Waterlogging panorama construction method based on panorama stitching and space projection

Abstract

The invention discloses a waterlogging panorama construction method based on panorama stitching and spatial projection, and the method comprises the steps: image collection and feature extraction: obtaining a multi-view waterlogging image from a monitoring video, and extracting and matching key points to establish a spatial feature relation; panoramic stitching and projection transformation are carried out, a point cloud model is constructed, a projection matrix is generated, multi-view images are stitched into a hexahedral panoramic image, and the calculation efficiency is optimized; spherical mapping and visualization: mapping the panoramic image to the inner wall of the sphere to realize immersive display; target detection and positioning are carried out, a waterlogging area is identified through deep learning, the ponding range is calculated in combination with camera parameters, and accurate monitoring is achieved; and panoramic browsing and dynamic updating are carried out, an interactive panoramic system is constructed, the ponding range and the change trend are displayed in real time, and the emergency response capability is improved. Through fusion of computer vision and space projection technologies, multi-view panoramic stitching and three-dimensional display are realized, and the coverage range, the real-time performance and the visualization capability of waterlogging monitoring are improved.

Description

Technical Field

[0001] The present application belongs to the field of intelligent information processing and relates to a method for constructing a panoramic view of urban flooding based on panoramic stitching and spatial projection. Background Art

[0002] Waterlogging disasters not only paralyze urban traffic and damage public facilities, but also pose a serious threat to the safety of life and property of residents. Although a variety of technologies have been applied to waterlogging monitoring and emergency management, such as sensor networks, hydrological models, and remote sensing technologies, these methods generally have problems such as limited monitoring coverage, lack of real-time performance, and delayed data updates, which make it difficult to meet the needs of rapid perception and precise positioning of waterlogging disasters. In addition, most of the current monitoring methods focus on single-point waterlogging detection and lack the ability to fully perceive the scope of waterlogging and three-dimensional visualization, which has caused great obstacles to the overall assessment and scientific decision-making of waterlogging.

[0003] The rapid development of video sensors, computer vision, and spatial information technology has brought new technical opportunities for waterlogging monitoring and visualization. Panoramic image stitching and spherical projection technology can reconstruct panoramic scenes from multi-view images. Combined with deep learning target detection and segmentation technology, it can achieve accurate identification and positioning of waterlogging areas. In addition, by dynamically updating waterlogging data and mapping it to a spherical projection model, the scope of waterlogging and its changing trend can be intuitively displayed in three-dimensional space, providing real-time and three-dimensional support for waterlogging monitoring and emergency response. Therefore, the waterlogging panorama construction technology based on panoramic stitching and spatial projection can effectively solve the shortcomings of traditional monitoring methods and has important research value and application prospects. Summary of the invention

[0004] In order to overcome the above defects, this application proposes a method for constructing a panoramic view of waterlogging based on panoramic stitching and spatial projection. The specific steps of this application are as follows:

[0005] S1, image acquisition and key feature extraction, collect multi-view flood scene images from surveillance video data, use SURF algorithm to extract key points and feature vectors of images, and establish the corresponding relationship of spatial features by matching the key points of consecutive frame images, providing a basis for subsequent splicing;

[0006] S2, panoramic image stitching and transformation matrix calculation, build a point cloud model based on the matched key points, use the corner detection algorithm to identify feature points, generate 8 linear equations and solve the 3×3 projection transformation matrix, stitch the multi-view images into a complete six-sided panorama, and use layered processing to generate multi-resolution pyramid tiles to adapt to different resolution requirements;

[0007] S3, spatial projection and spherical mapping, maps the spliced ​​panoramic image to the inner wall of the sphere through the projection transformation matrix, constructs a three-dimensional spherical projection model, and sets the observer's viewpoint at the center of the sphere, thereby achieving a full-scale and three-dimensional presentation of the waterlogging scene;

[0008] S4, target detection and positioning data inference, uses deep learning algorithms to detect and segment targets in panoramic images, combines camera installation parameters, builds a mathematical model to infer the specific location and size of the target, and provides accurate waterlogged area positioning and dynamic monitoring support for panoramic browsing;

[0009] S5, panoramic browsing and dynamic updating of waterlogging, combines the spherical panoramic image with the positioning information of the detected waterlogging area to build a panoramic browsing system for waterlogging, supports real-time dynamic updating and interactive viewing, intuitively displays the scope, depth and changing trend of the waterlogging area, and improves emergency response capabilities.

