established between the addressing nodes corresponding to the regions to which they belong. The adjacency transfer table is generated by summarizing all the addressing nodes, the internal adjacency edges, and the cross-domain adjacency edges.

[0009] Optionally, in one possible implementation of the first aspect, mapping the UAV hotspot coordinates in the continuous infrared image to the topology addressing space to generate a topology state sequence includes: Extract the temperature connectivity region of each frame in the infrared continuous image, and determine the geometric center of the temperature connectivity region as the hot spot coordinates of the UAV. Based on the relative position of the UAV hotspot coordinates and the boundaries of each region in the topology addressing space, the target region into which the UAV hotspot coordinates fall is determined. Extract the identifier of the target region and associate the identifier with the timestamp of the corresponding image frame to generate a region record with timestamp; Arrange the timestamped region records in chronological order to generate the topological state sequence.

[0010] Optionally, in one possible implementation of the first aspect, determining the flight obstruction state based on the adjacency relationship of adjacent addresses in the topological state sequence, generating a set of frozen nodes based on the end addresses of the topological state sequence in response to a hotspot missing event triggered by the flight obstruction state, and mapping the UAV hotspot coordinates to a reproducible region in response to a hotspot recovery event, includes: In response to the hot spot coordinates of the UAV being truncated in the hot spot projection within the gap region of the pipe gallery, a hot spot missing event corresponding to the pipe gallery occlusion state is triggered, and the frozen node set is generated based on the addressing node corresponding to the gap region of the pipe gallery at the end of the topology state sequence before truncation, and the hot spot recovery event is triggered when the hot spot projection is restored to connectivity. In response to the UAV hotspot coordinates entering the curved normal projection shading zone of the outer annular region of the tower, the hotspot missing event corresponding to the cooling tower shading state is triggered, and the frozen node set is generated based on the addressing node corresponding to the outer annular region at the end of the topology state sequence before entry. The hotspot recovery event is triggered when the UAV hotspot coordinates pass through the curved normal projection shading zone.

[0011] Optionally, in one possible implementation of the first aspect, generating the frozen node set based on the addressing nodes corresponding to the end of the tube gallery gap region in the topological state sequence before truncation includes: Extract the pixel distribution pattern of the temperature-connected region in the gap region of the pipe gallery from two consecutively acquired infrared images; When the pixel distribution pattern is divided into multiple connected regions by the parallel pipeline outline, and the multiple connected regions are respectively located on both sides of the parallel pipeline outline, and the centroid line connecting any two adjacent connected regions is parallel to the parallel pipeline outline, it is determined that the hot spot projection truncation has occurred. Extract the current pipe gallery gap region where the UAV hotspot coordinates are located before the hotspot projection is truncated; Based on the adjacency transfer table, a target addressing node with an internal adjacency edge is determined that exists between the addressing node corresponding to the current pipe gallery gap region and the pipe gallery gap region corresponding to the target addressing node is taken as the adjacent pipe gallery gap region. The addressing nodes corresponding to the current pipe gallery gap area and the addressing nodes corresponding to the adjacent pipe gallery gap areas are used together as node elements to generate the frozen node set.

[0012] Optionally, in one possible implementation of the first aspect, the process of determining the surface normal projection occlusion band includes: Extract the sequence of curvature extrema points of the hyperbolic tower wall profile; The extension depth along the normal direction at the corresponding position is determined based on the curvature values ​​of the curvature extremum point sequence. Connect the end points of each of the aforementioned extension depths to generate the outer boundary line; Connect the endpoints of the hyperbolic tower wall outline with the corresponding endpoints of the outer boundary line to generate a side closed line; The closed strip-shaped area enclosed by the hyperbolic tower wall outline, the outer boundary line, and the side closure line is defined as the surface normal projection shading zone, and the outer boundary line is defined as the outer boundary of the surface normal projection shading zone.

[0013] Optionally, in one possible implementation of the first aspect, the response to the UAV hotspot coordinates entering the curved normal projection shielding band of the annular region outside the tower includes: The traverse trajectory is constructed based on the hot spot coordinates of the UAV corresponding to two consecutively acquired infrared images. Extract the intersection point between the crossing trajectory and the outer boundary of the surface normal projection occlusion zone, and calculate the tangential vector of the crossing trajectory at the intersection point; Calculate the angle between the tangential vector and the normal vector of the outer boundary of the surface normal projection occlusion zone at the intersection point; When the included angle indicates that the traversal trajectory extends toward the hyperbolic tower wall outline, it is determined that the hot spot coordinates of the UAV have entered the normal projection shading zone of the curved surface.

[0014] Optionally, in one possible implementation of the first aspect, triggering the hotspot recovery event when the UAV hotspot coordinates exit the surface normal projection occlusion zone includes: Acquire image frames after the hot spot loss event is triggered, and extract the intersection point of the exit trajectory formed by the UAV hot spot coordinates and the outer boundary of the surface normal projection shading zone in the image frames. Determine the target tower outer ring region into which the intersection point falls in the topology addressing space, and extract the starting tower outer ring region contained in the frozen node set; Determine whether the target tower outer ring area and the starting tower outer ring area are adjacent or overlap in the normal arrangement sequence composed of the multiple equidistant offset lines; When the target tower outer annular region is adjacent to or overlaps with the starting tower outer annular region, the hot spot recovery event is triggered, and the target tower outer annular region is identified as the reproduction region.

[0015] Optionally, in one possible implementation of the first aspect, triggering the hotspot recovery event when the UAV hotspot coordinates exit the surface normal projection shading zone further includes: When there are multiple intersection points on the outer boundary of the surface normal projection occlusion zone in the image frame, determine the candidate tower outer annular region into which each intersection point falls. Based on the axis of symmetry of the hyperbolic tower wall contour, the initial tower outer annular region is mirrored to generate a mirrored tower outer annular region. From the candidate tower outer ring region, select the region that is adjacent to or overlaps with the mirror tower outer ring region in the normal arrangement sequence composed of the multiple equidistant offset lines to obtain the effective penetration region; The effective penetration area is defined as the outer ring area of ​​the target tower, and the hot spot recovery event is triggered based on the coordinates of the UAV hot spot that falls into the outer ring area of ​​the target tower.

[0016] Optionally, in one possible implementation of the first aspect, verifying the transfer path between the frozen node set and the reproduced region in the adjacency transfer table, and outputting the tracking and recovery result, includes: In the adjacency transfer table, an undirected graph traversal is performed starting from all the addressable nodes in the frozen node set to generate a reachable node set. Determine whether the addressing node corresponding to the reproduced region is included in the set of reachable nodes; When the addressing node corresponding to the reproduced region is included in the reachable node set, the shortest acyclic path between any addressing node in the frozen node set and the addressing node corresponding to the reproduced region is extracted to obtain the node connectivity sequence. The tracking and recovery result is generated based on the arrangement order of each addressed node in the node connectivity sequence.

[0017] Optionally, in one possible implementation of the first aspect, extracting the shortest acyclic path between any addressable node in the frozen node set and the addressable node corresponding to the reconstructed region to obtain a node connectivity sequence includes: When the frozen node set contains addressing nodes belonging to the pipe gallery gap area, and the addressing node corresponding to the reproduced area belongs to the outer ring area of ​​the tower, the cross-domain adjacency edge connecting the corresponding addressing node is extracted from the adjacency transfer table. Mark the two addressing nodes of the cross-domain adjacent edge as nodes that must be passed through the path; When generating the shortest acyclic path, the necessary nodes of the path are included as constraints in the shortest acyclic path.

[0018] A second aspect of this application provides an anti-interference drone target tracking system, comprising: The boundary recognition module is used to identify the physical structure boundary of the industrial park based on the visible light image of the park, construct a topological addressing space based on the physical structure boundary of the park, and generate an adjacency transfer table. The sequence mapping module is used to map the coordinates of UAV hotspots in continuous infrared images to the topology addressing space, generating a topology state sequence; The status response module is used to determine the flight obstruction status based on the adjacency relationship of adjacent addresses in the topology status sequence, generate a set of frozen nodes based on the end address of the topology status sequence in response to the hot spot missing event triggered by the flight obstruction status, and map the UAV hot spot coordinates to the reproduction area in response to the hot spot recovery event. The path verification module is used to verify the transfer path between the frozen node set and the reproducible region in the adjacency transfer table, and output the tracking and recovery results.

