A building change detection method and system based on multi-period unmanned aerial vehicle inspection video

Abstract

This invention discloses a method and system for detecting building changes based on multi-period UAV inspection videos. Specifically, it includes: acquiring multi-period UAV inspection videos and recording imaging attitude information to achieve unified management of cross-period inspection data; aligning cross-period time indexes and unifying spatial scale processing to construct a comparable video sequence; extracting building contour points, structural lines, and structural component regions to form the building's structural state; establishing a mapping relationship between structural states in adjacent inspection periods to form cross-period structural state pairs; constructing homotopy parameter intervals and setting target state nodes, performing multi-target structural deformation tracking and recording fracture and bifurcation information; generating free tracking results and constrained tracking results and calculating the differences; locating the cross-period interval where building changes occur and outputting the change detection results. This invention achieves precise location of building changes through cross-period homotopy tracking and counterfactual constraint comparison.

Description

Technical Field

[0001] This invention relates to the field of UAV visual inspection technology, and in particular to a method and system for detecting building changes based on multi-phase UAV inspection videos. Background Technology

[0002] With the development of drone technology and image acquisition technology, using drones for periodic building inspections has become a common technique. By acquiring video data at different inspection cycles, the building's appearance, structural distribution, and the condition of local areas can be recorded, providing a data foundation for subsequent analysis. In practical applications, drone inspections typically acquire building images in the form of video frame sequences. Due to inconsistencies in flight altitude, attitude angle, imaging scale, and time index between different inspection cycles, multiple inspection videos exhibit differences in both time and space.

[0003] In existing technologies, the analysis of multi-period UAV inspection videos mostly focuses on single-period image processing or simple inter-frame comparison, lacking a processing mechanism for systematically correlating video data from different inspection cycles. In multi-period inspection scenarios, buildings often exhibit changes in perspective, scale, and partial occlusion in videos acquired at different times, making it difficult to accurately reflect changes in the building's true condition by directly comparing video frames from different inspection cycles. Furthermore, existing methods typically rely on manual comparison or simple pixel-level difference analysis when determining changes in buildings, making it difficult to distinguish between apparent differences caused by changes in imaging conditions and changes in the building's structure itself.

[0004] Therefore, in multi-stage drone inspection application scenarios, how to effectively align video data acquired in different inspection cycles and establish the correlation between building structural states to achieve reliable detection of building changes remains a technical problem to be solved. Summary of the Invention

[0005] One objective of this invention is to propose a method and system for detecting building changes based on multi-period UAV inspection videos. This invention introduces a cross-period time index alignment and spatial scale unification processing mechanism, and combines homotopy parameter interval construction and multi-target tracking model to perform continuous deformation tracking and difference judgment on the building structural state in multi-period UAV inspection videos, forming a building change detection process with cross-period consistency, accurate change location and structural judgment stability.

[0006] A method and system for detecting building changes based on multi-period UAV inspection videos according to an embodiment of the present invention includes the following steps: Acquire at least two UAV inspection video frame sequences and record the corresponding imaging attitude information; perform cross-period time index alignment and spatial scale unification processing on the multi-period UAV inspection video frame sequences to form cross-period aligned video sequences; locate the target area of ​​the building and extract the building structural state in the cross-period aligned video sequences. The building structural state consists of a set of building outline points, a set of structural lines, and a set of structural component areas; establish a mapping relationship between the building structural states of adjacent inspection periods to form cross-period structural state pairs; construct a homotopy parameter interval based on the cross-period structural state pairs, and set several target point state nodes within the homotopy parameter interval. Perform local structural deformation calculations between adjacent target point state nodes through a multi-target solution method. Record fracture markers and bifurcation markers during the deformation process to generate cross-period state homotopy tracking evidence; generate free tracking results without introducing building geometric consistency constraints and constrained tracking results with introducing building geometric consistency constraints during the cross-period state homotopy tracking process, and calculate the difference between the two types of tracking results; determine the building change judgment marker based on the difference and locate the cross-period interval where the change occurs, and output the building change detection result.

[0007] Optionally, the step of performing cross-period time index alignment and spatial scale unification processing to form a cross-period aligned video sequence specifically includes: Read the video frame sequences of each UAV inspection period, associate the frame number and imaging attitude information of each video frame, and establish a time index reference sequence based on the video frame sequence of the first inspection period. For the video frame sequences of non-reference inspection periods, calculate the time index mapping position of the video frames based on the relationship between the attitude change and the frame number change in the imaging attitude information, and rearrange the video frame sequences of each period according to the time index mapping position to form a cross-period time-aligned video frame sequence. Analyze the imaging height parameter in the imaging attitude information corresponding to each video frame in the cross-period time-aligned video frame sequence, calculate the scale mapping factor of the non-reference inspection period video frames relative to the reference inspection period video frames, and perform spatial scale resampling processing on the cross-period time-aligned video frame sequence according to the scale mapping factor to generate a cross-period aligned video sequence.

[0008] Optionally, the extraction of the building structural state includes the following steps: Read each video frame in the cross-period aligned video sequence, locate the target building area according to the spatial continuity of the video frame, and generate the corresponding video frame building area image; perform boundary parsing processing on the video frame building area image to extract the building contour point set, and perform sequential connection processing on the contour point set to generate the building structure line set; within the area covered by the building structure line set, perform region division processing on the video frame building area image to generate the building structure component area set, and combine the building contour point set, the building structure line set, and the building structure component area set to form the building structure state.

[0009] Optionally, establishing the mapping relationship of building structural status between adjacent inspection periods specifically includes: Read the building structural status corresponding to adjacent inspection periods, and perform point-level indexing on the building outline point sets in the two periods to form an outline point index set; between the building structural status of adjacent inspection periods, perform spatial location matching calculation on the outline point index set to generate a cross-period outline point correspondence set; based on the cross-period outline point correspondence set, perform line segment-level correspondence calculation on the building structural line set to generate a cross-period structural line correspondence set; based on the cross-period structural line correspondence set, perform region-level correspondence calculation on the building structural component region set to generate a cross-period structural component region correspondence set; combine the cross-period outline point correspondence set, the cross-period structural line correspondence set, and the cross-period structural component region correspondence set to form the building structural status mapping relationship between adjacent inspection periods.

