A method and system for batch automatic matching and correction of panoramic photo data

Abstract

This invention provides a batch automatic matching and correction method and system for panoramic photo data, relating to the field of image processing technology. The method includes: acquiring batch panoramic photo data and preprocessing it to obtain an overlapping region set and a standardized image set; obtaining a fine point matching set and a non-rigid mapping relationship through point feature extraction and non-rigid transformation fitting; calculating the vertical structure saliency index by combining the spatial distribution direction, gradient principal direction, and local edge structure information of fine points in each candidate overlapping region; determining whether the vertical structure saliency index is greater than a threshold, and if so, marking it as a target candidate overlapping region and performing vertical line enhancement matching to obtain a line matching set; otherwise, setting the line matching set to empty; generating an enhanced homonymous point set and a matching weight set based on the fine point matching set, the line matching set, and the non-rigid mapping relationship; solving for geometric correction parameters based on the enhanced homonymous point set and the matching weight set, and outputting the corrected panoramic photo dataset.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for batch automatic matching and correction of panoramic photo data. Background Technology

[0002] With the development of panoramic imaging equipment, mobile acquisition terminals, and multi-view image stitching technology, panoramic photos have been widely used in fields such as virtual display, map surveying, security inspection, architectural space modeling, and intelligent driving environment perception due to their ability to completely record scene spatial information, large field of view, and intuitive expression.

[0003] In existing technologies, the matching and correction of panoramic photos or multiple images typically involves preprocessing the images first, then extracting feature points based on overlapping areas between images, and completing initial matching through feature descriptor similarity. Subsequently, affine transformation, projection transformation, or other geometric models are combined to achieve image registration and correction. For some complex scenes, some technical solutions introduce non-rigid transformation models to enhance adaptability to local deformations. Overall, existing technologies can already achieve a certain degree of alignment and geometric correction of overlapping areas between panoramic photos, possessing basic automatic processing capabilities.

[0004] However, most existing methods still rely on point feature matching. In areas with weak texture, repetitive texture, or local non-rigid deformation, they are prone to insufficient matching points, uneven spatial distribution, and high mismatch rates, leading to unstable foundations for subsequent geometric correction and making it difficult to obtain reliable overall registration results. At the same time, existing technologies usually lack adaptive enhancement matching mechanisms for areas with significant vertical structures, making it difficult to effectively utilize vertical structural features such as columns, window frames, and wall boundaries to form stronger geometric constraints. This results in structural misalignment, boundary cracking, and poor correction continuity in areas such as building edges and facade outlines. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method and system for batch automatic matching and correction of panoramic photo data. This method can solve the problems that most existing methods still rely on point feature matching, which can easily lead to insufficient matching points, uneven spatial distribution, and high mismatch rates in weak texture areas, repetitive texture areas, and scenarios with local non-rigid deformation. This results in an unstable foundation for subsequent geometric correction and makes it difficult to obtain reliable overall registration results. At the same time, existing technologies usually lack adaptive enhancement matching mechanisms for areas with significant vertical structures, making it difficult to effectively utilize vertical structural features such as columns, window frames, and wall boundaries to form stronger geometric constraints. This can lead to structural misalignment, boundary cracking, and poor correction continuity in areas such as building edges and facade outlines.

[0006] A first aspect of this invention provides a method for batch automatic matching and correction of panoramic photo data, comprising: S1: Obtain batch panoramic photo data to be processed; S2: Preprocess the batch of panoramic photo data to determine a set of candidate overlapping regions containing multiple candidate overlapping regions and a standardized image set; S3: Based on the candidate overlapping region set and the standardized image set, determine the fine point matching set and non-rigid mapping relationship through point feature extraction and non-rigid transformation fitting; S4: Calculate the vertical structure saliency index of each candidate overlapping region based on the spatial distribution direction, gradient principal direction, and local edge structure information of the fine points in the fine point matching set in each candidate overlapping region; S5: Determine whether the vertical structure saliency index of each candidate overlapping region is greater than the threshold; if so, mark the candidate overlapping region as the target candidate overlapping region and proceed to S6; otherwise, set the line matching set corresponding to the candidate overlapping region to an empty set and proceed to S7. S6: Based on the target candidate overlapping region and the corresponding standardized image set, obtain the line matching set of the target candidate overlapping region through vertical line enhancement matching, and proceed to S7; S7: Generate an enhanced set of points with the same name and a corresponding set of matching weights based on the fine point matching set, the line matching set, and the non-rigid mapping relationship; S8: Solve for the geometric correction parameters based on the enhanced set of corresponding points and the matching weight set, and output the corrected panoramic photo dataset.

[0007] A second aspect of this invention provides a batch automatic matching and correction system for panoramic photo data, comprising: a processor and a memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the batch automatic matching and correction method for panoramic photo data as described in the first aspect.

[0008] A third aspect of the present invention provides a readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the panoramic photo data batch automatic matching and correction method as described in the first aspect.

[0009] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: In this embodiment of the invention, by constructing candidate overlapping regions and a standardized image set, a fine-grained point feature matching and non-rigid transformation fitting mechanism is introduced to effectively improve the matching quantity and stability in weak texture, repetitive texture, and local non-rigid deformation scenarios. Furthermore, by fusing point matching and line matching to generate an enhanced set of corresponding points and matching weights, the problem of uneven spatial distribution of matching points and mismatches is significantly improved. Simultaneously, a vertical structure saliency index is constructed based on the spatial distribution direction, gradient principal direction, and local edge structure information of fine-grained point matching. This adaptively filters regions with significant vertical structure features and introduces a vertical line enhanced matching mechanism to strengthen structural constraints such as columns, window frames, and wall boundaries. This improves the overall stability and continuity of geometric correction, avoids structural misalignment and boundary tearing problems at building edges and facades, and significantly enhances the accuracy and robustness of panoramic image registration results. Attached Figure Description

[0010] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0011] Figure 1 This is a flowchart illustrating a method for batch automatic matching and correction of panoramic photo data provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of a batch automatic matching and correction system for panoramic photo data provided in an embodiment of the present invention. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0014] The following description, in conjunction with the accompanying drawings, details the batch automatic matching and correction method for panoramic photo data provided by the present invention through specific embodiments and application scenarios.

[0015] Reference manual attached Figure 1 The diagram illustrates a flowchart of a batch automatic matching and correction method for panoramic photo data provided by an embodiment of the present invention.