[0010] The technical features and improvements of this application are:

[0011] For step S1, the present application collects multi-view images containing waterlogged areas from surveillance videos, uses advanced image feature extraction algorithms to efficiently extract key points, and establishes spatial correspondences between images. The core technologies include feature extraction based on the SURF algorithm, key point matching, and optimized application of the scale-invariant feature transform (SIFT) algorithm, thereby significantly improving the accuracy and reliability of registration; the present application performs registration based on the obvious features of the image, including corner points, key points, edge contours, etc. Unlike the traditional method of directly using grayscale information, the present application uses the SURF algorithm to extract key points with rotation invariance and scale invariance, thereby avoiding the decrease in matching accuracy caused by changes in illumination, scale, and perspective. The SURF algorithm quickly locates key points by calculating the determinant of the Hessian matrix H(x, y, σ) of the image:

[0012]

[0013] Among them, L xx , L yy and L xy are the second-order derivatives of the image at different scales σ. By analyzing the response of the determinant, local extreme points are extracted as key points. To further enhance the stability of feature points, this application combines the SIFT algorithm to construct a scale space, extract the local gradient information of key points, and generate descriptors:

[0014] f SIFT ={d 1 ,d 2 ,...,d 128} (2)

[0015] Among them, fSIFT is a feature vector. The 128-dimensional descriptor is obtained by counting the gradient direction histogram in the neighborhood of the feature point, which is rotation invariant and illumination invariant.

[0016] Keypoint matching is achieved by calculating the Euclidean distance between descriptors of different image feature points:

[0017]

[0018] When Dist is less than the set threshold, the two feature points are considered to match. To reduce mismatching, this application introduces the nearest neighbor ratio criterion, which is defined as:

[0019]

[0020] Among them, Dist 1 and Dist 2 are the distances between the nearest and second nearest feature points, respectively. When NNDR is less than a certain threshold, the match is considered reliable;

[0021] In order to improve the efficiency of key point extraction, this application optimizes the Gaussian pyramid and difference Gaussian pyramid (DoG) in the SIFT algorithm, and uses a fast downsampling method to construct a pyramid in each level of scale space:

[0022] I t =I t-1 ↓2 (5)

[0023] Among them, ↓2 represents the downsampling operation, and t is the pyramid level. In the differential Gaussian pyramid, the differential images of adjacent scales are calculated:

[0024] D(x,y,t)=G t+1 (x,y)-G t (x,y) (6)

[0025] Where G is the image after Gaussian blur. In the DoG pyramid, this application performs extreme value detection on potential feature points. For any pixel point d(t,r,x,y), when its grayscale value is an extreme value in a 3×3 neighborhood, the point is considered a potential feature point. At the same time, this application removes feature points located at the edge with lower curvature, and accurately calculates the position of the feature point by fitting the grayscale value curve of the pixel where it is located.