[0019] The anti-occlusion UAV target tracking method and system provided in this application have the following advantages: This application realizes target identity association based on topological path verification between topological addressing space, frozen node set and reconstructed region, avoiding identity misjudgment caused by simply relying on pixel coordinate continuity and kinematic prediction, and can adapt to the spatial transfer that occurs during UAV occlusion, thereby improving the accuracy of UAV target recovery and continuous tracking in complex physical structure occlusion scenarios. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the anti-occlusion UAV target tracking method provided in an embodiment of this application; Figure 2 This is an application scenario diagram of the anti-occlusion UAV target tracking method provided in the embodiments of this application; Figure 3 This is a schematic diagram of the anti-obstruction UAV target tracking system provided in the embodiments of this application; Figure 4 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0022] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0023] See Figure 1 This is a flowchart of the anti-occlusion UAV target tracking method provided in the embodiments of this application. Figure 1 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not impose any limitations on this.

[0024] The anti-obstruction UAV target tracking method provided in this application includes steps S1 to S4, as follows: Step S1: Identify the physical structure boundary of the industrial park based on the visible light image of the park, construct a topological addressing space based on the physical structure boundary of the park, and generate an adjacency transfer table.

[0025] It should be noted that conventional multi-target tracking typically relies on the continuity of pixel coordinates to maintain target identity. However, when large physical structures within an industrial park cause occlusion, the target pixel trajectory is easily interrupted. This application transforms the continuous physical space into a discrete topological structure with connectivity by identifying physical structure boundaries and constructing a topological addressing space and adjacency transfer table.

[0026] Understandably, the input received in step S1 is a visible light image of the industrial park, and the output includes a topology addressing space and an adjacency transfer table. The topology addressing space is a discretized spatial encoding result of the area accessible to UAVs in the industrial park, and the adjacency transfer table is a data structure that records the topological connectivity relationships between discrete spatial units. The adjacency transfer table can store the adjacency relationships between addressing nodes using an adjacency matrix, an adjacency list, or other graph structure data forms.

[0027] Step S11: Extract the outlines of parallel pipelines and hyperbolic tower walls from the visible light image as the physical structural boundaries of the park.

[0028] Understandably, step S11 is used to identify the boundary contours of physical structures with occlusion effects within the industrial park from the visible light image, forming a geometric reference for spatial division. The visible light image contains appearance information of physical structures such as pipe corridors, cooling towers, and factory buildings within the industrial park.

[0029] Parallel pipeline outlines refer to the edge outlines of multiple parallel pipelines within a pipe gallery structure in an industrial park, as seen in visible light images. Pipe galleries typically consist of multiple layers of horizontally or nearly horizontally arranged pipelines, with physical gaps between adjacent pipelines. The edges of these pipelines appear as a set of approximately parallel line segments in the image. Hyperbolic tower wall outlines refer to the outer contour curve of the hyperboloidal tower wall of a cooling tower, as seen in visible light images. Due to heat dissipation requirements, cooling tower walls are typically hyperbolic in shape, tapering inwards from the bottom to the top and then expanding outwards, appearing as arc-shaped edges with a clear curvature change in top-down or side-view images.

[0030] Specifically, extracting the contour lines of parallel pipelines can be achieved by processing the visible light image using an edge detection algorithm to identify edge segments with straight-line characteristics. The parallelism between these segments is then used to filter out the contour lines of parallel pipelines belonging to the same pipe gallery structure. Extracting the contour lines of the hyperbolic tower wall can be done using a curve fitting method. This involves identifying arc-shaped edges with continuous curvature changes from the image and verifying the fit using the hyperbolic equation to determine the edge curves belonging to the cooling tower wall.

[0031] Step S12: Determine the physical gap between the outlines of adjacent parallel pipelines, and define each physical gap as the gap area of ​​the pipe gallery.

[0032] Understandably, step S12 is used to discretize the passable space between adjacent pipes in the pipe gallery structure into independent topological units. Since the pipes appear as strip-shaped regions of a certain width in the image, the area between the edges of adjacent pipes is the physical gap that the drone can pass through.

[0033] Specifically, for each set of parallel pipeline outlines, following the order of the pipelines in the image, the opposite edges of two adjacent pipelines are sequentially selected, and the enclosed area between these two edges is defined as a physical gap. Each physical gap has a unique region identifier, which can serve as the identifier for the gap regions in the utility tunnel. For example, if a utility tunnel structure contains 5 parallel pipelines, then 4 physical gaps can be formed between adjacent pipelines, which are respectively defined as 4 utility tunnel gap regions.

[0034] In this embodiment, the pipe gallery gap region refers to a two-dimensional spatial region enclosed by the outlines of adjacent parallel pipes, which has a defined boundary on the image plane. Each pipe gallery gap region corresponds to an independent addressing unit in the topology addressing space, used to record the topology address of the UAV's spatial position when flying within the pipe gallery region.

[0035] Step S13: Extend outward along the hyperbolic tower wall outline to generate multiple equidistant offset lines, and define the area between adjacent equidistant offset lines as the outer annular area of ​​the tower.

[0036] Understandably, step S13 is used to discretize the passable space around the cooling tower into multiple concentric annular topological units. Here, "outward" refers to the direction away from the center of the cooling tower. The equidistant offset line is a curve generated by translating outward along the normal direction of the hyperbolic tower wall contour line by a fixed offset distance.

[0037] Specifically, along the normal direction of the hyperbolic tower wall outline, and according to a set offset distance, the first equidistant offset line, the second equidistant offset line, and so on, up to the Nth equidistant offset line, can be generated sequentially. The first equidistant offset line is an offset line adjacent to the hyperbolic tower wall outline, and the area between it and the outline can be marked as the first outer annular region. The area between the first and second equidistant offset lines can be marked as the second outer annular region; and so on, with the areas between adjacent equidistant offset lines sequentially defined as the corresponding outer annular regions. Each outer annular region has a unique region identifier.

[0038] The outer annular region refers to the strip-shaped space enclosed by two adjacent equidistant offset lines that surrounds the cooling tower wall. Since the equidistant offset lines are generated along the normal direction of the hyperbola edge, the width of the outer annular region can remain consistent at locations with greater curvature and those with less curvature, thus ensuring that each outer annular region has approximately equal radial width in the normal direction.

[0039] Step S14: Merge all pipe gallery gap areas with all tower outer ring areas to generate a topology addressing space.

[0040] It is understandable that step S14 is used to combine the pipe gallery gap area and the outer ring area generated in the previous steps into a unified addressing space, so that different types of areas can be managed in a unified manner under the same topology framework.

[0041] Specifically, the area identifiers of all pipe gallery gap areas and all outer ring areas are aggregated into a single set to form a topology addressing space. This topology addressing space contains all discretized spatial units accessible to drones within the industrial park.

[0042] In this embodiment, the topology addressing space is a unified spatial coding system composed of multiple types of regions. Among them, the pipe gallery gap region is used to represent the passable locations within the pipe gallery area, and the outer annular region is used to represent the passable locations around the cooling tower.

[0043] Step S15: Extract the entrance arc segment of the outer annular region of the pipe gallery gap area that is adjacent to the hyperbolic tower wall outline, and generate an adjacency transfer table based on the spatial connectivity between the exit endpoint and the entrance arc segment.

[0044] It should be noted that while the topology addressing space has been constructed in steps S11 to S14, topological connectivity between the regions has not yet been established; that is, the system does not yet know which regions are physically reachable. Conventional spatial connectivity determination usually relies on manual labeling or preset connectivity rules. When pipe corridors and cooling towers in an industrial park have spatial intersections or adjacencies, manual labeling is insufficient to accurately reflect the actual passage conditions between physical structures. Step S15 is used to establish an adjacency transfer table based on the spatial connectivity between regions.

[0045] It is understood that the input received in step S15 is the geometric boundary information of all the pipe gallery gap regions and the outer annular region of the tower in the topological addressing space, and the output is an adjacency transfer table. In this embodiment, spatial connectivity refers to the traversable path between two region boundaries that does not traverse the physical structure. This path can be represented as a straight connecting segment or as a continuous reachable relationship formed by multiple continuous boundary segments. By identifying and recording spatial connectivity, the actual passage conditions in physical space can be transformed into edge relationships between nodes in a graph structure, forming structured data describing spatial passage conditions.

[0046] Step S151: Extract the endpoint pointing from the pipeline to the outer annular area of ​​the tower as the outlet endpoint, and take the section of the outer annular area facing the pipe gallery gap area adjacent to the hyperbolic tower wall outline as the inlet arc segment.

[0047] Understandably, step S151 is used to determine the possible spatial connection interface between the pipe gallery gap region and the outer annular region of the tower. Here, the outlet endpoint refers to the boundary endpoint on the side of the pipe gallery gap region extending towards the outer annular region along the pipeline's direction. The inlet arc segment refers to the boundary section in the outer annular region adjacent to the hyperbolic tower wall outline, facing towards the pipe gallery gap region.