[0016] This invention provides a method for batch automatic matching and correction of panoramic photo data, which may include the following steps: S1: Obtain batch panoramic photo data to be processed.

[0017] It should be noted that batch panoramic image data can originate from multiple panoramic images captured consecutively under the same acquisition task, or from a collection of panoramic images of the same scene acquired at different time periods and from different perspectives. Batch panoramic image data is generally stored as image files and may also include basic supplementary information corresponding to each panoramic image, such as image resolution, encoding format, capture time, device identification, and possibly pose recording information. The panoramic image data has a 1:2 aspect ratio.

[0018] S2: Preprocess the batch of panoramic photo data to determine the candidate overlapping region set and the normalized image set, which contain multiple candidate overlapping regions.

[0019] In one possible implementation, S2 specifically includes: S201: Decode and unify the channels of the batch of panoramic photo data to obtain the original image set.

[0020] Specifically, the encoded files corresponding to each panoramic photo in the batch panoramic photo data are read, and each encoded file is decoded to obtain the corresponding pixel matrix. Then, each panoramic photo is uniformly converted into a preset channel format, preferably a three-channel RGB format, to obtain the original image set.

[0021] S202: Perform brightness normalization and noise suppression on the original image set to obtain a normalized image set.

[0022] Optionally, the luminance component is first extracted from each image in the original image set, and then the luminance component is normalized to reduce the overall luminance shift caused by differences in exposure and ambient light between different photos. Then, noise suppression processing is performed on the normalized image to reduce the impact of random noise on the subsequent extraction of candidate overlapping regions.

[0023] Furthermore, the normalized image is Gaussian smoothed to obtain a preprocessed image. Gaussian smoothing is an existing technology, and will not be described in detail here.

[0024] S203: Perform resolution unification and coordinate standardization on the normalized image set to obtain a standardized image set.

[0025] Optionally, all images in the normalized image set are uniformly adjusted to the target resolution to eliminate the impact of different image resolutions on subsequent related matching calculations. After resolution unification, the image coordinates are standardized to map each image to a unified coordinate reference.

[0026] S204: Calculate the candidate offsets for the standardized image set to obtain the candidate offset set.

[0027] Optionally, overlap correlation response calculation is performed on any two standardized images to determine candidate relative displacements in the horizontal and vertical directions. Since panoramic images typically have circular continuity in the horizontal direction, a cyclic displacement approach is preferably used for correlation matching in the horizontal direction.

[0028] S205: Based on the candidate offset set, determine the candidate overlapping region set containing multiple candidate overlapping regions.

[0029] Specifically, based on the candidate offsets in the candidate offset set, one standardized image is mapped to the coordinate reference system of another standardized image, and the effective overlapping region of the two images under the candidate offsets is determined. Then, regions with an overlapping area lower than a preset overlapping area are removed, and finally, a candidate overlapping region set is generated.

[0030] It should be noted that the above preprocessing steps can unify batches of panoramic photos from different sources and with inconsistent formats into a standardized image set with consistent structure and stable quality. This effectively reduces the impact of brightness differences, noise interference, and resolution inconsistencies on subsequent matching calculations. Simultaneously, by calculating candidate offsets and extracting candidate overlapping regions, the search range for subsequent feature matching can be quickly narrowed based on the circular structure characteristics of panoramic images. This avoids high-overhead global matching operations on the entire image, thus significantly improving computational efficiency while ensuring matching accuracy.

[0031] Optionally, after S2, the following can be included: After obtaining the candidate overlapping region set and the standardized image set, each candidate overlapping region is divided into multiple initial overlapping sub-blocks according to its region size, texture distribution, and edge response intensity. The texture richness and edge stability of each initial overlapping sub-block are calculated. When the texture richness and edge stability of an initial overlapping sub-block are both below a preset texture threshold, it is marked as a weakly matched sub-block and its subsequent matching priority is reduced. When the texture richness or edge stability of an initial overlapping sub-block meets the corresponding threshold requirements, it is determined as a valid overlapping sub-block. Furthermore, local brightness equalization, contrast enhancement, and boundary redundancy expansion are performed on the valid overlapping sub-blocks to maintain a preset width of overlapping boundary between adjacent overlapping sub-blocks. An index mapping relationship is established between each valid overlapping sub-block and its corresponding candidate overlapping region, ultimately generating a block-based overlapping region set. The segmented overlapping region set serves as a local refinement result of the candidate overlapping region set. It is used to limit the local processing range of feature detection, descriptor construction, and corresponding point matching in the subsequent point feature extraction process, so that the subsequent steps are still based on the candidate overlapping region set for overall processing, and local matching is performed separately in each effective overlapping sub-block.

[0032] S3: Based on the candidate overlapping region set and the standardized image set, determine the fine point matching set and non-rigid mapping relationship through point feature extraction and non-rigid transformation fitting.

[0033] Point feature extraction refers to detecting feature points with local saliency and discriminability within candidate overlapping regions, and further constructing descriptive information for each feature point that can characterize its neighborhood texture, gradient direction, and local structural features, thereby achieving the search for initial corresponding points between different images; non-rigid transformation fitting, after obtaining relatively reliable corresponding point pairs, uses these corresponding points as control constraints to establish a coordinate mapping model that can describe the effects of local deformation, viewpoint changes, and projection distortion, so that points in one image can be accurately mapped to corresponding positions in another image.

[0034] In one possible implementation, S3 specifically includes: S301: Based on the candidate overlapping region set and the normalized image set, construct a multi-scale feature response map within the candidate overlapping region to obtain the initial candidate feature point set.

[0035] Specifically, multi-scale convolution is performed on the standardized image within any candidate overlapping region to calculate the Hessian response value of each pixel position at different scales, and local extreme points are selected as initial candidate feature points based on the Hessian response value.

[0036] S302: Based on the initial candidate feature point set, construct the local direction description vector to obtain the candidate feature description set.

[0037] Specifically, for any initial candidate feature point, the local gradient response is statistically analyzed in its neighborhood, and its principal direction is calculated. Preferably, the neighborhood of the feature point is divided into multiple sub-regions, and the projection component of the gradient in the principal direction coordinate system is calculated for each sub-region. The statistics of each sub-region are then concatenated to form a description vector. Finally, a candidate feature description set is generated based on the principal direction and description vector corresponding to each feature point.

[0038] S303: Based on the candidate feature description set, obtain the initial set of corresponding point pairs through bidirectional matching within the region.