[0026] For step S2, in order to achieve accurate stitching of multi-view images, this application adopts a point cloud model construction method based on key point matching, uses a 3×3 projection transformation matrix for image registration and stitching, solves the geometric relationship between multi-view images, and introduces image fusion technology to eliminate stitching gaps and improve the visual quality of panoramic images. In addition, multi-resolution pyramid tiles are generated through layered processing to meet different resolution requirements and improve system efficiency; based on the image registration key point matching results, this application constructs a point cloud model to reflect the spatial relationship of multi-view images, further optimizes the feature point selection through the corner detection algorithm, and constructs a transformation equation group based on the matched key points. The general form of the 3×3 projection transformation matrix H is:

[0027]

[0028] The solution of the projection matrix H is based on the following projection relationship:

[0029]

[0030] Where (x, y) is the source image coordinate, and (x′, y′) is the target image coordinate. A linear equation system is established by at least four pairs of key points and the matrix parameters {h ij};

[0031] The image is transformed by using the solved projection transformation matrix H, and the multi-view images are accurately registered to a unified spatial coordinate system to generate a complete six-sided panoramic image. During the stitching process, dealing with issues such as image brightness differences and registration errors is the key to improving the quality of the panoramic image. This application uses image fusion technology to eliminate stitching gaps. According to the image representation layer, the fusion method is divided into pixel-level fusion, feature-level fusion, and decision-level fusion. This application mainly uses pixel-level fusion with moderate computational effort, and solves the brightness difference problem in the stitching area through the weighted average method. The fusion formula of the weighted average method is:

[0032] I 融合 (x) = w 1 (x)I 1 (x)+w 2 (x)I 2 (x) (9)

[0033] Among them, I 1 (x) and I 2 (x) are the grayscale values ​​of the two registered images at pixel x; w 1 (x) and w 2 (x) is the weight, satisfying w 1 (x)+w 2 (x) = 1, and the weights are dynamically allocated using the gradual in and gradual out method:

[0034]

[0035] Among them, d 1 and d 2 are the left and right boundaries of the overlapping area, respectively, and d x is the horizontal coordinate of pixel x. This method can effectively smooth the stitching area and reduce the impact of grayscale differences on the panoramic image. To adapt to different resolution requirements, this application generates multi-resolution pyramid tiles through layered downsampling, and downsamples the stitched panoramic image layer by layer:

[0036] I t =I t-1 ↓2 (11)

[0037] Among them, ↓2 represents the downsampling operation, t is the pyramid level, and the downsampling results of each layer are saved to form pyramid tiles, which support efficient loading of different terminal devices.

[0038] For step S3, this application proposes a spatial projection and spherical mapping method based on a projection transformation matrix, which maps the stitched panoramic image to the inner wall of a three-dimensional sphere, constructs a spherical projection model with an all-round perspective, and thus realizes a three-dimensional and immersive display of the waterlogging scene. The following are the technical details of this step and the technical solution features of the invention. The 3×3 projection transformation matrix H generated in step S2 is used to perform spatial mapping on the stitched panoramic image, converting it from a two-dimensional plane coordinate system to a spherical coordinate system. The basic formula for projection transformation is:

[0039]

[0040] Among them, (x, y) is the pixel coordinate of the panoramic image in the original plane coordinate system, and (x′, y′) is the image coordinate after projection transformation. Through the above transformation, each pixel point in the spliced ​​image finds the corresponding position on the inner wall of the sphere to form a preliminary spatial mapping; based on the projection transformation, the mapped panoramic image is projected onto the inner wall of the sphere to form a complete three-dimensional spherical model. The spherical projection is implemented using the following formula:

[0041] X=RsinθcosΦY=RsinθsinΦZ=Rcosθ (13)

[0042] Among them, (X, Y, Z) are the three-dimensional coordinates on the sphere, R is the radius of the sphere, θ and φ represent the latitude and longitude angles on the sphere respectively. Through spherical projection, each pixel point is mapped from the two-dimensional plane space to the spherical three-dimensional space, realizing spherical coverage of the panoramic image. In order to achieve immersive scene display, this application fixes the observer's virtual viewpoint at the center of the sphere to ensure that the user can view the waterlogging scene with a 360° all-round perspective. The user can observe the scene details at any angle and position on the sphere by rotating and moving the viewpoint; during the projection process, this application further optimizes the texture alignment algorithm of spherical mapping to ensure seamless connection of the image texture on the inner wall of the sphere. In view of the edge distortion problem that may be caused by panoramic stitching, a spherical uniform sampling strategy is adopted. By constraining the equidistant distribution of the spherical grid, pixel information is evenly distributed to reduce image distortion caused by projection.