[0010] Optionally, constructing the homotopy parameter interval and setting several target point state nodes within the homotopy parameter interval includes the following steps: Read the cross-period structural state pairs corresponding to adjacent inspection periods. Perform structural element counts on the building outline point set, building structural line set, and building structural component region set in the cross-period structural state pairs to generate initial structural state parameter vectors. Perform component-level difference calculations on the initial structural state parameter vectors of the two phases in the cross-period structural state pairs to generate structural change parameter vectors. Construct a continuous parameter axis according to the numerical distribution order of each component in the structural change parameter vector, and set the start and end parameter positions on the continuous parameter axis to form a homotopy parameter interval. Discretize the continuous parameter axis along the homotopy parameter interval with a fixed parameter step size to generate several parameter segment intervals. Read the corresponding parameter position values ​​in each parameter segment interval, and map the parameter position values ​​to the building outline point set, building structural line set, and building structural component region set to generate intermediate structural states. Perform structural integrity verification on each intermediate structural state and filter out intermediate structural states that do not meet the structural continuity conditions. Perform state numbering on intermediate structural states that pass the structural integrity verification to generate a target point state node set. According to the parameter position order of the target state nodes in the homotopy parameter interval, the target state node set is processed to perform sequential arrangement to form the target state node sequence in the homotopy parameter interval.

[0011] Optionally, the step of performing local structural deformation calculations between adjacent target point state nodes using a multi-target solution method, recording fracture markers and bifurcation markers during the deformation process, and generating cross-period state homotopy tracking evidence specifically includes: Read the sequence of target state nodes arranged sequentially within the homotopy parameter interval, and select adjacent target state nodes according to the parameter position order to form target state node pairs; For the starting and ending target state nodes in each target state node pair, read the corresponding building contour point set, building structural line set, and building structural component region set, respectively; Construct a structural corresponding constraint set between the starting and ending target state nodes, which includes contour point corresponding constraints, structural line corresponding constraints, and structural component region corresponding constraints; Under the constraints of the structural corresponding constraint set, perform continuous point-by-point position update calculations on the building contour point set of the starting target state node to generate an intermediate contour point set sequence; Based on the intermediate contour point set sequence, perform continuous segment reconstruction calculations on the building structural line set to generate an intermediate structural line set sequence; Based on the intermediate structural line set sequence, perform continuous region reconstruction calculations on the building structural component region set to generate an intermediate structural component region set sequence. During the generation of the intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence, a break mark is recorded when any structural line breaks during continuous updates; during the generation of the intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence, a bifurcation mark is recorded when any contour point or structural line corresponds to several continuous update paths at the same parameter position; the target point state node is aggregated with the corresponding intermediate contour point set sequence, intermediate structural line set sequence, intermediate structural component region set sequence, and the corresponding break mark and bifurcation mark to form cross-period state homotopy tracking evidence.

[0012] Optionally, generating free-tracking results that do not introduce architectural geometric consistency constraints includes the following steps: Read the target point state node pairs from the intertemporal state homotopy tracing evidence, and extract the associated intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence. Perform parameter-by-parameter independent position update processing on the position of each contour point in the intermediate contour point set sequence to form a free contour point evolution sequence. Based on the free contour point evolution sequence, perform segment-by-segment independent update processing on each structural line in the intermediate structural line set sequence to form a free structural line evolution sequence. Based on the free structural line evolution sequence, perform region-by-region independent update processing on each structural component region in the intermediate structural component region set sequence to form a free structural component region evolution sequence. During the generation of the free contour point evolution sequence, free structural line evolution sequence, and free structural component region evolution sequence, maintain the original recorded state of the fracture markers and bifurcation markers, and do not introduce architectural geometric consistency constraints. Combine the free contour point evolution sequence, free structural line evolution sequence, free structural component region evolution sequence, and the corresponding fracture markers and bifurcation markers to generate the free tracing result.

[0013] Optionally, the generation of constraint tracking results that introduce building geometric consistency restrictions specifically includes: Read the target state node pairs corresponding to the intertemporal state homotopy tracing evidence, and extract the intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence associated with the target state node pairs; introduce contour continuity constraints to the positions of each contour point in the intermediate contour point set sequence, and perform restricted update processing on the contour point positions according to the spatial distance relationship between adjacent contour points to generate a constrained contour point evolution sequence; based on the constrained contour point evolution sequence, introduce line segment connectivity constraints to each structural line in the intermediate structural line set sequence, and perform restricted reconstruction processing on the structural lines according to the connection relationship of the line segment endpoints to generate a constrained structural line evolution sequence; based on the constrained structural line evolution sequence, introduce region closure constraints to each structural component region in the intermediate structural component region set sequence, and perform restricted update processing on the structural component regions according to the region boundary closure relationship to generate a constrained structural component region evolution sequence; during the generation of the constrained contour point evolution sequence, constrained structural line evolution sequence, and constrained structural component region evolution sequence, keep the fracture marker and bifurcation marker synchronously updated; The constraint contour point evolution sequence, constraint structure line evolution sequence, constraint structure component region evolution sequence, and corresponding fracture and bifurcation markers are combined to generate constraint tracking results.

[0014] Optionally, the step of determining the building change judgment mark based on the difference, locating the time interval in which the change occurred, and outputting the building change detection result includes the following steps: Read the free tracking results and constrained tracking results. Align the corresponding contour point evolution sequences, structural line evolution sequences, and structural component region evolution sequences according to the parameter positions within the homotopy parameter interval. Under the alignment condition, calculate the difference in contour point positions, structural line morphology differences, and structural component region boundary differences to generate a structural difference component sequence. Perform component aggregation processing on the structural difference component sequence according to contour point components, structural line components, and structural component region components to form a cross-period structural difference vector sequence. Perform continuous interval division processing on the cross-period structural difference vector sequence along the homotopy parameter interval order to generate several cross-period difference intervals. Perform threshold comparison processing on the cross-period difference intervals to generate building change judgment marks. Map the parameter position range corresponding to the building change judgment marks to the inspection period index position to locate the cross-period interval where the change occurred, and output the building change detection results.