[0039] Specifically, for any two standardized images that have a candidate overlap relationship... and Feature description sets are established in the corresponding candidate overlapping regions. and .for Any description vector Calculate its relationship with Each description vector The distance is described, and the matching cost function is constructed by combining the relative positional consistency within the candidate overlapping regions:

[0040] in, Represents a standardized image The Middle i Candidate feature points and standardized image The Middle j The cost of matching between candidate feature points This represents the weighting coefficients describing the similarity terms. Represents a standardized image The Middle i The description vector corresponding to each candidate feature point Represents a standardized image The Middle j The description vector corresponding to each candidate feature point Represents Euclidean distance. Represents the distance scale parameter. This represents the weighting coefficient of the position consistency term. Representing a standardized image The Middle i The position coordinates of the candidate feature points Represents a standardized image The Middle j The position coordinates of the candidate feature points Represents a standardized image The Middle jEach candidate feature point is mapped to a normalized image based on its candidate offset. Coarse mapping position in coordinate reference system Indicates the first k Candidate overlapping regions, Indicates the first k The area of ​​each candidate overlapping region. and Obtained from calibration.

[0041] For feature point pairs that satisfy the bidirectional nearest neighbor constraint :

[0042] in, i Represents a standardized image The index of a candidate feature point in the data. Represents a standardized image The index of candidate feature points that establishes matching relationships with candidate feature points. t Represents a standardized image The candidate feature point index variable that participates in the traversal comparison. s Represents a standardized image The candidate feature point index variable that participates in the traversal comparison. Represents a standardized image The Middle i Candidate feature points and standardized image The Middle t The cost of matching between candidate feature points Represents a standardized image The Middle s Candidate feature points and standardized image The Middle j The cost of matching between candidate feature points In the standardized image Selecting from all candidate feature points to achieve the matching cost Minimum candidate feature point index j , In the standardized image Selecting from all candidate feature points to achieve the matching cost Minimum candidate feature point index i .

[0043] Record it in the initial set of corresponding point pairs.

[0044] S304: Perform control point filtering on the initial set of corresponding point pairs to obtain a control point matching set.

[0045] Specifically, to avoid control points being too concentrated in local texture-dense areas, a joint score of matching reliability and spatial dispersion is constructed for each pair of points in the initial set of corresponding point pairs, and the first preset number of point pairs with higher scores are selected as the control point matching set.

[0046] Optionally, the joint scoring is as follows:

[0047] in, Indicates the first i The candidate feature point and the first j The joint score of point pairs formed by candidate feature points, where exp represents the exponential function. This represents the matching cost decay scale parameter. Indicates the spatial dispersion enhancement coefficient. Indicates the first i The candidate feature point and the first j The minimum spatial distance between candidate feature points Indicates the average minimum spatial distance. It represents a small positive number in the same unit as distance.

[0048] It should be noted that the purpose of this step is to provide a moderate number of evenly distributed and reliable control points for the subsequent TPS solution, rather than directly using all coarse matching points for non-rigid fitting.

[0049] S305: Based on the control point matching set, a non-rigid mapping relationship is obtained by fitting the non-rigid transformation of the thin plate spline.

[0050] Specifically, each pair of control points in the control point matching set is used as the source control point and the target control point, respectively. The source control point is the coordinate of the control point participating in the fitting in the first standardized image, and the target control point is the coordinate of the control point corresponding to the source control point in the second standardized image. Based on this, a radial basis kernel matrix reflecting the local deformation propagation characteristics of the control points is constructed according to the spatial distance relationship between the source control points. Simultaneously, a low-order polynomial term matrix describing the overall translation, rotation, and affine trends is constructed. The radial basis kernel matrix and the low-order polynomial term matrix are then jointly assembled to form a set of fitting equations for the non-rigid transformation of the thin-plate spline. The fitting equations are solved along the lateral and longitudinal coordinate components of the target control point to obtain the lateral and longitudinal mapping parameters, respectively. Based on the lateral and longitudinal mapping parameters, a two-dimensional non-rigid mapping relationship is established from the coordinates of the first standardized image to the coordinates of the second standardized image.

[0051] S306: Map candidate feature points according to non-rigid mapping relationships and obtain a fine point matching set by combining geometric consistency screening.

[0052] Specifically, for any feature point to be verified within the candidate overlapping region of the first image, its position in the second image is calculated using a non-rigid mapping relationship. Then, the nearest candidate feature point is searched within its location neighborhood, and the non-rigid mapping residual is calculated. Simultaneously, geometric consistency screening is performed on all candidate corresponding point pairs, preferably using an interior point determination method based on random sampling consistency. When a point pair satisfies that the non-rigid mapping residual is less than a preset residual value and is determined to be an interior point in the geometric consistency screening, it is recorded as a fine-point matching interior point. Finally, all interior points constitute a fine-point matching set.

[0053] In this embodiment of the invention, an initial correspondence with local saliency and distinguishability can be established first within the candidate overlapping region using point feature extraction. Then, a non-rigid mapping relationship describing the local deformation and overall registration trend can be constructed through control point screening and thin-plate spline non-rigid transformation fitting. This avoids the accumulation of mismatches and mapping instability caused by relying solely on coarse matching point pairs. Simultaneously, the candidate feature points are re-mapped and verified using the non-rigid mapping relationship, and a fine point matching set is obtained by combining geometric consistency screening. This effectively improves the accuracy, uniformity of distribution, and reliability of matching corresponding points, providing stable and accurate basic constraints for subsequent calculation of the vertical structure saliency index, vertical line enhancement matching, and geometric correction parameter solving.

[0054] S4: Calculate the vertical structure saliency index of each candidate overlapping region based on the spatial distribution direction, gradient principal direction, and local edge structure information of the fine points in the fine point matching set in each candidate overlapping region.

[0055] Among them, the vertical structure saliency index is a comprehensive index used to measure the strength of vertical structural features in candidate overlapping regions. It judges whether there are obvious vertical structures such as columns, wall boundaries, and window frame edges in the region by comprehensively utilizing the spatial distribution direction of the fine point matching set in the region, the consistency of the main direction of the local gradient, and the degree of support of the edge line segment for the vertical structure.