[0043] The method for constructing a panoramic view of waterlogging based on panoramic stitching and spatial projection in this application realizes the panoramic stitching and three-dimensional display of multi-perspective waterlogging scenes. By stitching multi-angle images in the monitoring video into a complete panoramic image and projecting it onto the inner wall of the sphere, a spherical projection model with a 360° all-round perspective is constructed, so that the spatial distribution and dynamic changes of the waterlogging scene can be intuitively presented, providing strong technical support for waterlogging monitoring, assessment and emergency response. The method has the following advantages:

[0044] (1) This application adopts a stitching method that combines key point matching with a 3×3 projection transformation matrix to ensure the geometric registration accuracy and seamless fusion effect of multi-view images;

[0045] (2) This application uses a three-dimensional spherical projection model to map the planar panoramic image onto the inner wall of the sphere, making the spatial distribution of the waterlogging scene more intuitive and the observation more comprehensive;

[0046] (3) This application supports the dynamic update function of panoramic images, which can superimpose waterlogging detection results in real time and intuitively reflect the expansion or change process of the waterlogging range. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flowchart for constructing a panoramic view of flooding based on panoramic stitching and spatial projection in this application.

[0048] Figure 2 This is the process of constructing the DoG image pyramid in this application.

[0049] Figure 3 This is the SIFT feature point detection map in this application. DETAILED DESCRIPTION

[0050] The present application is further described in detail below with reference to the accompanying drawings and specific implementation methods:

[0051] A method for constructing a panoramic view of urban flooding based on panoramic stitching and spatial projection, such as Figure 1 As shown, it is a flowchart of the construction of the panorama of flooding based on panorama stitching and spatial projection of the present application, and the method comprises:

[0052] S1, collect multi-view waterlogging scene images from surveillance video data to ensure that the images cover different angles and fields of view of the waterlogged area. Use the SURF algorithm to extract key points from the image and generate the feature vector of the image. For any image point, its SURF feature vector is defined as follows:

[0053] f SURF ={d 1 ,d 2 ,...,d n} (14)

[0054] Among them, fSURF\mathbf{f}_{\text{SURF}}fSURF is the feature vector, and did_idi represents the local descriptor value of the key point.

[0055] Through the feature matching algorithm, the key points extracted from the continuous frames are paired, and the corresponding relationship of the spatial features is used to provide a reference for subsequent image stitching. The matching metric of the key points usually uses the Euclidean distance:

[0056]

[0057] When the matching distance Dist is less than the set threshold, the two key points are considered to be matched successfully.

[0058] S2, based on the key points matched in S1, builds a point cloud model of the image to analyze the spatial relationship between multi-view images. The corner detection algorithm is used to further extract the registration points, and 8 linear equations are generated based on the registration points to solve the 3×3 projection transformation matrix H between the images:

[0059]

[0060] The projection transformation matrix has 8 degrees of freedom. The least squares method is used to solve H that satisfies the following equations:

[0061]

[0062] Among them, (x,y)(x,y)(x,y) are the original image coordinates, and (x′,y′)(x',y')(x′,y′) are the target image coordinates.

[0063] The multi-view images are stitched into a complete six-sided panoramic image through H, and multi-resolution pyramid tiles are generated for the stitched images to adapt to different resolution requirements. In the pyramid structure, any t-th level image can be obtained by downsampling operation:

[0064] I t =I t-1 ↓2 (18)

[0066] S3, using the projection transformation matrix generated by S2, maps the stitched panoramic image to the inner wall of the three-dimensional spherical model. The spherical projection is implemented using the following formula:

[0067] X=RsinθcosΦY=RsinθsinΦZ=Rcosθ (19)

[0068] Among them, R is the radius of the sphere, θ and φ are the latitude and longitude angles on the sphere, respectively. The texture data of the panoramic image is accurately aligned with the spherical model through the projection matrix, and the observer's viewpoint is set to the center of the sphere, allowing users to observe the waterlogging scene with a 360° perspective.