[0048] Specifically, for each pipe gallery gap region, the endpoint in its boundary profile that points towards the location of the outer annular region along the pipeline extension direction is extracted, and this endpoint is taken as the outlet endpoint of the pipe gallery gap region. For the outer annular region adjacent to the hyperbolic tower wall profile, the segment in its boundary profile facing the location of the pipe gallery gap region is extracted, and this segment is taken as the entrance arc segment.

[0049] In this embodiment, the outlet endpoint is a geometric point on the boundary of the pipe gallery gap area, used to characterize the outlet position of the pipe gallery gap area towards the external space. The inlet arc segment is a geometric line segment or curve segment on the boundary of the outer annular area of ​​the tower, used to characterize the inlet range of the outer annular area towards the external space.

[0050] Step S152: Generate corresponding addressing nodes for each pipe gallery gap area and each tower outer ring area.

[0051] Understandably, step S152 is used to map each region in the topology addressing space to a node in the graph structure.

[0052] Specifically, a corresponding addressing node is generated for each pipe gallery gap region, and a corresponding addressing node is also generated for each outer annular region of the tower. Each addressing node carries the region identifier of its corresponding area, serving as the unique identifier of that node. In this embodiment, an addressing node refers to a vertex in the graph structure, used to represent an independent addressing unit in the topological addressing space.

[0053] Since the connectivity between regions is essentially a topological relationship rather than a geometric one, each region is abstracted into a uniquely corresponding addressable node. This transforms the subsequent description of connectivity between regions into a description of edge relationships between nodes, providing a unified data representation for the establishment of the adjacency transition table. Through this abstraction process, the region adjacency problem in physical space is transformed into a node connectivity problem in standard graph theory.

[0054] Step S153: When there is a connecting segment between any two pipe gallery gap regions that is not blocked by the outline of the parallel pipeline, establish an internal adjacent edge between the corresponding addressing nodes.

[0055] Understandably, step S153 is used to establish topological connectivity within areas of the same type. Here, an intersecting segment refers to a boundary line segment where the boundary contours of two pipe gallery gap areas are in direct contact or partially overlap. "Not blocked by parallel pipe contour lines" means that at the location of the intersecting segment, there is no solid line segment belonging to the parallel pipe contour lines separating the two areas.

[0056] In this embodiment, an internal adjacency edge refers to an edge connecting addressing nodes of the same type of region, indicating a topological relationship between two regions of the same type that can be directly reached without traversing other types of regions. Establishing an internal adjacency edge indicates that a UAV can enter one pipe rack gap region from another without crossing physical obstacles; therefore, the two regions have a one-time topological reachability relationship. An internal adjacency edge records at least the starting addressing node identifier, the target addressing node identifier, and the edge type, where the edge type distinguishes between internal adjacency edges and cross-domain adjacency edges.

[0057] Specifically, iterate through all pairwise combinations of pipe gallery gap regions. For any two pipe gallery gap regions, determine whether there is a connecting segment between their boundary contours. If a connecting segment exists, and no entity line segment of the parallel pipeline contour line passes through the connecting segment, then the two pipe gallery gap regions are considered physically connected, and an internal adjacency edge is established between their corresponding addressing nodes. If no connecting segment exists, or the connecting segment is completely blocked by the parallel pipeline contour line, then no internal adjacency edge is established.

[0058] Step S154: In the normal arrangement sequence composed of multiple equidistant offset lines, establish internal adjacency edges between the addressing nodes corresponding to the outer ring areas of the tower that are in adjacent positions.

[0059] Understandably, step S154 is used to establish topological connectivity within regions of the same type. Here, the normal arrangement sequence refers to an ordered sequence formed by arranging multiple equidistant offset lines along the normal direction of the hyperbolic tower wall outline in ascending order of offset distance. Being in adjacent positions means that in the normal arrangement sequence, the equidistant offset line intervals corresponding to two outer annular regions are adjacent in the sequence, with no other outer annular regions in between.

[0060] Specifically, the position of each outer-tower annular region in the sequence is determined based on the arrangement order of multiple equidistant offset lines along the normal direction. For two outer-tower annular regions that are adjacent in the normal direction arrangement sequence, an internal adjacency edge is established between their corresponding addressing nodes. Since adjacent outer-tower annular regions share the same equidistant offset line as a common boundary, the UAV can directly enter from one outer-tower annular region to another when moving along the normal direction. The two are directly topologically connected and there is no need to additionally determine whether there is an obstruction.

[0061] Step S155: When there is a straight connected segment without physical obstruction between the exit endpoint and the entrance arc segment, establish a cross-domain adjacency edge between the addressing nodes corresponding to the regions to which they belong.

[0062] Understandably, step S155 is used to establish topological connectivity between different types of regions. In this embodiment, a cross-domain adjacency edge refers to an edge connecting the addressing nodes corresponding to different types of regions, used to indicate a direct spatial connectivity between the pipe gallery gap region and the outer annular region of the tower. A straight-line connectivity segment without physical obstruction refers to a straight-line segment between the outlet endpoint and the inlet arc segment that does not pass through any parallel pipeline outline or hyperbolic tower wall outline enclosed by a physical region.

[0063] Specifically, it is determined whether there is a straight, unobstructed connecting segment between the exit endpoint and the entrance arc segment extracted in step S151. If a straight connecting segment exists, it indicates that there is a physical direct path between the exit of the pipe gallery gap area and the entrance of the outer ring area of ​​the tower. A cross-domain adjacency edge is established between the addressing node corresponding to the pipe gallery gap area to which the exit endpoint belongs and the addressing node corresponding to the outer ring area to which the entrance arc segment belongs. If no straight connecting segment exists, it indicates that there is a physical structural obstruction between the two, and no cross-domain adjacency edge is established.

[0064] Step S156: Summarize all addressing nodes, internal adjacency edges, and cross-domain adjacency edges to generate an adjacency transfer table.

[0065] Understandably, step S156 is used to aggregate all the addressing nodes, internal adjacency edges and cross-domain adjacency edges generated in the preceding steps into a unified graph theory data structure, forming an adjacency transfer table.

[0066] Specifically, all addressable nodes are treated as a vertex set, and all internal and cross-domain adjacency edges are treated as an edge set. Adjacency mapping relationships are established between nodes based on the starting and target addressable nodes corresponding to each edge. The connectivity state between each addressable node is recorded in the form of an adjacency matrix or adjacency linked list, generating an adjacency transition table. In the adjacency transition table, each addressable node records all directly connected addressable nodes to fully record the direct connectivity state between all addressable nodes.

[0067] Step S2: Map the coordinates of the UAV hotspots in the continuous infrared image to the topology addressing space to generate a topology state sequence.

[0068] It should be noted that conventional multi-target tracking typically calculates the target's trajectory directly in the pixel coordinate system. When a drone flies into the gaps in the pipe corridor of an industrial park or is blocked by the wall of a cooling tower, the infrared hotspot may disappear or deform, causing the pixel trajectory to break. This application maps the drone's hotspot coordinates to the topology addressing space constructed in step S1, converting continuous coordinates into discrete topology addresses, and forming a topology state sequence in chronological order.

[0069] It is understandable that the input received in step S2 is the continuous infrared image and the topology addressing space generated in step S1, and the output is the topology state sequence. The continuous infrared image is video stream data containing the thermal characteristics of the UAV, which is collected by an infrared thermal imaging device, and the topology state sequence is an ordered set recording the topology address of the UAV within a continuous time window.

[0070] Step S21: Extract the temperature connectivity region of each frame in the continuous infrared image, and determine the geometric center of the temperature connectivity region as the coordinates of the UAV hot spot.

[0071] Understandably, step S21 is used to locate the physical position of the UAV from the infrared image, converting the planar thermal features into point coordinates to support geometric calculations with the region boundaries. Here, the temperature-connected region refers to a closed region in the infrared image where the temperature is higher than the background environment and the pixels are adjacent; this typically corresponds to the imaging characteristics of heat-generating components such as the UAV's engine, battery, or fuselage in the infrared band.

[0072] Specifically, for each frame of a continuous infrared image, regions with pixel values ​​higher than the background temperature threshold can be extracted using an image segmentation algorithm. Adjacent high-pixel-value regions are then merged to obtain temperature-connected regions. Image segmentation algorithms include thresholding or region growing algorithms. Subsequently, the geometric center of the temperature-connected region on the image plane is calculated. This geometric center can be the centroid of the pixel coordinates, and the two-dimensional pixel coordinates of this geometric center are determined as the coordinates of the UAV hotspot.