[0056] It should be noted that, in order to address the problem that existing automatic matching and correction methods for panoramic photos use the same enhancement matching strategy for all candidate overlapping regions, which can easily lead to the introduction of invalid line features in unstructured regions, increase the computational burden, and fail to fully utilize the line structure constraint in regions with obvious facade boundaries, window frame edges, columnar outlines, etc., resulting in a lack of targeted matching enhancement, this invention innovatively introduces a vertical structure saliency index to achieve structured screening of candidate overlapping regions. This ensures that vertical line enhancement matching is triggered only in regions with significant vertical structure features, thereby avoiding invalid calculations and the introduction of erroneous line constraints, while improving the structural matching capability in facade, boundary, and columnar regions.

[0057] In one possible implementation, the calculation method for the vertical structure salience index specifically includes: Based on the spatial distribution direction of the fine points in the fine point matching set within each candidate overlapping region, calculate the vertical distribution coefficient of each candidate overlapping region:

[0058] in, Indicates the first k Vertical distribution coefficients of candidate overlapping regions Indicates that it belongs to the symbol. Indicates the first k A subset of fine-grained point matching within candidate overlapping regions, where cos represents the cosine function. Indicates the first u The displacement direction angle of a fine-point matching point pair Represents pi (π). Represents the two-parameter arctangent function, ( , ) indicates the first u A fine-point matching point pair in the standardized image Coordinates in ( , ) indicates the relationship with the first u Each fine-point matching pair corresponds to a fine-point matching pair in the standardized image. The coordinates in the diagram.

[0059] It should be noted that when The larger the value, the closer the displacement direction of the fine point matching within the candidate overlapping area is to a vertical distribution, and the more likely it corresponds to vertical structures such as columns, window frames, and wall boundaries.

[0060] Based on the principal gradient direction in the neighborhood corresponding to the fine point in the fine point matching set, calculate the vertical gradient consistency coefficient of each candidate overlapping region:

[0061] in, Indicates the first k Vertical gradient consistency coefficient of candidate overlapping regions Indicates the first u Each fine-point matching point corresponds to the gradient principal direction angle in radians within the local neighborhood.

[0062] Based on the local edge structure information in each candidate overlapping region, calculate the vertical edge support coefficient of each candidate overlapping region:

[0063] in, Indicates the first k Vertical edge support coefficients of candidate overlapping regions m Indicates the first candidate in the overlapping region m Index of edge line segments, Indicates the first k The total number of edge segments in each candidate overlapping region Indicates the first k The set of edge line segments in the candidate overlapping regions This represents an indicator function; it takes the value 1 when the condition within the parentheses is true, and takes the value 0 when the condition within the parentheses is false. Indicates the first m The coordinates of the endpoint closer to the vertical reference point among the two endpoints of the edge line segment. Indicates the first m The coordinates of the endpoint of the edge line segment that is farther from the vertical reference point. Indicates the first k The vertical reference point coordinates corresponding to each candidate overlapping region Indicates the angle between rays. Indicates the angle threshold.

[0064] The vertical structure significance index of each candidate overlapping region is calculated based on the vertical distribution coefficient, the vertical gradient consistency coefficient, and the vertical edge support coefficient.

[0065] Optionally, the vertical structure significance index is specifically:

[0066] in, Indicates the first k Vertical structure significance index of candidate overlapping regions The weights representing the vertical distribution coefficients, The weights representing the vertical gradient consistency coefficients, This represents the weight of the support coefficient at the vertical edge.

[0067] It should be noted that, , , as well as Obtained from calibration.

[0068] S5: Determine whether the vertical structure saliency index of each candidate overlapping region is greater than the threshold. If so, mark the candidate overlapping region as the target candidate overlapping region and proceed to S6. Otherwise, set the line matching set corresponding to the candidate overlapping region to an empty set and proceed to S7.

[0069] It should be noted that those skilled in the art can set the threshold value according to actual needs, and this invention does not impose any limitations on it.

[0070] Furthermore, the vertical structure saliency index of each candidate overlapping region is obtained, and the vertical structure saliency index of each candidate overlapping region is compared with a threshold. When the vertical structure saliency index of a candidate overlapping region is greater than the threshold, it is determined that the candidate overlapping region has obvious vertical structure features, and the candidate overlapping region is marked as a target candidate overlapping region. Then, the target candidate overlapping region and its corresponding normalized image are sent to step S6 to perform vertical line enhancement matching. When the vertical structure saliency index of a candidate overlapping region is less than or equal to the threshold, it is determined that the candidate overlapping region does not have significant vertical structure features. In this case, vertical line enhancement matching is no longer performed on the candidate overlapping region. Instead, the line matching set corresponding to the candidate overlapping region is directly set to an empty set, and the fine point matching result corresponding to the candidate overlapping region is sent to step S7 together. In this way, vertical line enhancement matching can be selectively introduced based on the differences in structural features of different candidate overlapping regions, thereby avoiding the introduction of invalid line constraints or mismatch results in regions with indistinct vertical structures.

[0071] S6: Based on the target candidate overlapping regions and the corresponding standardized image set, obtain the line matching set of the target candidate overlapping regions through vertical line enhancement matching, and proceed to S7.

[0072] The vertical line enhancement matching method refers to the process of extracting vertical line segments from the image after completing the basic point feature matching and establishing the non-rigid mapping, targeting the candidate overlapping area with obvious vertical structural features, and using vertical reference points, non-rigid mapping relationships, spatial projection constraints, and color consistency of the two-sided neighborhood of the line segments to determine the correspondence of candidate vertical lines in different images.

[0073] It should be noted that, in order to address the problem that existing automatic matching and correction methods for panoramic photos, which rely solely on point feature matching, tend to suffer from sparse matching, insufficient structural constraints, and unstable boundary alignment in areas with weak texture, repetitive texture, facade boundaries, and local occlusion, leading to local misalignment of geometric correction results in vertically significant areas such as window frame edges, wall outlines, and columnar structures, this invention innovatively introduces a vertical line enhanced matching mechanism. This mechanism can supplement the point matching constraints with explicit geometric constraints on vertical structures, thereby enhancing the matching stability and alignment accuracy of structural boundaries in the overlapping areas of target candidates.

[0074] Furthermore, instead of blindly performing line feature matching in all candidate overlapping regions, this invention extracts initial vertical line candidates based on vertical reference points within the target candidate overlapping regions selected in step S5. Then, combined with the non-rigid mapping relationship obtained in the previous steps, the candidate vertical lines are guided by cross-image mapping to narrow the search range. Furthermore, spatial constraints are formed through hierarchical spatial projection, and color consistency constraints are constructed through the dominant colors of the neighborhoods on both sides of the line segment. Based on this, candidate line pairs are generated and geometric confidence is screened to finally obtain the line matching set.