[0069] S4, use the deep learning model to perform target detection and semantic segmentation on the waterlogged area in the panoramic image. The input of the model is the panoramic image I, and the output is the detection result D of the waterlogged area:

[0070] D={(x i ,y i ,w i ,h i ,p i )|i=1,2,...,n} (20)

[0071] Among them, (x i ,y i ) is the target center coordinate, w i 、h i is the target width and height, p i is the target confidence;

[0072] Combined with the camera installation parameters (such as height H, angle α, focal length f, etc.), a mathematical model is constructed to calculate the location and depth of the flooded area. The actual location coordinates (X, Y, Z) can be calculated using the following formula:

[0073]

[0074] S5 combines the spherical panorama generated by S3 with the waterlogged area location data detected by S4 to build a waterlogged panoramic browsing system. Through the dynamic update module, the latest waterlogged area data is superimposed on the spherical panorama in real time to achieve an intuitive display of dynamic changes. It supports user interaction and real-time viewing of the scope and depth of waterlogging at any location. The final output is a real-time updated waterlogged panoramic map I final :

[0075] I final =I base +D update (twenty two)

[0076] Among them, I base As the basic panorama, D update The method for constructing a panoramic view of waterlogging based on panoramic stitching and spatial projection in this application collects multi-view waterlogging scene images, uses SURF feature extraction and matching to realize key point registration of images, combines 3×3 projection transformation matrix to complete accurate stitching of panoramic images, and maps panoramic images to the inner wall of the sphere through three-dimensional spherical mapping to realize three-dimensional display of waterlogging scenes; on this basis, the deep learning algorithm is used to detect and segment the waterlogging area in the panoramic image, and the precise position and dynamic changes of the waterlogging area are calculated by combining camera parameters, and the results are superimposed on the spherical panoramic model in real time to construct a dynamically updated waterlogging panoramic browsing system. This method effectively integrates image registration, target detection, spatial projection and dynamic visualization technology, and can comprehensively and intuitively display the dynamic changes of waterlogging scenes, providing efficient and accurate technical support for waterlogging monitoring and emergency response. By realizing the rapid identification, dynamic analysis and three-dimensional visualization of waterlogging areas, this method significantly improves the scientificity and real-time nature of urban waterlogging disaster management, and has important practical value and promotion significance.

[0077] Although the content of the present application has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present application. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present application can be made. Therefore, the protection scope of the present application should be limited by the appended claims.

Claims

1. A method for constructing a panoramic view of urban flooding based on panoramic stitching and spatial projection, its characteristics and The specific steps are as follows: S1, image acquisition and key feature extraction, collect multi-view flood scene images from surveillance video data, use SURF algorithm to extract key points and feature vectors of images, and establish the corresponding relationship of spatial features by matching the key points of consecutive frame images, providing a basis for subsequent splicing; S2, panoramic image stitching and transformation matrix calculation, build a point cloud model based on the matched key points, use the corner detection algorithm to identify feature points, generate 8 linear equations and solve the 3×3 projection transformation matrix, stitch the multi-view images into a complete six-sided panorama, and use layered processing to generate multi-resolution pyramid tiles to adapt to different resolution requirements; S3, spatial projection and spherical mapping, maps the spliced ​​panoramic image to the inner wall of the sphere through the projection transformation matrix, constructs a three-dimensional spherical projection model, and sets the observer's viewpoint at the center of the sphere, thereby achieving a full-scale and three-dimensional presentation of the waterlogging scene; S4, target detection and positioning data inference, uses deep learning algorithms to detect and segment targets in panoramic images, combines camera installation parameters, builds a mathematical model to infer the specific location and size of the target, and provides accurate waterlogged area positioning and dynamic monitoring support for panoramic browsing; S5, panoramic browsing and dynamic updating of waterlogging, combines the spherical panoramic image with the positioning information of the detected waterlogging area to build a panoramic browsing system for waterlogging, supports real-time dynamic updating and interactive viewing, intuitively displays the scope, depth and changing trend of the waterlogging area, and improves emergency response capabilities.