[0073] Step S22: Based on the relative position of the UAV hotspot coordinates and the boundaries of each region in the topology addressing space, determine the target region in which the UAV hotspot coordinates fall.

[0074] Understandably, step S22 is used to convert continuous pixel coordinates into discrete topological addresses, realizing the mapping from pixel space to topological space. Here, relative position refers to the spatial inclusion relationship between the UAV hotspot coordinates as a two-dimensional point and the boundaries of each region in the topological addressing space as closed polygons or curves enclosing the region. The target region refers to the actual location of the UAV in the pipe gallery gap region or the outer ring region of the tower in the current frame.

[0075] Specifically, all regions in the topology addressing space are traversed. For each region, it is determined whether the UAV hotspot coordinates are located within the closed area enclosed by the region's boundary. This determination can be made using computational geometry algorithms such as the point-within-a-polygon method. Specifically, a ray is emitted from the UAV hotspot coordinates in any direction, and the number of intersections between the ray and the region boundary is counted. If the number of intersections is odd, the UAV hotspot coordinates are considered to fall within that region. If the UAV hotspot coordinates fall within the closed area of ​​a region, that region is designated as the target region. Each region corresponds to a unique region identifier generated in step S1. Once the target region is determined, the region identifier corresponding to the target region is used as the topology address of the current frame.

[0076] In this embodiment, there is a case branch: if the drone hotspot coordinates do not fall into any region of the topology addressing space, then there is no valid target region in the current frame.

[0077] Step S23: Extract the identifier of the target area and associate the identifier with the timestamp of the corresponding image frame to generate a time-stamped area record.

[0078] It is understandable that step S23 is used to bind spatial address with time information to form a basic data unit containing spatiotemporal information.

[0079] Specifically, the corresponding identifier is extracted from the target region determined in step S22, and the timestamp of the current processed image frame is obtained. The identifier and timestamp are associated, for example, by combining them into key-value pairs or structure objects, to generate a region record with timestamps. If it is determined in step S22 that there is no valid target region in the current frame, a region record with timestamps containing a timestamp and an empty identifier or a specific occlusion identifier can be generated to record the state of the UAV outside the topological space at that moment.

[0080] Step S24: Arrange the time-stamped region records in chronological order to generate a topological state sequence.

[0081] Understandably, step S24 is used to organize discrete timestamped region records into a complete trajectory with temporal direction, forming a topological state sequence. This topological state sequence is a set of region identifiers arranged in chronological order, used to characterize the topological flight trajectory of the UAV over a period of time.

[0082] Specifically, multiple timestamped region records generated from consecutive frames of images are aggregated into a list or queue and sorted in ascending order according to the timestamps contained in each timestamped region record. After sorting, the region identifiers in each timestamped region record are extracted and combined into a topological state sequence in chronological order.

[0083] Step S3: Determine the flight occlusion state based on the adjacency relationship of adjacent addresses in the topology state sequence. In response to the hot spot missing event triggered by the flight occlusion state, generate a set of frozen nodes based on the end address of the topology state sequence, and map the UAV hot spot coordinates to the reproducible area in response to the hot spot recovery event.

[0084] It should be noted that conventional multi-target tracking algorithms typically rely on kinematic models for trajectory prediction when a target is occluded. When a UAV is obscured for an extended period by multi-layered pipes in an industrial park or the curved walls of a cooling tower, the predicted trajectory diverges rapidly, leading to loss of target identity. In the case of deep physical structure occlusion as described in this application, the UAV may change layers or circle the tower during the occlusion period, resulting in a highly uncertain reappearance location. Kinematic prediction alone cannot accurately correlate the target before and after occlusion. Therefore, this application monitors the adjacency changes of addresses in the topology state sequence in step S3, identifies hotspot loss and recovery events caused by physical occlusion, expands the last recorded topology address before occlusion into a set of frozen nodes containing potential transfer paths, and maps the reappearing coordinates to the reconstructed region, forming a set of starting and ending points for path verification.

[0085] Step S31: Determine the triggering method of the hot spot missing event based on the type of obstruction structure corresponding to the flight obstruction state.

[0086] It is understandable that step S31, as a scene routing step, is used to route subsequent occlusion determination and event triggering logic to the corresponding processing branch based on the current physical environment of the drone. In this embodiment, the occlusion structure types mainly include pipe gallery structures and cooling tower structures.

[0087] Step S311: In response to the truncation of the hot spot projection in the pipe gallery gap area by the UAV hot spot coordinates, a hot spot missing event corresponding to the pipe gallery occlusion state is triggered, and a set of frozen nodes is generated based on the addressing node corresponding to the pipe gallery gap area at the end of the topology state sequence before truncation. When the hot spot projection is restored to connectivity, a hot spot recovery event is triggered.

[0088] Understandably, step S311 is used to handle occlusion event triggering in the utility tunnel scenario. When a drone hotspot coordinate is detected to be truncated in the hotspot projection within the utility tunnel gap region, a hotspot missing event corresponding to the utility tunnel occlusion state is triggered. At this time, the addressing node corresponding to the utility tunnel gap region at the end of the topology state sequence before truncation is extracted, and a frozen node set is generated based on this. When a hotspot projection is detected to be restored to connectivity, a hotspot restoration event is triggered.

[0089] Step S312: In response to the UAV hot spot coordinates entering the curved normal projection shading zone of the outer annular region of the tower, a hot spot missing event corresponding to the cooling tower shading state is triggered, and a set of frozen nodes is generated based on the addressing node corresponding to the outer annular region at the end of the pre-entry topology state sequence. When the UAV hot spot coordinates pass through the curved normal projection shading zone, a hot spot recovery event is triggered.

[0090] Understandably, step S312 is used to handle occlusion event triggering in the cooling tower scenario. When the drone's hotspot coordinates are detected to enter the curved normal projection occlusion zone of the outer annular region, a hotspot missing event corresponding to the cooling tower occlusion state is triggered. At this time, the addressing node corresponding to the outer annular region at the end of the pre-entry topology state sequence is extracted, and a frozen node set is generated based on this. When the drone's hotspot coordinates are detected to pass through the curved normal projection occlusion zone, a hotspot recovery event is triggered.

[0091] Step S32: Generate a set of frozen nodes based on the addressing nodes corresponding to the gap region of the pipe gallery at the end of the topological state sequence before truncation.

[0092] It should be noted that this step is a further expansion of the pipe gallery scenario in step S311. In the pipe gallery obstruction scenario, when a drone is obstructed after flying behind a pipe, it is very likely to change altitude during the obstruction period, thus switching from the current pipe gallery gap to another adjacent pipe gallery gap. If only the single area where it was located before obstruction is used as the search starting point, when the drone flies out from another adjacent pipe gallery gap, the system will not be able to find a connecting path, resulting in identity association failure. Therefore, this application packages the current area and its directly connected adjacent areas on the topology network together in step S32 to generate a frozen node set containing multiple potential starting points, in order to cover the inter-layer transfers that the drone may make during obstruction.

[0093] Step S321: Extract the pixel distribution pattern of the temperature connectivity region in the gap region of the pipe gallery from two consecutively acquired infrared images.

[0094] Understandably, step S321 is used to acquire the visual feature changes of the UAV hotspot within the gaps in the pipe gallery, as a basis for determining whether occlusion has occurred. Here, pixel distribution morphology refers to the shape, area, and segmentation of the temperature-connected region on the image plane.

[0095] Specifically, in two consecutively acquired infrared images, the gap region of the pipe gallery determined in step S22 is located, and the set of pixels of the temperature connected region located within the boundary range of the region is extracted to analyze its connectivity, contour shape and other pixel distribution patterns.

[0096] Step S322: When the pixel distribution pattern is divided into multiple connected regions by the parallel pipeline outline, and the multiple connected regions are located on both sides of the parallel pipeline outline, and the centroid line connecting any two adjacent connected regions is parallel to the parallel pipeline outline, it is determined that hot spot projection truncation has occurred.

[0097] Understandably, step S322 is used to determine, through strict geometric conditions, whether the drone is visually deeply obscured by the utility tunnel pipes.

[0098] Specifically, it is determined whether the pixel distribution pattern is segmented by the parallel pipeline contour line extracted in step S11. If the originally single temperature connected region is cut by the parallel pipeline contour line, forming multiple discrete connected regions, and these connected regions are located on both sides of the parallel pipeline contour line, that is, the pipeline runs across the middle of the hot spot, the geometric center of any two adjacent connected regions is calculated. If their connecting line is parallel to the parallel pipeline contour line, it indicates that the UAV's hot spot is split by the pipeline entity along the pipeline's direction. In this case, it is determined that hot spot projection truncation has occurred.