[0075] In one possible implementation, S6 specifically includes: S601: Determine the vertical reference point based on the target candidate overlapping region and the corresponding standardized image set, and extract the initial vertical line candidate set based on the vertical reference point.

[0076] Specifically, for any two standardized images corresponding to the overlapping region of a target candidate... and Each segment determines its own vertical reference point. This reference point can be determined by the projection of the camera center onto the imaging plane. After determining the vertical reference points, edge extraction and line segment detection are performed on the normalized image within the overlapping area of ​​the target candidates to obtain a set of edge segments. For any segment in the edge segment set, the endpoint closer to the vertical reference point and the endpoint farther from the vertical reference point are determined, and the deviation angle of the segment relative to the vertical reference point is calculated. Segments with deviation angles less than a deviation angle threshold are identified as initial vertical line candidates and added to the initial vertical line candidate set.

[0077] S602: Based on the non-rigid mapping relationship, perform cross-image mapping on the initial vertical candidate set to generate a candidate line pair set.

[0078] Specifically, for any perpendicular line in the initial perpendicular line candidate set of the first normalized image within any target candidate overlapping region, multiple sampling points are obtained by uniformly sampling the perpendicular line along its length direction using the established non-rigid mapping relationship. Then, using the non-rigid mapping relationship, each sampling point is mapped one by one to the coordinate system of the second normalized image, thus forming a corresponding mapping trajectory. Subsequently, within the corresponding target candidate overlapping region of the second normalized image, each perpendicular line in its initial perpendicular line candidate set is traversed, and the average point-to-line distance between the mapping trajectory and each candidate perpendicular line is calculated to characterize the non-rigid mapping guidance residual between the two perpendicular lines. If the non-rigid mapping guidance residual is less than a preset residual value, the line pair formed by the two perpendicular lines is recorded as a valid candidate line pair and included in the candidate line pair set.

[0079] S603: Perform layered spatial projection on each candidate line pair in the candidate line pair set, calculate the number of corresponding points at the same position for each candidate line pair, and obtain the spatial constraint strength.

[0080] Specifically, for any candidate line pair in the candidate line pair set, the corresponding line segments from the two standardized images are projected onto a set of discrete hierarchical spatial planes. Then, in each hierarchical spatial plane, it is calculated whether the two projected line segments intersect, so as to obtain the number of points with the same position and name for the candidate line pair in different spatial layers.

[0081] S604: Based on the candidate line pair set, determine the color consistency constraint result by extracting the dominant color of the two-sided neighborhood of the line segment.

[0082] In one possible implementation, S604 specifically includes: S6041: Construct a two-sided polar coordinate neighborhood for each candidate line segment based on the two vertical reference points corresponding to each candidate line segment in the candidate line pair set.

[0083] Specifically, for each candidate line segment in any candidate line pair, a polar coordinate reference system is established with the vertical reference point in the corresponding standardized image as the pole. Then, the neighborhood of the candidate line segment is divided into a clockwise side neighborhood and a counterclockwise side neighborhood, using the observation vector of the candidate line segment relative to the vertical reference point as the boundary, thus obtaining the two-sided polar coordinate neighborhood of each candidate line segment.

[0084] S6042: Based on the polar coordinate neighborhood of the line segment, perform bilateral rotation sampling on each candidate line segment to obtain the set of pixels in the bilateral neighborhood.

[0085] Specifically, for any candidate line segment, it is rotated slightly in both clockwise and counterclockwise directions around its corresponding vertical reference point, and the set of pixels adjacent to the candidate line segment is extracted at the rotated position to obtain the clockwise side neighboring pixel set and the counterclockwise side neighboring pixel set of the candidate line segment.

[0086] S6043: Perform color space conversion and abnormal pixel removal on the set of pixels in both neighboring regions to obtain a stable subset of color pixels.

[0087] Specifically, the set of neighboring pixels on both sides of each candidate line segment is converted from the RGB color space to the Lab color space. Then, the color difference between each neighboring pixel and the average color of that side is calculated, and abnormal pixels with a color difference greater than a preset color difference are removed to obtain the corresponding stable color pixel subset.

[0088] S6044: Extract the dominant color of the two adjacent neighborhoods of the current line segment based on the stable color pixel subset.

[0089] Specifically, the primary color descriptor is calculated for each stable color pixel subset on both sides of any candidate line segment to obtain the primary color on the clockwise side and the primary color on the counterclockwise side of the candidate line segment.

[0090] Optionally, after converting each pixel from the RGB color space to the Lab color space, the channel components in each pixel subset are summed and normalized according to the number of pixels on that side to obtain the average color vector of that neighborhood. The average color vector is then used as the primary color descriptor for that neighborhood.

[0091] Furthermore, to improve the stability of the primary color, before calculating the average color, pixels in the stable color pixel subset are preferably further filtered according to color differences, removing extreme pixels that deviate from the overall color distribution on that side. This reduces the impact of noise points, highlight points, or boundary mixed pixels on the primary color calculation. After filtering, statistical calculations are performed on the remaining pixels to obtain a more stable primary color descriptor.

[0092] S6045: Determine the color consistency constraint result based on the color difference between the primary colors of the two neighboring regions.

[0093] Specifically, for any candidate line pair, the color difference between the dominant colors of the corresponding line segments in the clockwise and counterclockwise neighborhoods of the candidate line pair in the two standardized images is calculated respectively. The color difference is preferably calculated using a color difference measurement method based on the CIEDE2000 standard. If the color difference between the dominant colors of the clockwise and counterclockwise neighborhoods is less than a preset color consistency threshold, the candidate line pair is considered to satisfy the color consistency condition on at least one side, thus determining that the candidate line pair meets the color consistency constraint. Conversely, if the color difference of the dominant colors of the candidate line pair in both the clockwise and counterclockwise neighborhoods is greater than the preset color consistency threshold, the candidate line pair is determined not to meet the color consistency constraint. The final color consistency constraint result is then obtained.

[0094] It should be noted that the determination method of satisfying color consistency on at least one side is adopted because the neighboring area on the other side of the candidate line segment may be significantly changed due to occlusion, background switching or perspective changes under different shooting angles. Therefore, the single-sided consistency determination is more suitable for constructing color constraints for candidate line pairs in panoramic photos.