2. The method for constructing a panoramic view of urban flooding based on panoramic stitching and spatial projection according to claim 1, characterized in that For step S1, the present invention collects multi-view images containing waterlogged areas from surveillance videos, uses advanced image feature extraction algorithms to efficiently extract key points, and establishes spatial correspondences between images. The core technologies include feature extraction based on the SURF algorithm, key point matching, and optimized application of the scale-invariant feature transform (SIFT) algorithm, thereby significantly improving the accuracy and reliability of registration. The present application performs registration based on obvious features of the image, including corner points, key points, edge contours, etc. Different from the traditional method of directly using grayscale information, the present application uses the SURF algorithm to extract key points with rotation invariance and scale invariance, thereby avoiding the decrease in matching accuracy caused by changes in illumination, scale, and viewing angle. The SURF algorithm quickly locates key points by calculating the determinant of the Hessian matrix H(x, y, σ) of the image: Among them, Lxx, Lyy and Lxy are the second-order derivatives of the image at different scales σ. By analyzing the response of the determinant, local extreme points are extracted as key points. In order to further enhance the stability of the feature points, this application combines the SIFT algorithm to construct a scale space, extract the local gradient information of the key points, and generate a descriptor: f SIFT ={d1,d2,...,d 128 } (2) Among them, f SIFT is a feature vector. The 128-dimensional descriptor is obtained by counting the gradient direction histogram in the neighborhood of the feature point, which is rotation invariant and illumination invariant. Keypoint matching is achieved by calculating the Euclidean distance between descriptors of different image feature points: When Dist is less than the set threshold, the two feature points are considered to match; to reduce mismatching, this application introduces the nearest neighbor ratio criterion, which is defined as: Among them, Dist1 and Dist2 are the distances of the nearest and second nearest feature points respectively. When NNDR is less than a certain threshold, the match is considered reliable. In order to improve the efficiency of key point extraction, this application optimizes the Gaussian pyramid and difference Gaussian pyramid (DoG) in the SIFT algorithm. In each level of scale space, a fast downsampling method is used to construct a pyramid: I t =I t-1 ↓2 (5) Among them, ↓2 represents the downsampling operation, and t is the pyramid level. In the differential Gaussian pyramid, the differential image of adjacent scales is calculated: D(x,y,t)=G t+1 (x,y)-G t (x,y) (6) Among them, G is the image after Gaussian blur. In the DoG pyramid, the present application performs extreme value detection on potential feature points. For any pixel point d(t, r, x, y), when its grayscale value is an extreme value in a 3×3 neighborhood, the point is regarded as a potential feature point. At the same time, the present application eliminates feature points located at the edge with lower curvature, and accurately calculates the position of the feature points by fitting the grayscale value curve of the pixel where it is located.