[0099] If the pixel distribution pattern remains uniformly connected, or if the centroid connection is not parallel to the pipeline outline even though the pixels are segmented, then it is determined that no hot spot projection truncation has occurred.

[0100] Step S323: Extract the current pipe gallery gap region where the UAV hot spot coordinates are located before the hot spot projection is truncated.

[0101] Understandably, step S323 is used to determine the last confirmed topology location before the drone enters the obstruction state.

[0102] Specifically, from the topological state sequence generated in step S2, the pipe gallery gap region where the UAV hot spot coordinates fall in the previous frame or several frames of the image before the hot spot projection truncation occurs is extracted and used as the current pipe gallery gap region.

[0103] Step S324: Based on the adjacency transfer table, determine the target addressing node that has an internal adjacency edge with the addressing node corresponding to the current pipe gallery gap area, and take the pipe gallery gap area corresponding to the target addressing node as the adjacent pipe gallery gap area.

[0104] Understandably, step S324 is used to locate potential transfer locations in the topology network that are directly connected to the current location.

[0105] Specifically, the adjacency transfer table generated in step S1 is queried to find all other addressable nodes that have an internal adjacency edge with the addressable node corresponding to the current pipe gallery gap area, and these addressable nodes are designated as target addressable nodes. Then, the pipe gallery gap areas corresponding to these target addressable nodes are determined as adjacent pipe gallery gap areas.

[0106] Step S325: Combine the addressing nodes corresponding to the current pipe gallery gap area and the addressing nodes corresponding to the adjacent pipe gallery gap areas as node elements to generate a frozen node set.

[0107] Understandably, step S325 is used to construct a set containing multiple starting points to address unknown transitions during occlusion.

[0108] Specifically, the addressing nodes corresponding to the current pipe gallery gap area, as well as the addressing nodes corresponding to all adjacent pipe gallery gap areas determined in step S324, are put into a set data structure to generate a frozen node set.

[0109] Step S33: Determination of the surface normal projection occlusion zone.

[0110] It should be noted that this step, and subsequent steps S34 to S36, are further developments of the cooling tower scene in step S312. In the cooling tower occlusion scene, because the cooling tower wall has a hyperboloid shape, its projected width in the image undergoes non-linear perspective contraction with changes in height and viewing angle. If a fixed-width occlusion area is used, occlusion may be missed in areas with greater curvature, and occlusion may be misjudged in areas with less curvature. Therefore, this application dynamically generates a surface normal projection occlusion band based on the local curvature of the hyperboloid tower wall outline in step S33, making it geometrically fit the actual occlusion range of the cooling tower entity in the image.

[0111] Step S331: Extract the sequence of curvature extrema points of the hyperbolic tower wall profile.

[0112] Understandably, step S331 is used to obtain the key geometric features of the cooling tower wall profile. The curvature extremum point sequence refers to a set of pixels or geometric points on the hyperbolic tower wall profile whose curvature values ​​reach local maxima or minima, arranged sequentially.

[0113] Specifically, the curvature values ​​of each point on the hyperbolic tower wall contour line extracted in step S11 are calculated, and points where the curvature values ​​change significantly or reach extreme values ​​are selected and sorted according to the extension direction along the edge curve to generate a sequence of curvature extreme value points.

[0114] Step S332: Based on the curvature extremum point sequence, determine the extension depth along the normal direction at the corresponding position according to the curvature value of each extremum point.

[0115] Specifically, for each extremum point in the curvature extremum point sequence, the depth to which it extends outward along the normal direction of the hyperbolic tower wall profile is calculated. There is a mapping relationship between the extension depth and the curvature value of that extremum point. A larger extension depth is set for locations with greater curvature and steeper tower wall curvature, and a smaller extension depth is set for locations with less curvature and gentler tower wall curvature.

[0116] Step S333: Connect the end points of each extension depth to generate the outer boundary line.

[0117] Specifically, the endpoints obtained by translating each extreme point along the normal direction to the corresponding extension depth are connected sequentially to form a smooth curve, which serves as the outer boundary line.

[0118] Step S334: Connect the endpoints of the hyperbolic tower wall outline with the corresponding endpoints of the outer boundary line to generate a side closed line.

[0119] Specifically, the two endpoints of the hyperbolic tower wall contour within the effective range of the image, as well as the two corresponding endpoints on the outer boundary line, are extracted, and the corresponding endpoints are connected in pairs to generate two side closed lines.

[0120] Step S335: The closed strip-shaped area enclosed by the hyperbolic tower wall outline, the outer boundary line, and the side closure line is defined as the surface normal projection shading zone, and the outer boundary line is defined as the outer boundary of the surface normal projection shading zone.

[0121] Specifically, the closed polygonal or curved polygonal region enclosed by the hyperbolic tower wall outline, the outer boundary line, and the two side closure lines is defined as the surface normal projection shading zone. Simultaneously, the outer boundary line is separately marked as the outer boundary of the surface normal projection shading zone to determine whether a drone enters or exits this area.

[0122] Step S34: Responding to the UAV hotspot coordinates entering the curved surface normal projection shading zone of the annular region outside the tower.

[0123] It should be noted that after determining the surface normal projection shielding zone in step S33, it is necessary to determine whether the UAV's flight trajectory has actually entered the shielding zone to rule out the possibility that the UAV is merely flying parallel to the edge of the shielding zone. Therefore, this application analyzes the motion vector direction of the UAV when crossing the outer boundary of the shielding zone in step S34 to ensure that only trajectories flying towards the tower wall are considered to have entered the shielding zone.

[0124] Step S341: Construct the traverse trajectory based on the UAV hotspot coordinates corresponding to the two consecutively acquired infrared images.

[0125] Specifically, the thermal spot coordinates of the UAV are extracted from two consecutively acquired infrared images. The thermal spot coordinates of the previous frame are connected with the thermal spot coordinates of the current frame to form a directed line segment, which serves as the traverse trajectory.

[0126] Step S342: Extract the intersection point of the traversal trajectory and the outer boundary of the surface normal projection occlusion zone, and calculate the tangential vector of the traversal trajectory at the intersection point.

[0127] Specifically, the geometric intersection point between the crossing trajectory and the outer boundary of the surface normal projection occlusion zone determined in step S335 is calculated. If an intersection point exists, it is taken as the intersection point. The direction vector of the crossing trajectory is extracted as the tangential vector at the intersection point, which represents the direction of motion of the UAV when crossing the boundary.

[0128] Step S343: Calculate the angle between the tangential vector and the normal vector at the intersection point of the surface normal projection shading zone outer boundary.

[0129] Specifically, the normal vector at the intersection point of the outer boundary of the surface normal projection shading zone is calculated, and this normal vector points inward into the shading zone. Then, the angle between the tangential vector and this normal vector is calculated.

[0130] Step S344: When the included angle represents the trajectory extending toward the hyperbolic tower wall outline, determine that the UAV hotspot coordinates have entered the surface normal projection occlusion zone.

[0131] Specifically, it is determined whether the included angle is less than a set angle threshold. If the included angle is less than the threshold, it means that the tangential vector and the inward-pointing normal vector are roughly in the same direction, indicating that the trajectory extends towards the hyperbolic tower wall outline, meaning that the UAV is flying into the shielding zone. In this case, it is determined that the UAV's hotspot coordinates have entered the surface normal projection shielding zone. If the included angle is greater than or equal to the threshold, it means that the UAV is flying tangentially along the boundary or flying away from the boundary, and it is determined that it has not entered.

[0132] Step S35: Trigger a hot spot recovery event when the UAV hot spot coordinates pass through the surface normal projection occlusion zone.

[0133] It should be noted that when a drone exits the cooling tower shielding zone, its exit position may not be within the same annular area as its entry position due to radial displacement along the tower wall during the shielding period. Forcing the exit position to be exactly the same as the entry position would lead to the omission of many normal recovery events. Therefore, this application, through step S35, allows the exit position to be adjacent to or coincide with the starting position in the normal arrangement sequence to accommodate reasonable radial displacement during the shielding period.

[0134] Step S351: Acquire the image frame after the hot spot missing event is triggered, and extract the intersection point of the exit trajectory formed by the UAV hot spot coordinates and the outer boundary of the surface normal projection occlusion zone in the image frame.