[0095] In this embodiment of the invention, by extracting the main colors of the neighborhood on both sides of the line segment, the color information on both sides of the candidate line segment can be extracted more stably, reducing the impact of occlusion, perspective changes, specular reflection, and boundary mixed pixels on color comparison. At the same time, by adopting a determination method that satisfies color consistency on at least one side, it can adapt to the situation where the neighborhood on one side changes under different shooting angles, thereby reducing the probability of true matching line pairs being mistakenly eliminated and improving the reliability of color consistency constraints.

[0096] S605: Based on the spatial constraint strength and color consistency constraint results, and combined with the non-rigid mapping guided residuals, an initial line matching candidate set is generated.

[0097] Specifically, for any candidate line pair in the candidate line pair set, the calculated spatial constraint strength and color consistency constraint results are first obtained. Simultaneously, combined with the corresponding non-rigid mapping guided residual from step S602, a fusion evaluation index is constructed to characterize the matching reliability of the candidate line pair. Based on this, all candidate line pairs are sorted according to the fusion evaluation index, and a one-to-one mutually exclusive greedy selection strategy is adopted to sequentially select candidate line pairs with higher scores and add them to the initial line matching candidate set. During the selection process, if any line segment in a candidate line pair is already occupied by a previously selected candidate line pair, that candidate line pair is discarded to ensure the uniqueness and consistency of the matching results. Finally, all candidate line pairs that satisfy the above fusion evaluation and mutual exclusion constraints constitute the initial line matching candidate set.

[0098] S606: Perform geometric confidence screening on the initial line matching candidate set to obtain the line matching set.

[0099] Specifically, for any candidate line pair in the initial line matching candidate set, based on the imaging model of the corresponding standardized image, the two perpendicular lines in the candidate line pair are back-projected in reverse perspective, thereby constructing their respective spatial solution geometry in three-dimensional space. Further, the intersection relationship between the spatial geometry structures formed by the back projections of the two perpendicular lines is calculated, and the geometric consistency of the candidate line pair in space is evaluated accordingly. Based on this, geometric reliability judgment conditions are constructed by calculating the consistency index of the spatial intersection lines and their deviation from the ideal vertical direction. When a candidate line pair simultaneously satisfies that the intersection consistency is greater than the target value and the perpendicularity deviation is less than the target value, the candidate line pair is judged as a geometrically reliable match and included in the line matching set. Otherwise, it is judged as a mismatch and removed. Finally, all candidate line pairs that pass the geometric reliability screening constitute the line matching set and proceed to S7.

[0100] In this embodiment of the invention, through the processing in step S6, stable line matching constraints can be further introduced for the vertically significant structures in the target candidate overlapping region, in addition to the basic constraints provided by fine point matching. This effectively compensates for the problems of sparse point matching and insufficient boundary constraints in weak texture regions, repetitive texture regions, structural boundary regions, and locally occluded regions. At the same time, this step is not a simple line segment detection and matching, but rather a combination of vertical reference point constraints, non-rigid mapping guidance, spatial constraints formed by layered spatial projection, color consistency constraints formed by the main colors of the two-sided neighborhood of the line segment, and geometric confidence screening. This allows unreliable candidate line pairs to be eliminated layer by layer, improving the accuracy and stability of the line matching results.

[0101] S7: Generate an enhanced set of points with the same name and the corresponding set of matching weights based on the fine point matching set, the line matching set, and the non-rigid mapping relationship.

[0102] In one possible implementation, S7 specifically includes: S701: Unify the coordinate representation of the fine point matching set and the line matching set to generate the basic fusion matching unit set.

[0103] Specifically, for each set of point matching units in the fine point matching set, the coordinate correspondence between them in the first and second normalized images is preserved. For each set of line matching units in the line matching set, discrete sampling is performed along the length direction of the corresponding matching line segment, converting each set of matching line segments into multiple pairs of sampling points with sequential correspondence. Then, the point matching units and the sampling point pairs converted from the line matching units are uniformly represented as basic fusion matching units, thereby generating a set of basic fusion matching units.

[0104] It should be noted that this step does not directly involve the entire line segment in the subsequent solution. Instead, it discretizes the line matching result into pairs of sampling points with consistent order, so that the line constraint can be incorporated into the subsequent enhanced point set construction process in a structured point constraint manner, thereby taking into account the consistency of point matching and line matching in terms of data representation.

[0105] S702: Calculate the mapping consistency coefficient of each basic fusion matching unit in the basic fusion matching unit set according to the non-rigid mapping relationship.

[0106] Specifically, for any matching unit in the basic fusion matching unit set, a non-rigid mapping relationship is used to map its coordinates in the first standardized image to the coordinate system of the second standardized image, and the deviation between the mapped predicted position and the true corresponding position is calculated. Then, the mapping consistency coefficient of the matching unit is determined based on the deviation.

[0107] Alternatively, the mapping consistency coefficient can be expressed as:

[0108] in, Indicates the first n The mapping consistency coefficient of each basic fusion matching unit. This represents the square of the Euclidean distance. Indicates the first n The mapping positions of the basic fusion matching units are predicted by non-rigid mapping relationships. This indicates the actual location of the current basic fusion matching unit. This represents the mapping residual scaling parameter.

[0109] S703: Determine the structural contribution coefficient of each basic fusion matching unit based on the vertical structural significance index of the candidate overlapping region and the matching unit type.

[0110] Specifically, the vertical structure saliency index corresponding to the candidate overlapping region to which each basic fusion matching unit belongs is obtained. When the basic fusion matching unit comes from the fine point matching set, the regional structure contribution coefficient of the point matching unit is determined based on the vertical structure saliency index. When the basic fusion matching unit comes from the sampled point pairs obtained by discretization of the line matching set, the line structure enhancement coefficient is superimposed on the vertical structure saliency index to increase the influence of structured constraints in the vertical structure saliency region on the subsequent geometric correction solution.

[0111] Alternatively, the structural contribution coefficient can be expressed as:

[0112] in, Indicates the first n The structural contribution coefficients of each basic fusion matching unit This represents the structural reinforcement factor (the structural reinforcement factor is set according to actual needs). Indicates the first n The matching unit type flag of the first basic fusion matching unit, when the first... n When the basic fusion matching units are derived from the fine point matching set When the first n When each basic fusion matching unit is derived from the sampling point pairs obtained by discretizing the line matching set, .