3. The method for constructing a panoramic view of urban flooding based on panoramic stitching and spatial projection according to claim 1, characterized in that: For step S5, the present invention adopts a point cloud model construction method based on key point matching to achieve accurate stitching of multi-view images, uses a 3×3 projection transformation matrix to perform image registration and stitching, solves the geometric relationship between multi-view images, and introduces image fusion technology to eliminate stitching gaps and improve the visual quality of panoramic images. In addition, multi-resolution pyramid tiles are generated through layered processing to meet different resolution requirements and improve system efficiency. This application builds a point cloud model based on the key point matching results of image registration to reflect the spatial relationship of multi-view images, further optimizes the feature point selection through the corner detection algorithm, and builds a transformation equation group based on the matched key points. The general form of the 3×3 projection transformation matrix H is: The solution of the projection matrix H is based on the following projection relationship: Where (x, y) is the source image coordinate, (x′, y′) is the target image coordinate. A linear equation system is established through at least four pairs of key points and the matrix parameters {hij} are solved by the least squares method; The image is transformed by using the solved projection transformation matrix H, and the multi-view images are accurately registered to a unified spatial coordinate system to generate a complete six-sided panoramic image. During the stitching process, dealing with issues such as image brightness differences and registration errors is the key to improving the quality of the panoramic image. This application uses image fusion technology to eliminate stitching gaps. According to the image representation layer, the fusion method is divided into pixel-level fusion, feature-level fusion, and decision-level fusion. This application mainly uses pixel-level fusion with moderate computational effort, and solves the brightness difference problem in the stitching area through the weighted average method. The fusion formula of the weighted average method is: I 融合 (x)=w1(x)I1(x)+w2(x)I2(x) (9) Among them, I1(x) and I2(x) are the grayscale values ​​of the two registered images at pixel x; w1(x) and w2(x) are weights, satisfying w1(x)+w2(x)=1, and the weights are dynamically allocated using the fade-in and fade-out method: Wherein, d1 and d2 are the left and right boundaries of the overlapping area, respectively, and dx is the horizontal coordinate of pixel x. This method can effectively smooth the stitching area and reduce the impact of grayscale differences on the panoramic image. To adapt to different resolution requirements, this application generates multi-resolution pyramid tiles through layered downsampling, and downsamples the stitched panoramic image layer by layer: I t =I t-1 ↓2 (11) Among them, ↓2 represents the downsampling operation, t is the pyramid level, and the downsampling results of each layer are saved to form pyramid tiles, which support efficient loading of different terminal devices.

4. The method for constructing a panoramic view of urban flooding based on panoramic stitching and spatial projection according to claim 1, characterized in that: For step S5, the present invention application proposes a spatial projection and spherical mapping method based on a projection transformation matrix, which maps the spliced ​​panoramic image to the inner wall of a three-dimensional sphere, constructs a spherical projection model with an all-round perspective, and thus realizes a three-dimensional and immersive display of the waterlogging scene. The following are the technical details of this step and the technical solution features of the invention. The 3×3 projection transformation matrix H generated in step S2 is used to perform spatial mapping on the spliced ​​panoramic image, converting it from a two-dimensional plane coordinate system to a spherical coordinate system. The basic formula for projection transformation is: Among them, (x, y) is the pixel coordinate of the panoramic image in the original plane coordinate system, and (x′, y′) is the image coordinate after projection transformation. Through the above transformation, each pixel point in the spliced ​​image finds the corresponding position on the inner wall of the sphere to form a preliminary spatial mapping; based on the projection transformation, the mapped panoramic image is projected onto the inner wall of the sphere to form a complete three-dimensional spherical model. The spherical projection is implemented using the following formula: X=RsinθcosΦY=RsinθsinΦZ=Rcosθ (13) Among them, (X, Y, Z) are the three-dimensional coordinates on the sphere, R is the radius of the sphere, θ and φ represent the latitude and longitude angles on the sphere respectively. Through spherical projection, each pixel point is mapped from the two-dimensional plane space to the spherical three-dimensional space, realizing spherical coverage of the panoramic image. In order to achieve immersive scene display, this application fixes the observer's virtual viewpoint at the center of the sphere to ensure that the user can view the waterlogging scene with a 360° all-round perspective. The user can observe the scene details at any angle and position on the sphere by rotating and moving the viewpoint; during the projection process, this application further optimizes the texture alignment algorithm of spherical mapping to ensure seamless connection of the image texture on the inner wall of the sphere. In view of the edge distortion problem that may be caused by panoramic stitching, a spherical uniform sampling strategy is adopted. By constraining the equidistant distribution of the spherical grid, pixel information is evenly distributed to reduce image distortion caused by projection.

Images