[0135] Specifically, after the hot spot loss event is triggered in step S34, the acquired image frames are continuously monitored. When it is detected that the hot spot coordinates of the UAV intersect with the outer boundary of the curved surface normal projection shielding zone again, and the direction of movement is away from the tower wall, the intersection point is extracted as the exit intersection point, and the line segment connecting the last known coordinates in the shielding zone and the exit intersection point is taken as the exit trajectory.

[0136] Step S352: Determine the target tower outer ring region in which the intersection point falls in the topology addressing space, and extract the starting tower outer ring region contained in the frozen node set.

[0137] Specifically, according to the method in step S22, it is determined which outer ring region the intersection point falls into in the topology addressing space, and this is taken as the target outer ring region. At the same time, the outer ring region where the UAV was located before the missing event was triggered is extracted from the set of frozen nodes generated in step S31, and this is taken as the starting outer ring region.

[0138] Step S353: Determine whether the target tower outer ring area and the starting tower outer ring area are adjacent or overlapped in the normal arrangement sequence composed of multiple equidistant offset lines.

[0139] Specifically, in step S13, the normal arrangement sequence composed of multiple equidistant offset lines is queried to obtain the index positions of the target tower outer ring region and the starting tower outer ring region in the sequence. If the index positions are the same, they are determined to be overlapping. If the index positions differ by one, they are determined to be adjacent.

[0140] Step S354: When the target tower outer ring area is adjacent to or overlaps with the starting tower outer ring area, a hot spot recovery event is triggered, and the target tower outer ring area is determined as the reproduction area.

[0141] Specifically, if the judgment result of step S353 is adjacent or overlapping, it indicates that the radial displacement of the UAV during the obstruction period is within a reasonable range. In this case, a hot spot recovery event is triggered, and the annular area outside the target tower is identified as the reproduction area. If they are not adjacent, for example, separated by multiple annular areas, they are considered to be other interfering heat sources, and a hot spot recovery event is not triggered.

[0142] Step S36: When there are multiple penetration points, filter the valid penetration area based on mirror mapping and trigger the hot spot recovery event.

[0143] It should be noted that this step is a further development of the cooling tower scenario in step S312 under the condition of multiple heat source interference. In complex industrial park environments, there may be other heat-generating equipment behind the cooling tower, such as exhaust pipes or other aircraft, which can cause the illusion of multiple heat sources simultaneously passing through the boundary in the image at the same time the drone passes through the shielding zone. If only the adjacent or overlapping verification in step S35 is relied upon, the background heat source may be mistaken for the target. Therefore, this application introduces the geometric symmetry of the cooling tower in step S36, and uses mirror mapping to filter out the effective penetration area that conforms to the laws of physical symmetry, thereby locking the true target among multiple candidate heat sources.

[0144] Step S361: When there are multiple intersection points on the outer boundary of the surface normal projection occlusion zone in the image frame, determine the candidate tower outer ring region into which each intersection point falls.

[0145] Specifically, when multiple intersection points are detected on the outer boundary of the surface normal projection occlusion zone in the image frame, the outer ring region of each intersection point in the topology addressing space is determined and used as a candidate outer ring region.

[0146] Step S362: Based on the axis of symmetry of the hyperbolic tower wall contour, mirror the initial outer annular region of the tower to generate a mirrored outer annular region.

[0147] Specifically, the axis of symmetry of the hyperbolic tower wall contour line in step S11 is extracted. Using this axis of symmetry as a mirror line, the initial annular region outside the tower extracted in step S352 is geometrically mirrored to generate a mirrored annular region outside the tower. This mirrored annular region outside the tower represents the symmetrical position that the UAV should theoretically pass through after flying around the tower along a symmetrical trajectory.

[0148] Step S363: From the candidate tower outer ring region, select the region that is adjacent to or overlaps with the mirror tower outer ring region in the normal arrangement sequence composed of multiple equidistant offset lines to obtain the effective penetration region.

[0149] Specifically, all candidate tower outer ring regions are traversed, and regions adjacent to or overlapping with the mirror tower outer ring region are selected from the normal arrangement sequence composed of multiple equidistant offset lines. The selected regions are determined as valid penetration regions.

[0150] Step S364: Determine the effective penetration area as the outer ring area of ​​the target tower, and trigger the hot spot recovery event based on the hot spot coordinates of the UAV that fell into the outer ring area of ​​the target tower.

[0151] Specifically, the effective penetration area is defined as the annular region outside the target tower, and a hotspot recovery event is triggered based on the coordinates of the hotspots of UAVs that fall into the annular region outside the target tower. Other candidate heat sources that do not fall into the effective penetration area are filtered out.

[0152] Step S4: Verify the transfer paths of the frozen node set and the reproducible region in the adjacency transfer table, and output the tracking and recovery results.

[0153] It should be noted that conventional multi-target tracking algorithms typically associate a target with its closest spatial distance or highest similarity in appearance features when the target reappears. However, in the complex physical environment of industrial parks, when a drone flies out of the obstruction area of ​​a pipe gallery or cooling tower, the visually reappearing heat spot may not be the original target. For example, another drone may fly out from an adjacent pipe gallery, or a background heat source behind a cooling tower may be misdetected. Furthermore, the original target may have undergone significant spatial shifts during the obstruction period, resulting in its reappearance location being far from the predicted location. When the deep obstruction reappearance situation described in this application occurs, relying solely on distance or appearance features is highly prone to misidentification or loss of the target.

[0154] Understandably, the input received in step S4 is the set of frozen nodes and the re-enactment region generated in step S3, as well as the adjacency transfer table generated in step S1, and the output is the tracking recovery result. The tracking recovery result is used to indicate whether the target appearing before and after occlusion is the same object, and when consistency is confirmed, the re-enacted topology address is appended to the historical trajectory.

[0155] Step S41: In the adjacency transition table, perform an undirected graph traversal starting from all addressable nodes in the frozen node set to generate a reachable node set; determine whether the addressable node corresponding to the reconstructed region is included in the reachable node set; when the addressable node corresponding to the reconstructed region is included in the reachable node set, extract the shortest acyclic path between any addressable node in the frozen node set and the addressable node corresponding to the reconstructed region to obtain a node connectivity sequence; generate the tracking recovery result based on the arrangement order of each addressable node in the node connectivity sequence.

[0156] Step S41 is used to complete the topological path verification between the frozen node set and the reconstructed area, and generate the tracking recovery result when it is confirmed that there is a physical reachability relationship between the two. Among them, when the process of generating the shortest acyclic path involves cross-domain transfer scenarios, the path generation method is further limited by step S42.

[0157] Step S411: In the adjacency transition list, perform an undirected graph traversal starting from all the addressable nodes in the frozen node set to generate a reachable node set.

[0158] Understandably, step S411 is used to explore all physically reachable spatial locations in the topological network starting from multiple potential starting points. Here, undirected graph traversal refers to a graph search algorithm, such as breadth-first search, that explores all reachable vertices layer by layer outward from a given set of starting points without considering edge direction.

[0159] Specifically, all addressable nodes in the frozen node set generated in step S3 are used as the starting set. An undirected graph traversal is performed in the adjacency transition table generated in step S1 to record all visited addressable nodes, thus generating a reachable node set. The reachable node set contains addressable nodes corresponding to all areas that the UAV can theoretically reach during the obstruction period.

[0160] Step S412: Determine whether the addressing node corresponding to the reproducible region is included in the set of reachable nodes.

[0161] Specifically, it determines whether the addressable node corresponding to the reproduced area is included in the reachable node set generated in step S411. There are two possible outcomes. First, if the addressable node corresponding to the reproduced area is included in the reachable node set, it means there is a physically reachable topological path from the potential location before occlusion to the reproduced location, and the system will continue to perform path extraction. Second, if the addressable node corresponding to the reproduced area is not included in the reachable node set, it means the reproduced hotspot is completely disconnected from the location before occlusion in terms of physical topology, and is determined to be an interfering heat source or a new target. The system does not generate a node connectivity sequence, and the tracking recovery result indicates an inconsistent identity.

[0162] Step S413: When the addressable node corresponding to the reproducible region is included in the reachable node set, extract the shortest acyclic path between any addressable node in the frozen node set and the addressable node corresponding to the reproducible region to obtain the node connectivity sequence.

[0163] Specifically, when step S412 determines that the addressable node corresponding to the reproducible region is included in the reachable node set, the shortest acyclic path from each addressable node in the starting point set to the corresponding addressable node in the reproducible region is calculated. The path with the shortest path length or the fewest adjacent edges is selected, and its included addressable nodes are extracted sequentially to obtain the node connectivity sequence. In this embodiment, when the starting region corresponding to the frozen node set and the reproducible region belong to the same type of region, the shortest acyclic path can be directly extracted; when they belong to different types of regions, the shortest acyclic path generation process is further executed according to step S42 to ensure that the generated path conforms to the cross-domain physical access constraints.