[0113] S704: Generate a matching weight set based on the mapping consistency coefficient and the structural contribution coefficient.

[0114] Specifically, for any basic fusion matching unit, its corresponding mapping consistency coefficient and structural contribution coefficient are jointly fused to obtain the comprehensive matching weight of that matching unit. Then, the comprehensive matching weights corresponding to all basic fusion matching units are summarized to generate a matching weight set.

[0115] S705: Based on the matching weight set, the basic fusion matching units are filtered and enhanced to generate an enhanced set of points with the same name.

[0116] Specifically, all basic fusion matching units are sorted according to the matching weight set. Matching units with higher weights and more balanced spatial distribution are retained first. For sampling point pairs obtained by discretizing line matching units, the sequential continuity and spacing stability between adjacent sampling point pairs are further checked. When the continuity constraint is satisfied, the entire pair is retained as a structural enhancement subset. For discrete line sampling point pairs that do not satisfy the continuity constraint, they are retained or removed according to the single-point constraint method. Finally, all matching units that pass the screening form an enhanced set of corresponding points.

[0117] It should be noted that the innovation of this step lies in the fact that instead of simply breaking down the line matching into several independent points and mixing them in directly, a "structural continuity preservation" mechanism is introduced, which allows sampling points from the same matching vertical line to strengthen the local geometric structure in a group manner, thereby forming stronger corrective constraints on areas such as facade boundaries, window frame edges, and columnar structures in subsequent steps.

[0118] In this embodiment of the invention, through the processing in step S7, the fine point matching set, line matching set, and non-rigid mapping relationship can be uniformly incorporated into the same fusion framework, so that the matching constraints from different sources are consistent in data expression, which facilitates the unified solution of subsequent geometric correction parameters. At the same time, by calculating the mapping consistency coefficient and structural contribution coefficient respectively, and generating a matching weight set accordingly, the matching units with higher reliability and stronger structural significance can play a greater constraining role in subsequent solutions, thereby improving the stability and pertinence of the fusion results.

[0119] S8: Solve for the geometric correction parameters based on the enhanced set of corresponding points and the matching weight set, and output the corrected panoramic photo dataset.

[0120] In one possible implementation, S8 specifically includes: S801: Construct a weighted geometric correction solution model based on the enhanced set of corresponding points and the matching weight set.

[0121] Specifically, each pair of corresponding points in the enhanced corresponding point set is used as a geometric correction constraint term, and each weight in the matching weight set is used as a weighting coefficient for the corresponding constraint term. With the goal of aligning the corresponding points in the first standardized image as closely as possible with their corresponding points in the second standardized image after geometric correction, a weighted solution model for the geometric correction parameters is constructed. The objective function of the weighted solution model is specifically expressed as follows:

[0122] in, J This represents the objective function of the weighted geometric correction solution model. q This represents the index of the matching unit in the enhanced set of corresponding points. express qMatching weights corresponding to each enhanced corresponding point unit, Indicates geometric correction parameters, Represents geometric correction parameters Z The determined geometric mapping function corresponds to the i-th point in the normalized image. The mapped position obtained after transformation. Indicates the first in the enhanced set of corresponding points q Each matching unit in the standardized image The coordinates of the point in the diagram represent the coordinates of the point in the diagram. The corresponding location is in the standardized image The coordinates of the corresponding points in the [reference].

[0123] S802: Solve the parameters of the weighted geometric correction solution model to obtain the geometric correction parameters.

[0124] Specifically, the constructed weighted geometric correction solution model is solved as a weighted least squares optimization problem. First, the parameter results or identity mapping results corresponding to the aforementioned non-rigid mapping relationship are used as the initial values ​​of the geometric correction parameters. Then, for each pair of corresponding points in the enhanced point set, the predicted position under the current geometric correction parameters and the residual between them and the actual corresponding position are calculated. Based on this, the geometric mapping function is linearized to the first order to obtain the parameter Jacobian matrix, and a weighted normal equation is constructed by combining it with the matching weight set. Subsequently, the parameter increment is solved iteratively, and the geometric correction parameters are updated using the parameter increment until the parameter increment is less than a preset convergence value, or the overall weighted residual change is less than a preset residual value, at which point the iteration stops. Finally, the converged parameter results are determined as the geometric correction parameters.

[0125] S803: Based on the geometric correction parameters, perform coordinate mapping and pixel resampling on the standardized image set to obtain the corrected image set.

[0126] Specifically, for each standardized image in the standardized image set, pixel coordinate mapping is performed according to geometric correction parameters to transform the original pixel positions to the corrected target coordinate positions. For pixels that fall into non-integer coordinate positions after mapping, pixel resampling is performed using a preset interpolation method to generate the corresponding corrected image, thus obtaining the corrected image set.

[0127] Optionally, the preset interpolation method is bilinear interpolation or bicubic interpolation.

[0128] S804: Based on the corrected image set and the candidate overlapping region set, perform overlapping region fusion and boundary cropping on the corrected batch of panoramic images to obtain the corrected panoramic photo dataset.

[0129] Specifically, each corrected image in the corrected image set is mapped to a unified panoramic coordinate reference system according to its corresponding geometric position. Then, based on the candidate overlapping region set, the overlapping regions between the corrected images are determined, and the pixels within each overlapping region are fused to reduce brightness abruptness and structural misalignment at the seams, resulting in a panoramic stitching result. Next, invalid boundary regions, blank regions, or regions exceeding the effective imaging range in the panoramic stitching result are cropped to obtain valid panoramic images. Finally, each valid panoramic image is encapsulated and output to obtain the corrected panoramic photo dataset.

[0130] Reference manual attached Figure 2 The diagram shows a schematic representation of a batch automatic matching and correction system for panoramic photo data provided in an embodiment of the present invention.