[0015] The beneficial effects of this invention are: (1) By introducing a cross-period time index alignment and spatial scale unification processing mechanism, this invention maps the UAV video frame sequences obtained in different inspection periods to a unified time and space reference frame. Even when there are attitude and scale differences in the inspection data of multiple periods, a stable and consistent data foundation can still be established, providing continuous and comparable video sequence conditions for subsequent building structural status analysis, and avoiding the uncertainty caused by direct comparison between multiple periods of data.

[0016] (2) This invention constructs inter-period structural state pairs and sets target state nodes within the homotopy parameter range. It combines multiple target tracking methods to perform continuous deformation calculations on the structural state of the building, so that the change process of the building between adjacent inspection periods is described in the form of continuous state. At the same time, the structural change path is recorded through fracture markers and bifurcation markers, thereby forming traceable and reproducible inter-period state homotopy tracking evidence.

[0017] (3) In the homotopy tracing process, the present invention constructs free tracing results and constrained tracing results with building geometric consistency restrictions respectively. By calculating and determining the difference between the two types of tracing results, the difference between the actual structural changes of the building and the difference in imaging conditions can be distinguished, so that the change determination can be located to a specific time interval, thereby enhancing the stability and consistency of the building change detection results. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a method and system for detecting building changes based on multi-period UAV inspection videos proposed in this invention; Figure 2 This is a schematic diagram of cross-period state homotopy tracking for a building change detection method and system based on multi-period UAV inspection videos proposed in this invention. Figure 3 This is a schematic diagram illustrating the difference determination between free tracking and constrained tracking in a building change detection method and system based on multi-period UAV inspection videos proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-3 A method and system for detecting building changes based on multi-period UAV inspection videos, comprising the following steps: Acquire at least two UAV inspection video frame sequences and record the corresponding imaging attitude information; perform cross-period time index alignment and spatial scale unification processing on the multi-period UAV inspection video frame sequences to form cross-period aligned video sequences; locate the target area of ​​the building and extract the building structural state in the cross-period aligned video sequences. The building structural state consists of a set of building outline points, a set of structural lines, and a set of structural component areas; establish a mapping relationship between the building structural states of adjacent inspection periods to form cross-period structural state pairs; construct a homotopy parameter interval based on the cross-period structural state pairs, and set several target point state nodes within the homotopy parameter interval. Perform local structural deformation calculations between adjacent target point state nodes through a multi-target solution method. Record fracture markers and bifurcation markers during the deformation process to generate cross-period state homotopy tracking evidence; generate free tracking results without introducing building geometric consistency constraints and constrained tracking results with introducing building geometric consistency constraints during the cross-period state homotopy tracking process, and calculate the difference between the two types of tracking results; determine the building change judgment marker based on the difference and locate the cross-period interval where the change occurs, and output the building change detection result.

[0021] In this embodiment, performing cross-period time index alignment and spatial scale unification processing to form a cross-period aligned video sequence specifically includes: Read the video frame sequences of each UAV inspection period, associate the frame number and imaging attitude information of each video frame, and establish a time index reference sequence based on the video frame sequence of the first inspection period. For the video frame sequences of non-reference inspection periods, calculate the time index mapping position of the video frames based on the relationship between the attitude change and the frame number change in the imaging attitude information, and rearrange the video frame sequences of each period according to the time index mapping position to form a cross-period time-aligned video frame sequence. Analyze the imaging height parameter in the imaging attitude information corresponding to each video frame in the cross-period time-aligned video frame sequence, calculate the scale mapping factor of the non-reference inspection period video frames relative to the reference inspection period video frames, and perform spatial scale resampling processing on the cross-period time-aligned video frame sequence according to the scale mapping factor to generate a cross-period aligned video sequence.

[0022] In this embodiment, extracting the structural state of a building includes the following steps: The process involves reading video frames from a cross-period aligned video sequence, locating the target building region according to the spatial continuity of the video frames, and generating corresponding video frame building region images. Boundary parsing is performed on the video frame building region images to extract building contour point sets. Sequential connection processing is then performed on these contour point sets to generate a set of building structural lines. Sequential connection processing connects discrete contour points one by one into continuous structural lines based on their spatial proximity. Within the area covered by the set of building structural lines, the video frame building region images are divided into regions to generate a set of building structural component regions. Finally, the building contour point sets, building structural line sets, and building structural component region sets are combined to form the building structural state.

[0023] In this embodiment, the boundary resolution process specifically includes: The image of the building area in the video frame is converted to grayscale, transforming the original color pixel values ​​into a single-channel grayscale pixel value matrix.

[0024] Smoothing filtering is performed on the grayscale pixel value matrix, and local smoothing calculations are performed on adjacent pixel values ​​to suppress local noise interference and maintain the pixel distribution characteristics of the main edge regions.

[0025] After smoothing filtering, gradient calculation is performed on the grayscale pixel value matrix to obtain the gradient magnitude information corresponding to each pixel position, forming a pixel gradient magnitude matrix.

[0026] Based on the pixel gradient magnitude matrix, edge candidate determination processing is performed on the pixel positions to filter pixel positions whose gradient magnitudes meet preset conditions and form a set of edge candidate pixels. The edge candidate determination processing determines the pixel positions that may constitute the building boundary by comparing the pixel gradient magnitudes with thresholds and combining them with neighborhood continuity filtering.

[0027] Connectivity analysis is performed on the set of edge candidate pixels to combine spatially continuous edge candidate pixels and generate several boundary pixel chains.