[0164] Step S414: Generate tracking recovery results based on the arrangement order of each addressed node in the node connectivity sequence.

[0165] Specifically, when the node connectivity sequence is obtained, a tracking recovery result is generated based on the arrangement order of each addressed node in the node connectivity sequence. The tracking recovery result includes an identifier confirming the identity, and the identifier of the reproduced area is appended to the end of the topology state sequence generated in step S2 to restore continuous tracking of the UAV.

[0166] Step S42: When the frozen node set contains addressable nodes belonging to the gap area of ​​the pipe gallery, and the addressable node corresponding to the reproducible area belongs to the outer ring area of ​​the tower, extract the cross-domain adjacency edge connecting the corresponding addressable node in the adjacency transfer table; mark the addressable nodes at both ends of the cross-domain adjacency edge as the path must-pass nodes; when generating the shortest acyclic path, include the path must-pass nodes as constraints in the shortest acyclic path.

[0167] Step S42 is a specific implementation of generating the shortest acyclic path in step S413, and is used to impose physical access constraints on the generation process of the shortest acyclic path in cross-domain transfer scenarios.

[0168] Step S421: When the frozen node set contains addressable nodes belonging to the gap area of ​​the pipe gallery, and the addressable node corresponding to the reproducible area belongs to the outer ring area of ​​the tower, extract the cross-domain adjacency edge connecting the corresponding addressable node from the adjacency transfer table.

[0169] Specifically, in step S413, during the extraction of the shortest acyclic path, the region types of the starting and ending points are first determined. When it is detected that the frozen node set contains addressing nodes belonging to the pipe gallery gap region, and the addressing node corresponding to the reproduced region belongs to the outer ring region, it is determined that the current scenario is a cross-domain transfer scenario. At this time, the cross-domain adjacency edge connecting the addressing nodes corresponding to these two types of regions is extracted from the adjacency transfer table. The cross-domain adjacency edge is a specific adjacency edge established in step S1 that connects the pipe gallery gap region and the outer ring region, representing the unique physical passage interface between the two.

[0170] Step S422: Mark the two end addressing nodes of the cross-domain adjacent edge as nodes that must be passed through the path.

[0171] Specifically, the addressable nodes at both ends of the cross-domain adjacent edge extracted in step S421 are marked as mandatory nodes on the path. These two addressable nodes are the exit node on the pipe gallery side and the inlet node on the cooling tower side. Mandatory nodes on the path refer to vertices that must be traversed during the path search process.

[0172] Step S423: When generating the shortest acyclic path, include the necessary nodes of the path as constraints in the shortest acyclic path.

[0173] Specifically, when calculating the shortest path, the path must pass through all necessary nodes to ensure that the generated node connectivity sequence truly reflects the transfer process of the UAV from the pipe gallery to the cooling tower periphery through a specific physical interface. If the starting point and the ending point belong to the same type of area, such as both inside the pipe gallery or both outside the cooling tower, it belongs to the same-domain transfer scenario, and this constraint is not imposed; the node connectivity sequence is directly generated according to the conventional shortest path algorithm.

[0174] Steps S421 to S423 together complete the process of generating the shortest acyclic path in the cross-domain transfer scenario in step S413. By introducing the constraint of mandatory nodes along the path, the shortest acyclic path must pass through the actual cross-domain physical interface, thereby ensuring that the node connectivity sequence conforms to the actual physical traffic rules.

[0175] See Figure 3 This is a schematic diagram of the structure of the anti-occlusion UAV target tracking system provided in the embodiments of this application, including: The boundary recognition module is used to identify the physical structure boundary of the industrial park based on visible light images, construct a topological addressing space based on the physical structure boundary of the park, and generate an adjacency transfer table. The sequence mapping module is used to map the coordinates of UAV hotspots in continuous infrared images to the topology addressing space, generating a topology state sequence. The state response module is used to determine the flight occlusion state based on the adjacency relationship of adjacent addresses in the topology state sequence, generate a set of frozen nodes based on the end address of the topology state sequence in response to the hot spot missing event triggered by the flight occlusion state, and map the UAV hot spot coordinates to the reproduction area in response to the hot spot recovery event. The path verification module is used to verify the transfer path between the frozen node set and the reproducible region in the adjacency transfer table, and output the tracking and recovery results.

[0176] See Figure 2 This is an application scenario diagram of this application. The image acquisition device 202 is positioned facing the physical structure 201 of the industrial park, used to acquire visible light images of the physical structure 201 and continuous infrared images of the target drone 203 during flight. When the target drone 203 is flying between or behind the physical structure 201 (such as a pipe gallery or cooling tower) and is obstructed, this application divides the space between adjacent pipes in the pipe gallery into a pipe gallery gap region 204, and divides the space surrounding the cooling tower into an outer ring region 205 to construct a topology addressing space. When the target drone 203 reappears after exiting the obstruction area, the system maps its hotspot coordinates to a reproduction region 206, and combines this with the frozen node set generated before obstruction. The frozen node set and the reproduction region 206 are used to represent the topology space range corresponding to target tracking in the adjacency transfer table verification performed by the method of this application.

[0177] Figure 3 The system of the illustrated embodiment can be used to perform corresponding operations. Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0178] See Figure 4This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. The electronic device 40 includes: a processor 41, a memory 42, and a computer program; wherein, The memory 42 is used to store computer programs, and the memory may also be flash memory. Computer programs may be, for example, application programs or functional modules that implement the methods described above.

[0179] Processor 41 is used to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0180] Alternatively, the memory 42 can be either standalone or integrated with the processor 41.

[0181] When the memory 42 is a device independent of the processor 41, the device may also include: Bus 43 is used to connect memory 42 and processor 41.

[0182] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. An anti-obstruction UAV target tracking method, characterized in that, include: Based on visible light images of industrial parks, the physical structure boundaries of the parks are identified, a topological addressing space is constructed according to the physical structure boundaries of the parks, and an adjacency transfer table is generated. Map the coordinates of the UAV hotspots in the continuous infrared image to the topology addressing space to generate a topology state sequence; The flight obstruction state is determined based on the adjacency relationship of adjacent addresses in the topology state sequence. In response to the hot spot missing event triggered by the flight obstruction state, a set of frozen nodes is generated based on the end address of the topology state sequence. In response to the hot spot recovery event, the hot spot coordinates of the UAV are mapped to the reproduction area. Verify the transfer path between the frozen node set and the reproduced region in the adjacency transfer table, and output the tracking and recovery results.

2. The method according to claim 1, characterized in that, The method for recognizing the physical structure boundary of an industrial park based on visible light images, constructing a topological addressing space based on the physical structure boundary, and generating an adjacency transfer table includes: The outlines of parallel pipelines and hyperbolic tower walls in the visible light image are extracted as the physical structure boundaries of the park. Determine the physical gaps between adjacent parallel pipeline outlines, and define each physical gap as a pipe gallery gap region; Multiple equidistant offset lines are generated by extending outward along the hyperbolic tower wall outline, and the area between adjacent equidistant offset lines is defined as the outer annular area of ​​the tower. All the gap regions of the pipe gallery are merged with all the annular regions outside the tower to generate the topology addressing space; Extract the entrance arc segment of the outer annular region adjacent to the hyperbolic tower wall outline of the outlet endpoint of the pipe gallery gap area, and generate the adjacency transfer table based on the spatial connectivity between the outlet endpoint and the entrance arc segment.

3. The method according to claim 2, characterized in that, The step of extracting the outlet endpoint of the pipe gallery gap region and the entrance arc segment of the outer annular region adjacent to the hyperbolic tower wall contour line, and generating the adjacency transfer table based on the spatial connectivity between the outlet endpoint and the entrance arc segment, includes: Extract the endpoint along the pipeline pointing to the annular region outside the tower as the outlet endpoint, and take the section of the annular region outside the tower adjacent to the hyperbolic tower wall outline that faces the pipe gallery gap region as the inlet arc segment; Generate corresponding addressing nodes for each of the pipe gallery gap regions and each of the tower outer annular regions; When there is a connecting segment between any two boundaries of the pipe gallery gap area that is not blocked by the outline of the parallel pipeline, an internal adjacent edge is established between the corresponding addressing nodes. In the normal arrangement sequence composed of the multiple equidistant offset lines, the internal adjacency edge is established between the addressing nodes corresponding to the outer annular region of the tower that are in adjacent positions; When there is a straight connecting segment without physical obstruction between the exit endpoint and the entrance arc segment, a cross-domain adjacency edge is established between the addressing nodes corresponding to the regions to which they belong. The adjacency transfer table is generated by summarizing all the addressing nodes, the internal adjacency edges, and the cross-domain adjacency edges.