[0131] This invention provides a batch automatic matching and correction system 20 for panoramic photo data, including: a processor 201 and a memory 202; The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-described automatic batch matching and correction method for panoramic photo data and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention 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 of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for batch automatic matching and correction of panoramic photo data, characterized in that, include: S1: Obtain batch panoramic photo data to be processed; S2: Preprocess the batch of panoramic photo data to determine a set of candidate overlapping regions containing multiple candidate overlapping regions and a standardized image set; S3: Based on the candidate overlapping region set and the standardized image set, determine the fine point matching set and non-rigid mapping relationship through point feature extraction and non-rigid transformation fitting; S4: Calculate the vertical structure saliency index of each candidate overlapping region based on the spatial distribution direction, gradient principal direction, and local edge structure information of the fine points in the fine point matching set in each candidate overlapping region; S5: Determine whether the vertical structure saliency index of each candidate overlapping region is greater than the threshold; if so, mark the candidate overlapping region as the target candidate overlapping region and proceed to S6; otherwise, set the line matching set corresponding to the candidate overlapping region to an empty set and proceed to S7. S6: Based on the target candidate overlapping region and the corresponding standardized image set, obtain the line matching set of the target candidate overlapping region through vertical line enhancement matching, and proceed to S7; S7: Generate an enhanced set of points with the same name and a corresponding set of matching weights based on the fine point matching set, the line matching set, and the non-rigid mapping relationship; S8: Solve for the geometric correction parameters based on the enhanced set of corresponding points and the matching weight set, and output the corrected panoramic photo dataset.

2. The method for batch automatic matching and correction of panoramic photo data according to claim 1, characterized in that, S2 specifically includes: S201: Decode and unify the channels of the batch of panoramic photo data to obtain the original image set; S202: Perform brightness normalization and noise suppression on the original image set to obtain a normalized image set; S203: Perform resolution unification and coordinate standardization on the normalized image set to obtain a standardized image set; S204: Calculate the candidate offsets for the standardized image set to obtain a candidate offset set; S205: Determine the candidate overlapping region set, which contains multiple candidate overlapping regions, based on the candidate offset set.

3. The method for batch automatic matching and correction of panoramic photo data according to claim 1, characterized in that, S3 specifically includes: S301: Based on the candidate overlapping region set and the standardized image set, construct a multi-scale feature response map within the candidate overlapping region to obtain an initial candidate feature point set; S302: Based on the initial candidate feature point set, construct a local direction description vector to obtain a candidate feature description set; S303: Based on the candidate feature description set, an initial set of corresponding point pairs is obtained through bidirectional matching within the region; S304: Perform control point filtering on the initial set of corresponding point pairs to obtain a control point matching set; S305: Based on the control point matching set, the non-rigid mapping relationship is obtained by fitting the non-rigid transformation of the thin plate spline; S306: Based on the non-rigid mapping relationship, the candidate feature points are mapped, and combined with geometric consistency screening, the fine point matching set is obtained.

4. The method for batch automatic matching and correction of panoramic photo data according to claim 1, characterized in that, The calculation method for the vertical structure salience index specifically includes: Based on the spatial distribution direction of the fine points in the fine point matching set in each of the candidate overlapping regions, calculate the vertical distribution coefficient of each of the candidate overlapping regions; Based on the gradient principal direction in the neighborhood corresponding to the fine point in the fine point matching set, calculate the vertical gradient consistency coefficient of each candidate overlapping region. Based on the local edge structure information in each of the candidate overlapping regions, the vertical edge support coefficient of each of the candidate overlapping regions is calculated; The vertical structure significance index of each candidate overlapping region is calculated based on the vertical distribution coefficient, the vertical gradient consistency coefficient, and the vertical edge support coefficient.

5. The method for batch automatic matching and correction of panoramic photo data according to claim 1, characterized in that, S6 specifically includes: S601: Determine a vertical reference point based on the target candidate overlapping region and the corresponding standardized image set, and extract an initial vertical line candidate set based on the vertical reference point; S602: Based on the non-rigid mapping relationship, perform cross-image mapping on the initial vertical line candidate set to generate a candidate line pair set; S603: Perform hierarchical spatial projection on each candidate line pair in the candidate line pair set, calculate the number of corresponding points at the same position for each candidate line pair, and obtain the spatial constraint strength. S604: Based on the candidate line pair set, determine the color consistency constraint result by extracting the dominant color of the two-sided neighborhood of the line segment; S605: Based on the spatial constraint strength and the color consistency constraint results, and combined with the non-rigid mapping guided residual, an initial line matching candidate set is generated; S606: Perform geometric confidence screening on the initial line matching candidate set to obtain the line matching set.

6. The method for batch automatic matching and correction of panoramic photo data according to claim 5, characterized in that, Specifically, S604 includes: S6041: Construct a two-sided polar coordinate neighborhood of a line segment based on the two vertical reference points corresponding to each candidate line segment in the candidate line pair set; S6042: Based on the polar coordinate neighborhood of the line segment, perform bilateral rotation sampling on each candidate line segment to obtain a set of pixels in the bilateral neighborhood. S6043: Perform color space conversion and abnormal pixel removal on the bilateral neighboring pixel set to obtain a stable color pixel subset; S6044: Extract the dominant color of the two-sided neighborhood of the current line segment based on the stable color pixel subset; S6045: Determine the color consistency constraint result based on the color difference between the primary colors of the two neighboring regions.

7. The method for batch automatic matching and correction of panoramic photo data according to claim 1, characterized in that, Specifically, S7 includes: S701: Perform a unified coordinate expression on the fine point matching set and the line matching set to generate a basic fusion matching unit set; S702: Calculate the mapping consistency coefficient of each basic fusion matching unit in the basic fusion matching unit set according to the non-rigid mapping relationship; S703: Determine the structural contribution coefficient of each of the basic fusion matching units based on the vertical structural significance index and matching unit type of the candidate overlapping region; S704: Generate the matching weight set based on the mapping consistency coefficient and the structural contribution coefficient; S705: The basic fusion matching unit is filtered and enhanced according to the matching weight set to generate the enhanced homonymous point set.

8. The method for batch automatic matching and correction of panoramic photo data according to claim 1, characterized in that, S8 specifically includes: S801: Construct a weighted geometric correction solution model based on the enhanced set of corresponding points and the matching weight set; S802: Solve the parameters of the weighted geometric correction solution model to obtain the geometric correction parameters; S803: Based on the geometric correction parameters, perform coordinate mapping and pixel resampling on the standardized image set to obtain the corrected image set; S804: Based on the corrected image set and the candidate overlapping region set, perform overlapping region fusion and boundary cropping on the corrected batch of panoramic images to obtain the corrected panoramic photo dataset.

9. A batch automatic matching and correction system for panoramic photo data, characterized in that, include: Processor and memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the batch automatic matching and correction method for panoramic photo data as described in any one of claims 1 to 8.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the panoramic photo data batch automatic matching and correction method as described in any one of claims 1 to 8.

Images