[0028] Perform sequential sorting on the boundary pixel chain and output the building outline point set according to the spatial adjacency order between pixels.

[0029] In this embodiment, the region division process specifically includes: Read the set of building structure lines, determine the spatial position of each structure line in the building area image of the video frame, and use the structure lines as the boundary constraints of the area.

[0030] In the building area image of the video frame, the set of building structure lines is projected onto the pixel space to form several closed or semi-closed boundary areas defined by the structure lines.

[0031] After defining the boundary region, the pixel positions located within the structure line boundary are assigned to the corresponding region number according to the boundary region they are in.

[0032] After determining the region affiliation of all pixel locations, the pixel locations corresponding to the same region number are aggregated to form several spatially continuous pixel regions.

[0033] Each pixel region is output according to its region number to generate a set of building structural component regions.

[0034] In this embodiment, establishing the mapping relationship of building structural status between adjacent inspection periods specifically includes: The system reads the building structural status corresponding to adjacent inspection periods and performs point-level indexing on the building outline point sets in both periods to form an outline point index set. During point-level indexing, a unique index number is assigned to each outline point according to its spatial order within the building outline. A one-to-one correspondence is established between the index number and the spatial coordinates of the corresponding outline point, forming the outline point index set. Between adjacent inspection periods, spatial location matching calculations are performed on the outline point index sets to generate a cross-period outline point correspondence set. During spatial location matching calculations, the coordinates of outline points with the same index number in adjacent inspection periods are read, and the spatial distance between corresponding outline points is calculated. Outline point pairs that meet preset matching conditions are selected based on the spatial distance, forming a cross-period outline point correspondence set. Based on the set of correspondences between cross-period contour points, a segment-level correspondence calculation is performed on the set of building structural lines to generate a set of correspondences between cross-period structural lines. Based on this set, a region-level correspondence calculation is performed on the set of building structural component regions to generate a set of correspondences between cross-period structural component regions. During the segment-level correspondence calculation, the endpoint combinations of structural lines determined by the same set of cross-period contour point correspondences in adjacent inspection periods are read, and corresponding structural line pairs are constructed. A consistency comparison is performed on the endpoint index numbers and endpoint spatial distances of the structural line pairs. Structural line pairs with consistent index numbers and endpoint spatial distances meeting preset conditions are retained, forming a set of correspondences between cross-period structural lines. The set of correspondences between cross-period contour points, the set of correspondences between cross-period structural lines, and the set of correspondences between cross-period structural component regions are combined to form a mapping relationship of building structural status between adjacent inspection periods.

[0035] In this embodiment, constructing a homotopy parameter interval and setting several target point state nodes within the homotopy parameter interval includes the following steps: Read the cross-period structural state pairs corresponding to adjacent inspection periods. Perform structural element counts on the building outline point set, building structural line set, and building structural component region set in the cross-period structural state pairs to generate initial structural state parameter vectors. Perform component-level difference calculations on the initial structural state parameter vectors of the two phases in the cross-period structural state pairs to generate structural change parameter vectors. Construct a continuous parameter axis according to the numerical distribution order of each component in the structural change parameter vector, and set the start and end parameter positions on the continuous parameter axis to form homotopy parameter intervals. Discretize the continuous parameter axis along the homotopy parameter intervals with a fixed parameter step size to generate several parameter segment intervals. Read the corresponding parameter position values ​​in each parameter segment interval, and map the parameter position values ​​to the building outline point set, building structural line set, and building structural component region set to generate intermediate structural states. Perform structural integrity verification on each intermediate structural state, and filter out intermediate structural states that do not meet the structural continuity conditions. Structural integrity is defined by checking whether the building structural lines in the intermediate structural states form a continuous connection relationship and whether the structural component regions meet the closure or connectivity conditions. The intermediate structural states that pass the structural integrity check are processed by state numbering to generate a set of target state nodes. The target state node set is then sorted according to the parameter position order of the target state nodes in the homotopy parameter interval to form a sequence of target state nodes in the homotopy parameter interval.

[0036] In this embodiment, local structural deformation calculations are performed between adjacent target point state nodes using a multi-target solution method. During the deformation process, fracture markers and bifurcation markers are recorded to generate cross-period state homotopy tracking evidence, specifically including: The system reads the sequence of target state nodes arranged sequentially within the homotopy parameter interval, and selects adjacent target state nodes according to their parameter positions to form target state node pairs. For the starting and ending target state nodes in each pair, it reads the corresponding set of building contour points, set of building structural lines, and set of building structural component regions, respectively. A structural correspondence constraint set is constructed between the starting and ending target state nodes, including contour point correspondence constraints, structural line correspondence constraints, and structural component region correspondence constraints. This set defines the corresponding contour points, structural lines, and structural component regions in the starting and ending target state nodes, and updates them synchronously according to the established correspondence during local structural deformation calculations. Under the constraints of the structural correspondence constraint set, the system performs continuous point-by-point position update calculations on the building contour point set of the starting target state node, generating an intermediate contour point set sequence. Under the contour point correspondence defined by the structural correspondence constraint set, the system performs parameterized continuous updates on the coordinates of the building contour points in the starting target state node along the homotopy parameter interval, generating intermediate contour point coordinate sets corresponding to different parameter positions. The coordinates of intermediate contour points corresponding to adjacent parameter positions are sequentially associated to form a sequence of intermediate contour point sets.