4. The method according to claim 1, characterized in that, The step of mapping the UAV hotspot coordinates in the continuous infrared image to the topology addressing space to generate a topology state sequence includes: Extract the temperature connectivity region of each frame in the infrared continuous image, and determine the geometric center of the temperature connectivity region as the hot spot coordinates of the UAV. Based on the relative position of the UAV hotspot coordinates and the boundaries of each region in the topology addressing space, the target region into which the UAV hotspot coordinates fall is determined. Extract the identifier of the target region and associate the identifier with the timestamp of the corresponding image frame to generate a region record with timestamp; Arrange the timestamped region records in chronological order to generate the topological state sequence.

5. The method according to claim 1, characterized in that, The steps include determining the flight obstruction state based on the adjacency relationship of adjacent addresses in the topological state sequence, generating a set of frozen nodes based on the end addresses of the topological state sequence in response to a hotspot loss event triggered by the flight obstruction state, and mapping the UAV hotspot coordinates to a reconstructed region in response to a hotspot recovery event, including: In response to the hot spot coordinates of the UAV being truncated in the hot spot projection within the gap region of the pipe gallery, the hot spot missing event corresponding to the pipe gallery occlusion state is triggered, and the frozen node set is generated based on the addressing node corresponding to the gap region of the pipe gallery at the end of the topology state sequence before truncation, and the hot spot recovery event is triggered when the hot spot projection is restored to connectivity. In response to the UAV hotspot coordinates entering the curved normal projection shading zone of the outer annular region of the tower, the hotspot missing event corresponding to the cooling tower shading state is triggered, and the frozen node set is generated based on the addressing node corresponding to the outer annular region at the end of the topology state sequence before entry. The hotspot recovery event is triggered when the UAV hotspot coordinates pass through the curved normal projection shading zone.

6. The method according to claim 5, characterized in that, The process of generating the frozen node set based on the addressing nodes corresponding to the pipe gallery gap region at the end of the topological state sequence before truncation includes: Extract the pixel distribution pattern of the temperature-connected region in the gap region of the pipe gallery from two consecutively acquired infrared images; When the pixel distribution pattern is divided into multiple connected regions by the parallel pipeline outline, and the multiple connected regions are located on both sides of the parallel pipeline outline, and the centroid line connecting any two adjacent connected regions is parallel to the parallel pipeline outline, it is determined that the hot spot projection truncation has occurred. Extract the current pipe gallery gap region where the UAV hotspot coordinates are located before the hotspot projection is truncated; Based on the adjacency transfer table, a target addressing node that has an internal adjacency edge with the addressing node corresponding to the current pipe gallery gap region is determined, and the pipe gallery gap region corresponding to the target addressing node is taken as the adjacent pipe gallery gap region. The addressing nodes corresponding to the current pipe gallery gap area and the addressing nodes corresponding to the adjacent pipe gallery gap areas are used together as node elements to generate the frozen node set.

7. The method according to claim 5, characterized in that, The process of determining the surface normal projection occlusion zone includes: Extract the sequence of curvature extrema points of the hyperbolic tower wall profile; The extension depth along the normal direction at the corresponding position is determined based on the curvature values ​​of the curvature extremum point sequence. Connect the end points of each of the aforementioned extension depths to generate the outer boundary line; Connect the endpoints of the hyperbolic tower wall outline with the corresponding endpoints of the outer boundary line to generate a side closed line; The closed strip-shaped area enclosed by the hyperbolic tower wall outline, the outer boundary line, and the side closure line is defined as the surface normal projection shading zone, and the outer boundary line is defined as the outer boundary of the surface normal projection shading zone.

8. The method according to claim 5, characterized in that, The curved normal projection shielding band responding to the UAV hotspot coordinates entering the annular region outside the tower includes: The traverse trajectory is constructed based on the hot spot coordinates of the UAV corresponding to two consecutively acquired infrared images. Extract the intersection point between the crossing trajectory and the outer boundary of the surface normal projection occlusion zone, and calculate the tangential vector of the crossing trajectory at the intersection point; Calculate the angle between the tangential vector and the normal vector of the outer boundary of the surface normal projection occlusion zone at the intersection point; When the included angle indicates that the traversal trajectory extends toward the hyperbolic tower wall outline, it is determined that the hot spot coordinates of the UAV have entered the normal projection shading zone of the curved surface.

9. The method according to claim 5, characterized in that, The process of triggering the hotspot recovery event when the hotspot coordinates of the UAV cross the surface normal projection occlusion zone includes: Acquire image frames after the hot spot loss event is triggered, and extract the intersection point of the exit trajectory formed by the UAV hot spot coordinates and the outer boundary of the surface normal projection shading zone in the image frames. Determine the target tower outer ring region into which the intersection point falls in the topology addressing space, and extract the starting tower outer ring region contained in the frozen node set; Determine whether the target tower outer ring area and the starting tower outer ring area are adjacent or overlap in the normal arrangement sequence composed of multiple equidistant offset lines; When the target tower outer annular region is adjacent to or overlaps with the starting tower outer annular region, the hot spot recovery event is triggered, and the target tower outer annular region is identified as the reproduction region.

10. The method according to claim 9, characterized in that, The method of triggering the hotspot recovery event when the hotspot coordinates of the UAV cross the surface normal projection occlusion zone also includes: When there are multiple intersection points on the outer boundary of the surface normal projection occlusion zone in the image frame, determine the candidate tower outer annular region into which each intersection point falls. Based on the axis of symmetry of the hyperbolic tower wall contour, the initial tower outer annular region is mirrored to generate a mirrored tower outer annular region. From the candidate tower outer ring region, select the region that is adjacent to or overlaps with the mirror tower outer ring region in the normal arrangement sequence composed of the multiple equidistant offset lines to obtain the effective penetration region; The effective penetration area is defined as the outer ring area of ​​the target tower, and the hot spot recovery event is triggered based on the coordinates of the UAV hot spot that falls into the outer ring area of ​​the target tower.

11. The method according to claim 1, characterized in that, The process verifies the transfer path between the frozen node set and the reproduced region in the adjacency transfer table, and outputs the tracking and recovery results, including: In the adjacency transfer table, an undirected graph traversal is performed starting from all the addressable nodes in the frozen node set to generate a reachable node set. Determine whether the addressing node corresponding to the reproduced region is included in the set of reachable nodes; When the addressing node corresponding to the reproduced region is included in the reachable node set, the shortest acyclic path between any addressing node in the frozen node set and the addressing node corresponding to the reproduced region is extracted to obtain the node connectivity sequence. The tracking and recovery result is generated based on the arrangement order of each addressed node in the node connectivity sequence.

12. The method according to claim 11, characterized in that, The step of extracting the shortest acyclic path between any addressable node in the frozen node set and the addressable node corresponding to the reconstructed region to obtain a node connectivity sequence includes: When the frozen node set contains addressing nodes belonging to the pipe gallery gap area, and the addressing node corresponding to the reproduced area belongs to the outer ring area of ​​the tower, the cross-domain adjacency edge connecting the corresponding addressing node is extracted from the adjacency transfer table. Mark the two addressing nodes of the cross-domain adjacent edge as nodes that must be passed through the path; When generating the shortest acyclic path, the necessary nodes of the path are included as constraints in the shortest acyclic path.

13. An anti-interference UAV target tracking system, employing the anti-interference UAV target tracking method as described in any one of claims 1 to 12, characterized in that, include: The boundary recognition module is used to identify the physical structure boundary of the industrial park based on the visible light image of the park, construct a topological addressing space based on the physical structure boundary of the park, and generate an adjacency transfer table. The sequence mapping module is used to map the coordinates of UAV hotspots in continuous infrared images to the topology addressing space, generating a topology state sequence; The status response module is used to determine the flight obstruction status based on the adjacency relationship of adjacent addresses in the topology status sequence, generate a set of frozen nodes based on the end address of the topology status sequence in response to the hot spot missing event triggered by the flight obstruction status, and map the UAV hot spot coordinates to the reproduction area in response to the hot spot recovery event. The path verification module is used to verify the transfer path between the frozen node set and the reproducible region in the adjacency transfer table, and output the tracking and recovery results.

Images