[0037] Based on the intermediate contour point set sequence, continuous segment reconstruction calculations are performed on the building structural line set to generate an intermediate structural line set sequence. Based on the correspondence between each contour point in the intermediate contour point set sequence at adjacent parameter positions, the endpoint coordinates of each building structural line are synchronously updated according to the structural line endpoint index relationship, generating intermediate structural lines at corresponding parameter positions. Between adjacent parameter positions, continuous structural lines are generated by connecting them in the order of endpoint coordinates, and then arranged sequentially along the homotopy parameter interval to form an intermediate structural line set sequence. Based on the intermediate structural line set sequence, continuous region reconstruction calculations are performed on the building structural component region set to generate an intermediate structural component region set sequence. Based on the closed boundary relationship formed by each structural line at adjacent parameter positions in the intermediate structural line set sequence, the boundary coordinates of the regions enclosed by the corresponding structural lines are synchronously updated, generating intermediate structural component regions at corresponding parameter positions. The intermediate structural component regions generated at each parameter position are then continuously arranged along the homotopy parameter interval to form an intermediate structural component region set sequence. During the generation of the intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence, a break mark is recorded when any structural line breaks during continuous updates; during the generation of the intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence, a bifurcation mark is recorded when any contour point or structural line corresponds to several continuous update paths at the same parameter position; the target point state node is aggregated with the corresponding intermediate contour point set sequence, intermediate structural line set sequence, intermediate structural component region set sequence, and the corresponding break mark and bifurcation mark to form cross-period state homotopy tracking evidence.

[0038] In this embodiment, generating free-tracking results without introducing architectural geometric consistency constraints includes the following steps: Read the target state node pairs corresponding to the homotopy tracing evidence of the intertemporal state, and extract the intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence associated with the target state node pairs; perform parameter-by-parameter independent position update processing on the position of each contour point in the intermediate contour point set sequence to form a free contour point evolution sequence; read the intermediate contour point set sequence sequentially along the homotopy parameter interval, and read the spatial coordinate values ​​of the corresponding contour point at each parameter position; at the current parameter position, apply the displacement increment corresponding to the parameter position to the spatial coordinate value of the contour point, and the displacement increment is obtained by mapping the parameter step size between the current parameter position and the adjacent parameter positions; perform a component-by-component addition update operation on the spatial coordinate components of the contour point according to the displacement increment to generate the updated contour point coordinates corresponding to the parameter position; during the position update process, do not introduce spatial constraint relationships between adjacent contour points, nor introduce geometric constraints of structural lines or structural component regions; record the updated coordinates generated by the same contour point at different parameter positions to form a contour point position sequence arranged by parameter position; repeat the above position update processing for all contour points in the contour point set, and combine them to form a free contour point evolution sequence.

[0039] Based on the free contour point evolution sequence, each structural line in the intermediate structural line set sequence is updated independently segment by segment to form a free structural line evolution sequence. The free contour point evolution sequence is read sequentially along the homotopy parameter interval. At each parameter position, the updated coordinates of the two end contour points constituting the same structural line are read, and the structural lines at the corresponding parameter positions are reconstructed according to the endpoint connection relationship. The reconstruction results of each structural line at different parameter positions are recorded independently, without introducing connection constraints between adjacent structural lines, forming a free structural line evolution sequence arranged by parameter position. Based on the free structural line evolution sequence, each structural component region in the intermediate structural component region set sequence is updated independently region by region to form a free structural component region evolution sequence. The free structural line evolution sequence is read sequentially along the homotopy parameter interval. At each parameter position, based on the update results of the structural lines enclosing the structural component region, the boundary contour of the corresponding region is recalculated, and the updated region morphology is generated. The region morphology of each structural component region at different parameter positions is recorded independently, without introducing boundary consistency or topological constraints between adjacent regions, forming a free structural component region evolution sequence.

[0040] During the generation of the free contour point evolution sequence, the free structural line evolution sequence, and the free structural component region evolution sequence, the fracture markers and bifurcation markers are kept in their original recorded state, and no architectural geometric consistency constraints are introduced. The free contour point evolution sequence, the free structural line evolution sequence, the free structural component region evolution sequence, and the corresponding fracture markers and bifurcation markers are combined to generate free tracking results.

[0041] In this embodiment, the characteristic feature is that generating constraint tracking results that introduce building geometric consistency restrictions specifically includes: Read the target point state node pairs corresponding to the intertemporal state homotopy tracing evidence, and extract the intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence associated with the target point state node pairs; introduce contour continuity constraints on the positions of each contour point in the intermediate contour point set sequence, and perform constrained update processing on the contour point positions according to the spatial distance relationship between adjacent contour points to generate a constrained contour point evolution sequence; during the generation of the intermediate contour point set sequence, read the corresponding contour point set for each parameter position, and establish an index relationship set of adjacent contour point pairs according to the order relationship of contour points in the building contour; calculate the spatial distance value between adjacent contour points based on the index relationship set of adjacent contour point pairs, and write the spatial distance value into the contour continuity constraint parameter set; before performing position update on a single contour point, read the previous one corresponding to that contour point. The spatial distance constraint parameter between a contour point and the next contour point is used to perform a distance consistency check on the candidate update positions of the contour points. When the candidate update position causes the spatial distance between the candidate update position and the adjacent contour point to deviate from the contour continuity constraint parameter, a position correction operation is performed on the candidate update position along the contour tangential direction to limit the updateable range of the contour point under the current parameter position. The positions of the contour points that pass the distance consistency check and have completed the position correction are written into the contour point position buffer to form the constrained contour point set under the current parameter position. The above-mentioned restricted update process is repeated for each parameter position in sequence along the homotopy parameter interval to generate a constrained contour point evolution sequence containing contour continuity constraints. Based on the constraint contour point evolution sequence, line segment connectivity constraints are introduced for each structural line in the intermediate structural line set sequence. Constrained reconstruction processing is performed on the structural lines according to the connection relationships of the line segment endpoints, generating a constraint structural line evolution sequence. At each parameter position, the coordinates of the corresponding contour point in the constraint contour point evolution sequence are read. The contour point indices corresponding to the two endpoints of each structural line are locked, and the endpoints of the structural lines are restricted to be updated only within the local neighborhood formed by the corresponding contour point position or the line connecting adjacent contour points. Within this restricted neighborhood, the position of the endpoint coordinates of the structural lines is recalculated, and after the update, it is verified whether the structural lines maintain the endpoint connection relationship and the line segment continuity relationship. Structural lines that meet the constraints are written into the structural line state buffer, forming a constraint structural line evolution sequence sequentially along the homotopy parameter interval. Based on the constraint structural line evolution sequence, region closure constraints are introduced for each structural component region in the intermediate structural component region set sequence. Constrained update processing is performed on the structural component regions according to the region boundary closure relationship, generating a constraint structural component region evolution sequence. The region closure constraint is a state constraint on the boundary of the structural component region formed by connecting the beginning and end of the structural lines to form a closed contour, ensuring that the region boundary always maintains a closed connection relationship during the update process.During the generation of the constraint contour point evolution sequence, constraint structure line evolution sequence, and constraint structure component region evolution sequence, the fracture markers and bifurcation markers are updated synchronously. The constraint contour point evolution sequence, constraint structure line evolution sequence, constraint structure component region evolution sequence, and the corresponding fracture markers and bifurcation markers are combined to generate the constraint tracking result.

[0042] In this embodiment, determining the building change judgment marker based on the differences and locating the time interval where the change occurred, and outputting the building change detection results includes the following steps: The system reads both free-tracking and constrained-tracking results. It aligns the corresponding contour point evolution sequences, structural line evolution sequences, and structural component region evolution sequences according to the parameter positions within the homotopy parameter interval. Under the alignment condition, it calculates the differences in contour point positions, structural line morphology, and structural component region boundaries to generate a structural difference component sequence. It then performs component aggregation on the structural difference component sequence according to contour point components, structural line components, and structural component region components to form a cross-period structural difference vector sequence. Finally, it performs continuous interval division along the homotopy parameter interval to generate several cross-period difference intervals. At the same parameter position, it reads the differences in contour point positions, structural line morphology, and structural component region boundaries, and groups and aggregates the difference data according to preset component categories. Finally, it performs numerical merging operations on the difference data within each component category to generate contour point components, structural line components, and structural component region components at the corresponding parameter positions, thus forming a structural difference component sequence. Threshold comparison processing is performed on the inter-period difference intervals to generate building change judgment marks. The parameter position range corresponding to the building change judgment mark is mapped to the inspection period index position to locate the inter-period interval where the change occurred, and the building change detection result is output. Within each inter-period difference interval, the corresponding inter-period structural difference vector value is read and compared with a pre-set change judgment threshold. When the structural difference vector value within the interval meets the threshold comparison condition, a building change judgment mark is generated within the parameter position range corresponding to that interval, and this parameter position range is mapped to the inspection period index interval to determine the location of the change.

[0043] Example 1: To verify the feasibility of this invention in practical applications, it was applied to a building change detection scenario under multi-period UAV inspection video conditions. In this scenario, UAVs inspect the same building area multiple times at different times, acquiring a continuous sequence of inspection video frames. Due to different inspection cycles, changes in flight attitude, and inconsistent shooting altitudes, directly comparing frames from multiple periods of video can easily lead to structural misalignment, scale deviation, and misjudgment, especially when minor changes occur in the local structure of the building. Traditional methods based on pixel difference or simple feature matching are difficult to reliably identify the true changes. This invention addresses these problems by using cross-period time index alignment, spatial scale unification, and structural homotopy tracking to continuously model and determine building structural changes. In specific applications, video frame sequences are first extracted from multiple UAV inspection videos, and the imaging attitude information corresponding to each frame is recorded simultaneously. By establishing a unified time index reference, the video frames from different inspection periods are indexed and rearranged to ensure that semantically corresponding images in different periods remain consistent on the timeline. Subsequently, the video frames are resampled at spatial scale based on imaging height parameters to eliminate the proportional shift caused by differences in flight altitude across different inspection periods, forming a directly comparable cross-period aligned video sequence. Within this sequence, the target building area is located, and building outline point sets, structural line sets, and structural component area sets are extracted sequentially from the building area images. These three types of structural elements together constitute a description of the building's structural state. By performing point-level, line-level, and region-level correspondence calculations on the building's structural state between adjacent inspection periods, a stable cross-period structural state mapping relationship is established, providing a consistent data foundation for subsequent structural evolution analysis. Based on this, the system constructs a continuous homotopy parameter interval according to the structural change parameters between adjacent inspection periods, and sets several target point state nodes along this interval. Through a multi-target solution method, local structural deformation calculations are performed between adjacent target point state nodes to track the continuous evolution of the building outline, structural lines, and structural component areas. During the structural evolution process, the system automatically records the structural fracture location and bifurcation markers generated by multi-path evolution, forming complete cross-period state homotopy tracking evidence. To enhance the reliability of change determination, this invention simultaneously generates free tracking results without introducing building geometric consistency constraints, and constrained tracking results with contour continuity, line segment connectivity, and region closure constraints during homotopy tracking. By calculating the differences between the two types of tracking results in the homotopy parameter interval, it can effectively distinguish between evolutionary differences caused by actual structural changes and non-structural offsets caused by noise or local disturbances. In the change determination stage, the system performs interval processing on the distribution of structural differences in the parameter interval and generates building change determination markers based on preset thresholds. Finally, the determination results are mapped back to specific inspection period intervals, achieving accurate location of building changes.In practical applications, this invention demonstrates its ability to reliably identify various changes in buildings, including additions, demolitions, and structural adjustments, even under multi-stage inspection video conditions, maintaining high detection consistency despite complex backgrounds and scale variations. Compared to traditional methods, this invention avoids the limitations of relying on fixed windows or single-feature comparisons, exhibiting significant advantages in continuity, stability, and change localization accuracy.

[0044] Table 1: Comparison of Building Intertemporal Change Detection Results

[0045] As can be seen from the comparison data of the building's inter-period change detection results listed in Table 1, there are significant differences in the structural change characteristics between different inspection periods, and the change trend is consistent with the actual structural evolution of the building. In the comparison between Phase I and Phase II, the number of contour points changed by 2, the number of structural lines changed by 1, and the number of structural component areas remained unchanged. The corresponding maximum structural difference value was 0.18, which is lower than the change judgment threshold. Therefore, the system did not generate a change judgment mark. This result indicates that there was only a slight structural disturbance in this stage, which did not constitute an identifiable structural change.

[0046] In the comparison between phases two and three, the number of contour points increased to 5, the number of structural lines changed to 3, and one structural component area changed. The corresponding maximum structural difference value reached 0.46, which was significantly higher than the previous stage. Based on this, the system generated a change judgment mark, and the accuracy of the change interval positioning reached 0.95, indicating that the building structure underwent a relatively significant adjustment in this stage, and the present invention can accurately capture this change interval.

[0047] The comparison results from Phase III to Phase IV show that the number of changes in contour points, structural lines, and structural component areas are all at a low level, with the maximum structural difference value being only 0.12. The system did not generate a change judgment mark. This result indicates that the overall building structure remained stable during this stage, and the data fluctuations mainly came from local disturbances caused by differences in inspection conditions.

[0048] In the comparison between phases four and five, the number of contour points changed by 7, structural lines by 4, and structural component regions by 2. The maximum structural difference value rose to 0.63, one of the highest values ​​in the table. The system generated a change judgment mark, and the accuracy of the change interval positioning reached 0.96, reflecting significant structural changes in this stage. This invention effectively distinguishes between real structural changes and non-structural offsets through homotopy tracking and constraint comparison mechanisms.

[0049] The data from each inspection period in the table show that when the number of structural changes and the value of structural differences show a synchronous upward trend, the system can stably generate change judgment marks and maintain a high accuracy in locating the change range. This verifies the reliability and consistency of the present invention in detecting structural changes in buildings under multi-period UAV inspection conditions.

[0050] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method and system for detecting building changes based on multi-period UAV inspection videos, characterized in that, Includes the following steps: Acquire at least two video frame sequences of UAV inspections and record the corresponding imaging attitude information; Perform cross-period time index alignment and spatial scale unification processing on multi-period drone inspection video frame sequences to form cross-period aligned video sequences; In the cross-period aligned video sequence, the target area of ​​the building is located and the structural state of the building is extracted. The structural state of the building consists of a set of building outline points, a set of structural lines, and a set of structural component areas. A mapping relationship of the structural state of the building between adjacent inspection periods is established to form cross-period structural state pairs. Based on the cross-period structural state pairs, a homotopy parameter interval is constructed, and several target point state nodes are set within the homotopy parameter interval. Local structural deformation calculations are performed between adjacent target point state nodes through a multi-target solution method. During the deformation process, fracture markers and bifurcation markers are recorded to generate cross-period state homotopy tracking evidence. In the process of intertemporal state homotopy tracing, free tracing results without introducing building geometric consistency constraints and constrained tracing results with introducing building geometric consistency constraints are generated respectively, and the difference between the two types of tracing results is calculated. Based on the differences, determine the building change judgment mark and locate the time interval in which the change occurred, and output the building change detection result.

2. The method and system for detecting building changes based on multi-period UAV inspection videos according to claim 1, characterized in that, The process of performing cross-period time index alignment and spatial scale unification to form a cross-period aligned video sequence specifically includes: Read the video frame sequences of each UAV inspection period, associate the frame number and imaging attitude information of each video frame, and establish a time index reference sequence based on the video frame sequence of the first inspection period. For the video frame sequences of non-reference inspection periods, calculate the time index mapping position of the video frames based on the relationship between the attitude change and the frame number change in the imaging attitude information, and rearrange the video frame sequences of each period according to the time index mapping position to form a cross-period time-aligned video frame sequence. Analyze the imaging height parameter in the imaging attitude information corresponding to each video frame in the cross-period time-aligned video frame sequence, calculate the scale mapping factor of the non-reference inspection period video frames relative to the reference inspection period video frames, and perform spatial scale resampling processing on the cross-period time-aligned video frame sequence according to the scale mapping factor to generate a cross-period aligned video sequence.

3. The method and system for detecting building changes based on multi-period UAV inspection videos according to claim 2, characterized in that, The extraction of building structural status includes the following steps: Read each video frame in the cross-period aligned video sequence, locate the target building area according to the spatial continuity of the video frame, and generate the corresponding video frame building area image; perform boundary parsing processing on the video frame building area image to extract the building contour point set, and perform sequential connection processing on the contour point set to generate the building structure line set; within the area covered by the building structure line set, perform region division processing on the video frame building area image to generate the building structure component area set, and combine the building contour point set, the building structure line set, and the building structure component area set to form the building structure state.

4. The method and system for detecting building changes based on multi-period UAV inspection videos according to claim 3, characterized in that, The establishment of the building structural status mapping relationship between adjacent inspection periods specifically includes: The system reads the building structural status corresponding to adjacent inspection periods, performs point-level indexing on the building outline point sets in the two periods, forming an outline point index set; between the building structural statuses of adjacent inspection periods, it performs spatial location matching calculations on the outline point index sets to generate a cross-period outline point correspondence set; based on the cross-period outline point correspondence set, it performs line segment-level correspondence calculations on the building structural line sets to generate a cross-period structural line correspondence set; based on the cross-period structural line correspondence set, it performs region-level correspondence calculations on the building structural component region sets to generate a cross-period structural component region correspondence set; and combines the cross-period outline point correspondence set, the cross-period structural line correspondence set, and the cross-period structural component region correspondence set to form a building structural status mapping relationship between adjacent inspection periods.

5. The method and system for detecting building changes based on multi-period UAV inspection video as described in claim 4, characterized in that, The process of constructing a homotopy parameter interval and setting several target point state nodes within the homotopy parameter interval includes the following steps: Read the cross-period structural state pairs corresponding to adjacent inspection periods. Perform structural element counts on the building outline point set, building structural line set, and building structural component region set in the cross-period structural state pairs to generate initial structural state parameter vectors. Perform component-level difference calculations on the initial structural state parameter vectors of the two phases in the cross-period structural state pairs to generate structural change parameter vectors. Construct a continuous parameter axis according to the numerical distribution order of each component in the structural change parameter vector, and set the start and end parameter positions on the continuous parameter axis to form a homotopy parameter interval. Discretize the continuous parameter axis along the homotopy parameter interval with a fixed parameter step size to generate several parameter segment intervals. Read the corresponding parameter position values ​​in each parameter segment interval, and map the parameter position values ​​to the building outline point set, building structural line set, and building structural component region set to generate intermediate structural states. Perform structural integrity verification on each intermediate structural state and filter out intermediate structural states that do not meet the structural continuity conditions. Perform state numbering on the intermediate structural states that pass the structural integrity verification to generate a target point state node set. According to the parameter position order of the target state nodes in the homotopy parameter interval, the target state node set is processed to perform sequential arrangement to form the target state node sequence in the homotopy parameter interval.

6. The method and system for detecting building changes based on multi-period UAV inspection video as described in claim 5, characterized in that, The process of performing local structural deformation calculations between adjacent target point state nodes using a multi-target solution method, recording fracture and bifurcation markers during deformation, and generating cross-period state homotopy tracking evidence specifically includes: Read the sequence of target state nodes arranged sequentially within the homotopy parameter interval, and select adjacent target state nodes according to the parameter position order to form target state node pairs; For the starting and ending target state nodes in each target state node pair, read the corresponding building contour point set, building structural line set, and building structural component region set, respectively; Construct a structural corresponding constraint set between the starting and ending target state nodes, which includes contour point corresponding constraints, structural line corresponding constraints, and structural component region corresponding constraints; Under the constraints of the structural corresponding constraint set, perform continuous point-by-point position update calculations on the building contour point set of the starting target state node to generate an intermediate contour point set sequence; Based on the intermediate contour point set sequence, perform continuous segment reconstruction calculations on the building structural line set to generate an intermediate structural line set sequence; Based on the intermediate structural line set sequence, perform continuous region reconstruction calculations on the building structural component region set to generate an intermediate structural component region set sequence. During the generation of the intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence, a break mark is recorded when any structural line breaks during continuous updates; during the generation of the intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence, a bifurcation mark is recorded when any contour point or structural line corresponds to several continuous update paths at the same parameter position; the target point state node is aggregated with the corresponding intermediate contour point set sequence, intermediate structural line set sequence, intermediate structural component region set sequence, and the corresponding break mark and bifurcation mark to form cross-period state homotopy tracking evidence.

7. The method and system for detecting building changes based on multi-period UAV inspection videos according to claim 6, characterized in that, The process of generating free-tracking results without introducing architectural geometric consistency constraints includes the following steps: Read the target point state node pairs from the intertemporal state homotopy tracing evidence, and extract the associated intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence. Perform parameter-by-parameter independent position update processing on the position of each contour point in the intermediate contour point set sequence to form a free contour point evolution sequence. Based on the free contour point evolution sequence, perform segment-by-segment independent update processing on each structural line in the intermediate structural line set sequence to form a free structural line evolution sequence. Based on the free structural line evolution sequence, perform region-by-region independent update processing on each structural component region in the intermediate structural component region set sequence to form a free structural component region evolution sequence. During the generation of the free contour point evolution sequence, free structural line evolution sequence, and free structural component region evolution sequence, maintain the original recorded state of the fracture markers and bifurcation markers, and do not introduce architectural geometric consistency constraints. Combine the free contour point evolution sequence, free structural line evolution sequence, free structural component region evolution sequence, and the corresponding fracture markers and bifurcation markers to generate the free tracing result.

8. The method and system for detecting building changes based on multi-period UAV inspection videos according to claim 7, characterized in that, The specific results of generating constraint tracking that introduce building geometric consistency restrictions include: Read the target state node pairs corresponding to the intertemporal state homotopy tracing evidence, and extract the intermediate contour point set sequence, intermediate structural line set sequence, and intermediate structural component region set sequence associated with the target state node pairs; introduce contour continuity constraints to the positions of each contour point in the intermediate contour point set sequence, and perform restricted update processing on the contour point positions according to the spatial distance relationship between adjacent contour points to generate a constrained contour point evolution sequence; based on the constrained contour point evolution sequence, introduce line segment connectivity constraints to each structural line in the intermediate structural line set sequence, and perform restricted reconstruction processing on the structural lines according to the connection relationship of the line segment endpoints to generate a constrained structural line evolution sequence; based on the constrained structural line evolution sequence, introduce region closure constraints to each structural component region in the intermediate structural component region set sequence, and perform restricted update processing on the structural component regions according to the region boundary closure relationship to generate a constrained structural component region evolution sequence; during the generation of the constrained contour point evolution sequence, constrained structural line evolution sequence, and constrained structural component region evolution sequence, keep the fracture marker and bifurcation marker synchronously updated; The constraint contour point evolution sequence, constraint structure line evolution sequence, constraint structure component region evolution sequence, and corresponding fracture and bifurcation markers are combined to generate constraint tracking results.

9. A method and system for detecting building changes based on multi-period UAV inspection video as described in claim 8, characterized in that, The process of determining building change detection markers based on differences, locating the time intervals in which changes occur, and outputting building change detection results includes the following steps: Read the free tracking results and constrained tracking results. Align the corresponding contour point evolution sequences, structural line evolution sequences, and structural component region evolution sequences according to the parameter positions within the homotopy parameter interval. Under the alignment condition, calculate the difference in contour point positions, structural line morphology differences, and structural component region boundary differences to generate a structural difference component sequence. Perform component aggregation processing on the structural difference component sequence according to contour point components, structural line components, and structural component region components to form a cross-period structural difference vector sequence. Perform continuous interval division processing on the cross-period structural difference vector sequence along the homotopy parameter interval order to generate several cross-period difference intervals. Perform threshold comparison processing on the cross-period difference intervals to generate building change judgment marks. Map the parameter position range corresponding to the building change judgment marks to the inspection period index position to locate the cross-period interval where the change occurred, and output the building change detection results.

Images